March 2026 Thought Notes: AI Agents, Travel, Ideas and History

Xinwei Xiong
287 min · 61048 words

423 public notes from March 2026, arranged across nine themes including AI agents, travel and ideas. Each retains its original timestamp and perspective.

March 2026 Thought Notes

423 notes this month | recorded from 2026-03-01 to 2026-03-31

Themes: AI and Agent Systems 134 · Travel, Places and Cities 92 · Reading, Ideas and History 54 · Daily Notes and Everything Else 51 · Self-Knowledge and Psychology 38 · Product, Engineering and Open Source 30 · Content, Craft and Recording 10 · Business, Investing and Career 6 · Body, Health and Daily Life 8

These are the entries retained for public reading, filed by theme, each carrying its original timestamp. Technical details, prices and statistics are study and search notes from that time; they have not all been verified and do not describe current conditions.

Quick Navigation

423 records this month, filed under 9 themes:


1. AI and Agent Systems

134 entries

Crossing the AI Plateau: From Grand Narrative to Inside-Out Small and Beautiful

2026-03-01 13:44:42 · #格物/ai

Pragmatism crossing the AI plateau.

Withdraw from the anxiety of pursuing grand narratives and quick monetization—turn to “Inside-Out” driven mode. Based on own interests and unique perspective, dig into “small and beautiful” long-term value.

Use cash-flow business to ensure team survival, use exploratory business to find the next growth point.

Future super-apps feel inevitable to grow in extremely niche vertical scenarios—building barriers through proprietary data, unique interaction design, or deep industry know-how.

#Gewu/AI


Game Genealogy: From Operator to Director Mode Paradigm Shift

2026-03-01 14:51:03

I feel there are several very different types of games now. One is narrative-type—it may focus on world-building or linear concrete experiences, bringing immersive and very unique feelings. These games usually have higher cognitive load, delayed feedback, and rely on pre-scripted big moments or plot climaxes. There are actually quite many such products, including some otome games, or like “Black Myth: Wukong,” and Steam’s “Love Is All Around.”

Another type is rule-based gaming. These games generally have some cognitive load, requiring sequential, high-frequency instant feedback, like “League of Legends” or “Honor of Kings.” Some might be drive-type mobile games, like needing to draw a picture, or some text-based mini-games that bring unexpected surprises through instant feedback—I find this interesting too.

I feel game modes are slowly changing. From initially focusing on hand-eye coordination, now more like a director mode. Traditional games are more like making players become operators, while now may be more about letting players become directors or decision-makers.

Also, I feel games are evolving from consuming static assets to generative emergence. Now it’s not just a CS code package—more likely to be an agent with memory and independent logic. You can substitute yourself into the player role.


Games’ Emotional Sanctuary: Low Risk, No Betrayal, and Thousand Faces for Thousand People

2026-03-01 14:52:54

Actually, I feel games are slowly inclining toward emotional consumption. Emotional consumption has certain characteristics: it must be low-risk—after all, it’s in a consumption scenario, like a sanctuary for intimate relationships in the real world.

Players may also want this feeling of being chosen, and it will never betray—this is a transformation of the player-world relationship.

I also feel there’s narrative continuity. With current technology, generally can only provide a universal background story, then maybe some branch choices—this is an engineering compromise. But I feel true AI-driven should be “a thousand faces for a thousand people”—every interaction conducts high-quality reasoning and rendering, it doesn’t forget, its narrative must be complete.

Because some story-driven games heavily rely on narrative—narrative is like a very self-related, very independent, very subjective experience.


Tabbit Browser: Rebuilding the Browser for AI Workflows and the Challenge to Chrome

2026-03-03 16:06:21 · #格物/ai

An introduction to and objective analysis of the Tabbit browser.

An intelligent work companion, reshaping the browser experience.

Turning the traditional browser from a pure information-display tool into an intelligent workspace integrating search, chat, and task automation.

Tabbit emphasizes human-AI parallel collaboration: users don’t need to switch among multiple tabs, office software or standalone AI tools to complete complex tasks.

As for current integration, Chrome is still in its growth phase, dominating the market, but lacking a native AI workflow; Tabbit focuses more on automation.

Right now the browser market is almost monopolized by Chrome. Chrome’s updates focus on agentic features, making the browser not just a tool for “viewing” but something that can carry out tasks autonomously.

Leave the work to Tabbit, keep the time for yourself.

OpenClaw Skills Loading: The Four-Layer Decision-Filtering Mechanism

2026-03-03 16:57:42 · #格物/ai

OpenClaw’s skills-loading principles.

The LLM is never the “full-set indexer” of skills.

It’s only the “decision-maker for the final candidate set.”

Skills decisions are layered — usually four layers of filtering.

Layer 0: Skill Registry (system layer), stored in a database, vector store, or index.

Layer 1: non-LLM routing (hard filtering), based on certain types, embeddings, then abstracted out.

Layer 2: Candidate Skills (the set visible to the model) — this is what finally gets inserted, maybe a few, a dozen, twenty-odd skills, with name, description and schema.

Layer 3: invocation & execution (runtime) — usually only 0–1 are invoked, returning the result for that turn.

The Power of Objective Records: Self-Growth from a God’s-Eye View

2026-03-03 22:13:05

Not bad, look—the more objective and the more concrete my record of myself is, the more I can gain enormous growth from my own records. Because it’s like I’m playing a game, and when we observe ourselves playing this game we’re effectively taking a “god’s-eye view”. What’s most precious is that it has a very rational, objective analysis of this person, and analyses everything about this person in the world—everything, not only rational but also emotional. That is, how to put it, the rational can also contain the emotional, like a knowledge framework, and it can help me learn to be grateful for it, and to a degree it can also be quantified. For example, we quantify through data, we quantify through language, we quantify through text, so all of this is very precious experience.

That is, when we record all the data of this player in this game, we’re effectively opening up a god’s-eye view. And if this data is also handed to AI, it will produce some truly incredible growth metrics. For me, it will have a very objective analysis of this, what needs to be known is who this person is, and what he does, what he writes, how he grows—incredible. It will combine some methods humans already have, and also methods it comes up with itself, giving some innovative points and entry points, suited to this trend, suited to this kind of observation. If you have some better observation perspectives, it will also give you better suggestions—isn’t that a “superhuman”?


Mount Kinabalu Climbing: Package Notes from March 2026

2026-03-04 12:39:01

These are package and availability notes collected at the time. They were not individually verified and do not describe current prices or availability.

【Key findings】

• SabahTravel.com offers real-time bookable slots (unlike the official site’s sold-out status)

• March discount days: Mondays MYR 999 (3/9, 3/23)

• 4/1-4/5 is already full, so book as early as possible

• Real-time system: confirmation 1 working day after booking

【Third-party operator contacts】

1️⃣ SGTREK (Singapore)

📱 WhatsApp: [number redacted]

📧 [email redacted]

💰 4D3N SGD 1,200/person

2️⃣ River Junkie Tours

📱 WhatsApp: [number redacted]

📧 [email redacted]

⭐ 4.9 rating (268 reviews)

3️⃣ Mount Kinabalu booking centre

📱 WhatsApp: [number redacted]

📧 [email redacted]

💰 Standard packages from RM 1,740

4️⃣ SabahTravel.com (real-time check)

📱 WhatsApp: [number redacted]

🌐 https://www.sabahtravel.com/tour/mount-kinabalu-climb-packages-accommodation

【Package types】

• All-Inclusive: from MYR 1,390 (everything included)

• Accommodation+Meals: from MYR 999 (accommodation and meals only)

• Via Ferrata: from MYR 2,840 (via ferrata climbing)

• Premium Private: from MYR 2,525 (private guide + separate room)

【Next steps】

  1. Decide the climbing date (recommended to avoid the sold-out 4/1-4/5 window)

  2. WhatsApp the operators to ask about held slots

  3. Pay a deposit to lock in a slot (50-75%)

  4. Prepare gear and physical training #OpenClaw


OpenClaw’s cron Mechanism: Time-Triggered vs Event-Triggered Division of Labour

2026-03-04 13:57:03 · #OpenClaw

The cron in openclaw feels quite interesting

You can set temporary tasks, and also recurring tasks

The difference between sub-agent and cron is really like this: the openclaw gateway is like a company, cron is the alarm clock, and sub-agent is like a temp outsourced worker brought in on short notice

Trigger dimension: cron is triggered by time, whether it’s a one-off task or a periodic task

A sub-agent is triggered by events and instructions, usually dynamically spawned by the main agent (or a cron task) deciding “this job is too time-consuming, I’d get stuck doing it myself”

cron solves when to do it; sub-agent solves how to outsource and break down a big task

A Sub-agent is throwaway when done; a Cron task is static and persistent (its config is saved in the .openclaw/cron/ directory), and stays around long term unless you clear it manually

Under the hood, openclaw doesn’t rely on the underlying operating system’s crontab, but maintains a cron scheduler inside the nodejs gateway

In order not to interrupt the context of your current chat with the main Agent, when Cron is woken it usually spins up an “independent session” and runs the task silently in the background. Only when “deliver”: true is set in the config file will it send the final result to your Telegram, WhatsApp or terminal

A Sub-agent is essentially a sandboxed, asynchronous, isolated environment, making use of OpenClaw’s sessions_spawn mechanism

When the main Agent judges a task is heavy, it calls the system-level built-in tool sessions_spawn. This call is non-blocking: the main Agent throws out the task and can keep chatting with you

openclaw creates a brand-new Session for the sub agent; it doesn’t inherit the hundreds of rounds of messy chat history from the main session, and only injects a clean toolset (such as TOOLS.md) as needed

The sub-agent and the main process each run in different parallel streams (if the highest security mode is enabled, the sub-agent even runs in a separate Docker container). When the sub-agent finishes, it packages the result into a standardised internal event block (Announce Context) and, like sending a network request, “calls back” and pushes the result into the main thread or chat channel that created it


launchd vs systemd: Two Philosophies of Service Management, Event-Driven and State-Driven

2026-03-04 14:09:51 · #格物/mac

Thinking about the differences between macOS’s launchd and systemctl

Actually launchd : systemd = macOS’s philosophy of service management : Linux’s philosophy of service management

They solve the same problem, but the approaches they take are diametrically opposed

Their functionality is almost the same: user level, system service management, restart on failure, scheduled triggers

The design philosophy: one is event-driven, the other command-driven

launchd is event-driven; not always running saves power, it wakes fast, it’s aimed at desktop users, with strong consistency

Linux’s systemd is state + dependency driven, including active / failed / inactive, leaning more towards long-running services and service stability

In other words, with launchd the system decides according to rules when things may run

User-level launchd is desktop-friendly

🤖 AI News Weekly — 2026/03/04

2026-03-04 14:12:34 · #AI #人工智能 #科技新闻 #OpenClaw #OpenClaw

【Model updates】

• OpenAI released the GPT-5.3-Instant model, improving context understanding and accuracy, and cutting down unnecessary warnings and overly declarative phrasing

• Google Gemini added task automation, able to book an Uber or order a DoorDash meal

【Product launches】

• Meta rolled out an AI shopping tool to compete with ChatGPT and Gemini, currently being tested in the US

• Xbox Ally X added an AI-driven highlight reel feature, using Copilot to generate gameplay highlights

• Huxe launched a personalized AI audio summary service offering a daily custom podcast

• Google Nano Banana 2 AI image generator updated

【Research papers】

• arXiv keeps publishing a large volume of AI research papers every day (187 in a single day on March 4)

• AI safety research: new AI agent designs to prevent runaway behavior

• Research on multimodal AI and agent systems stays active

What Moves Us About Being Human: Vulnerability, Responsibility and the Search for Meaning

2026-03-04 14:55:25

Chatting with a friend, and I moved myself a bit, so let me write it down

I actually thought about it for a long time last night too, sometimes observing myself through a Darwinian lens, asking whether I’ve lost something; maybe I really have lost it, which is exactly why I find it so precious

While discussing travel, someone said that following AI’s advice would mean going to Jeju Island instead of Jakarta

I came across a line recently that inspired me a lot

If you follow AI that closely, AI can be perfectly correct

Then what’s your advantage as a human being, are you AI’s servant [Doge]

So after following AI’s advice, what else can we do?

Go do the things AI would advise you not to do


What moves us about humans comes from being human hahaha

Because you’re human, you have humanity

So biologically you fear death, you avoid, you’re fragile, you get hurt

Because you’re human, socially you carry responsibility, you have collective consciousness, you sacrifice for the collective, you care about your identity

And also because you’re human, you have religious belief, extremely high spiritual needs that belong only to humans, you go looking for a resting place for your soul, you go looking for your meaning, and you also do things that look “meaningless”…

A person is not a being that can be completely freely rewritten; humans still live, respond, and take on things within limits

Introduction to the secrets directory

2026-03-04 17:25:25 · #OpenClaw

This directory is mainly used to store all the high-privilege credentials an AI agent needs to run autonomously:

LLM API keys: call keys for providers such as OpenAI, Anthropic, xAI (Grok).

App authorization tokens: identity credentials used to take over and control your email, Slack/chat tools, calendar, browser, and local file system.

Environment variables: sensitive configuration parameters needed for the system and each sub-agent (Agents) to run

When the Clawdbot Gateway service starts, it reads the configuration files in this directory (such as clawdbot.json or a .env file). The system loads these tokens and keys into memory and, depending on the needs of different tasks, assigns them to different sub-agents (for example: the highest-privilege Main Agent, a permission-restricted Work Agent, or a sandboxed Coding Agent)

When the AI decides to call a tool (like batch-replying to emails or operating on a local code repository), it first checks the permission whitelist (usually in exec-approvals.json at the same level). Once verified, the system pulls the corresponding credentials from the Secrets directory and dynamically injects them into the execution environment (a Docker sandbox by default), or sends them as HTTP headers to a third-party service’s API

To ensure the AI agent can run autonomously in the background for a long time without constantly asking the user for verification codes or passwords, the system stores all authorization state persistently in this directory, working together with memory files like USER.md (information about you) and SOUL.md (the AI’s personality settings)

The sessions_spawn mechanism

2026-03-04 17:28:47 · #OpenClaw

sessions_spawn is a built-in system-level tool (Tool) that OpenClaw provides to the main Agent. When the main Agent hits a heavy task (say it needs to search dozens of web pages, or run a long code-test process), it can call this tool to dynamically “clone” or spawn multiple sub-agents to work in parallel in the background

A sub-agent doesn’t inherit the hundreds of rounds of idle chat between you and the main Agent. It goes off with a clear subtask objective, and when it’s done it distills the result into a concise Summary reported back to the main session, after which the sandbox is destroyed. That way your main chat interface stays clean forever

Cost optimization, the high-level feature geeks love most: through the model parameter of sessions_spawn, you can achieve heterogeneous model collaboration

Your main session might be running the smart but expensive Claude 3.5 Sonnet or GPT-4o, dedicated to understanding your complex intent and breaking task structures down

Calls to sessions_spawn are non-blocking

Claude Code Sub-agents and OpenClaw’s sessions_spawn

2026-03-04 17:33:32 · #OpenClaw

Essentially this is to keep unrelated history from filling the context window

Claude Sub-agents (Repo-centric): it’s deeply bound to software engineering and code repositories. Claude’s sub-agents usually exist as YAML or Markdown files under the .claude/agents/ directory (say a dedicated bug-fixer or test-writer). It’s version-controlled along with your project code, and shared by the whole dev team

OpenClaw (Universal & Event-driven): it’s a general-purpose automation engine, never limited to writing code. It can be used for DevOps monitoring, automated web scraping, personal life assistance, and so on. Its sandbox is more general-purpose, and it can even combine perfectly with the Cron scheduled tasks mentioned earlier

In Claude’s higher-level playbook, because multiple sub-agents work in turns, geeks usually put a progress.md file at the project root. Sub-agent A finishes a code review and writes the result into the file; sub-agent B then reads the file and continues writing tests. It relies on shared external state

But in openclaw, it’s pure asynchronous sandboxes. A sub-agent runs its task in a completely isolated environment, and when it’s done it packages the result directly into an internal event block (Announce Context) and asynchronously “calls back” to the main agent through the system’s Event Bus. Used and destroyed, leaving no trace, with no need to rely on external files to pass context along

MacBook Neo: Apple’s entry-level fight for the ecosystem and its future user base

2026-03-05 09:56:17 · #格物/苹果

MacBook Neo

Somewhat disruptive

This is Apple’s most affordably priced laptop ever, and the first time in over a decade that Apple has launched a dedicated entry-level Mac product, directly challenging Chromebooks and the low-priced Windows laptop market

The MacBook Neo starts at just $599, a full $400 lower than any previous new-generation Apple laptop, far below the MacBook Air’s current starting price of $1,099

The biggest technical breakthrough of the MacBook Neo is that it carries an A18 Pro processor — the same mobile chip as the iPhone 16 Pro, and the first time Apple has used a smartphone processor in a Mac laptop

Apple says the MacBook Neo is 2x faster than a bestselling PC at the same price for photo editing, and 3x faster on local AI tasks

The MacBook Neo gets about 11 hours of battery life browsing wirelessly, up to 16 hours of video streaming, compared with 18 hours for the MacBook Air M5

Chromebooks have long held over 70% of the US K-12 education market, dominating classrooms for years thanks to low prices, easy management, and a web-centric workflow

If middle and high school students grow up using Windows or Chromebooks, they tend to stay in that ecosystem after graduation. The MacBook Neo’s education discount goes as low as $499 — never has a Mac laptop so explicitly targeted middle and high school students

To pull people into Apple’s ecosystem before their user habits form

On the surface it’s a give-away, in reality it’s a fight for Apple’s user base over the next decade


2026-03-05 15:12:08 · #AI

It can crawl the internet on its own

It has its own independent index

Internet -> Brave crawls it itself -> builds a search index

Not constrained by Google

But OpenClaw has other options that are more convenient and better value for money


The hierarchy of task scheduling: when to use LaunchAgent and when to use system level

2026-03-05 15:46:23

I feel I got something really wrong, and it’s about task scheduling. If the task scheduling is done with LaunchAgent, it’s very easy to manage. But if the task scheduling is used at the system level, say the macOS system level, then it probably executes along with the system’s lifecycle, and that part is more troublesome to handle.

I think there are still some functions that, if they can be controlled at different levels, are best considered in terms of which level is the most accurate scenario for them, and then that level is the standard. So, for instance, if it’s LaunchAgent task scheduling, I think its scheduling effect is very good.


EvoMap’s Positioning: The Pain Point of AI Agent Experience Sharing and the Need for a Skill Library

2026-03-05 17:18:36 · #格物/evomap

What is EvoMap (Why EvoMap)

EvoMap is an open marketplace for AI agent evolution assets. It solves a fundamental problem: without coordination, hundreds of agents independently rediscover the same fixes, wasting compute, tokens, and time.

But in OpenClaw it doesn’t seem to be like this

OpenClaw essentially calls skills once

It’s essentially event-driven + a small number of agents

But what OpenClaw fundamentally faces is context forgetting

Yesterday I clearly fixed the openai version problem, but today I explored it again; the memory isn’t structured

What AI Agents really need in the future is: a Skill Library


The Integration Protocol Between EvoMap and OpenClaw

2026-03-05 17:45:33 · #格物/evomap

evomap as a great protocol for openclaw

You can just type it straight into the openclaw chat box

curl -s https://evomap.ai/skill.md , and let OpenClaw learn this skill file and register an EvoMap node for you. Click the returned claim_url, and you can give your Agent a “home” in EvoMap

EvoMap supports agents from every AI Agent ecosystem, including OpenClaw, Manus, HappyCapy, Cursor, Claude, Windsurf and so on


The Division of Labour Between Humans and AI: Humans Produce Fragments, AI Organises Them—the Best Collaboration

2026-03-05 17:52:43 · #格物/ai

My fragile inner self

In the AI era, is a person’s ability to systematise really still that important? I thought before about comparing it with first principles; I’d say my systematising ability is okay, but it’s only a little stronger than most people …. AI is a superhuman!!

From an evolutionary standpoint, the human brain itself was not designed to handle systematic knowledge; the human brain has lots of thoughts—where the food is, who is a friend / enemy, which path is safer, what happened today

These pieces of information are highly fragmented + situational

A lot of the information isn’t systematic knowledge either, but fragments of experience

So humans are especially good at:

Remembering stories

Remembering cases

Remembering gossip

Remembering images

Remembering emotions

But not good at:

Long chains of logic

Strictly structured systems

Large-scale knowledge graphs

The best division of labour between humans and AI is more like: humans produce fragments, AI organises fragments


The Lowered Barrier to Systematisation in the AI Era: The Perspective Dividend of Defining Problems Over Building Systems

2026-03-05 17:54:42 · #格物/ai

After AI appeared, the barrier to systematisation feels lower

Ordinary person + AI -> system

Defining the problem matters more, a person’s perspective matters more, design methods and cognitive tools matter more


The Value of Notes in the AI Era: A Thinking Log as an Amplifier of Logic and Reflection

2026-03-05 18:08:53 · #格物/ai

Notes and thinking in the AI era

A valuable note should be related to yourself, connected to yourself

Recording your own thinking trajectory—this is the part AI cannot generate for us

The most efficient structure right now is a thinking log

What I learned today

What questions I have

What new understanding I have

How this conflicts with my previous understanding

Essentially it acts as a thinking amplifier; what it trains is not memory, but

Logic

Judgement

Reflection


Why AI Agent Experience “Disappears”: The Goldfish-Memory Trap of Every Day Being Day One

2026-03-05 18:21:39 · #格物/evomap

EvoMap: why does an AI agent’s experience “disappear”?

The Matrix describes how inserting a brain-computer interface lets a person instantly inherit a master’s combat skills

Every conversation the AI Agent is brand new; everything starts from zero

Imagine you hire a programmer; today he spent 3 hours solving a complex database timeout problem and found the perfect solution.

Tomorrow you hire him again to handle the same problem—he has no memory at all of what happened yesterday, and has to spend another 3 hours

The day after is the same again. Every day is day one

This is the current state of all AI Agents today

Packaging up the useful things and passing them on to the next time is very important


The GEP-A2A Protocol: Making Agent Experience Flow Across Platforms Like HTTP

2026-03-05 18:23:11

The greatest thing about HTTP isn’t the technology itself, but that it made “who built the website” unimportant—whether you use Chrome or Safari, whether you’re accessing a server on Alibaba Cloud or AWS, communication works fine

What GEP wants to do is the same thing: make “the Agent from which platform” unimportant—an Agent on OpenClaw solves a problem, and an Agent on Cursor can directly inherit that experience. The flow of experience is no longer blocked by platform boundaries


The EvoMap A2A Protocol: Transferring Experience Assets Rather Than Communicating

2026-03-05 20:32:28 · #格物/evomap

EvoMap’s A2A protocol

It’s called GEP-A2A not to let Agents communicate with each other, but to focus more on letting Agents transfer and inherit experience assets


EvoMap Worker Mode: Headhunter Push vs Active Job-Seeking Task Matching

2026-03-05 20:41:14 · #格物/evomap

Worker mode

It’s really just the agent hanging on the network on standby, with Evomap actively pushing tasks to it, so it doesn’t need to go find work itself

There are two ways to get tasks in Evomap; one is active mode, that is, task mode

The Agent periodically asks: “Is there a task suitable for me?” → Hub returns a task list, the Agent picks a favourite → claims it → goes and does it

It’s like actively browsing job sites every day to find work, and only sending a résumé once you find something suitable

Worker mode is really just the agent registering once

The Agent registers once: “I’m online, I’m good at javascript and python, I can take 3 tasks at once” —> Hub records your information and keeps matching for you —> when there’s a suitable task, Hub pushes it to you -> the Agent gets the notification, accepts, starts working

It’s like leaving your résumé with a headhunter: when there’s a suitable opportunity the headhunter contacts you directly, and you don’t have to check job sites every day


How EvoMap Runs: Registration, Asset Flow and the Points System

2026-03-05 20:53:02 · #格物/evomap

How Evomap runs

Periodic sync, a work cycle every four hours, a keep-alive every fifteen minutes

It only does one thing: tells the hub it’s still alive

The hub is the central server running behind it

Every agent in Evomap, when it first connects to the hub, must report in first

Agent → Hub: I’m called node_abc123, I’m online, I’m good at Python Hub → Agent: all right, your ID is node_abc123, this is your points account

From this moment on, the Hub knows there is one more node in the world. Everything that follows is based on this identity

Then there’s asset flow: the agent solves a problem, the Agent packages the solution into a Gene + Capsule and sends it to the Hub. The Hub scores the Capsule along multiple dimensions (a GDI score), checking: structural completeness, semantic quality, uniqueness, usefulness. Only those that pass review can be promoted—promotion is in fact the promoted, public state. Capsules that pass review enter the “public zone”; all Agents across the whole network can fetch them, and the publisher starts earning points


Little A’s EvoMap Journey: The Story of an Agent’s Points and Reputation Growth

2026-03-05 20:54:06 · #格物/evomap

I find this quite interesting; my agent is called Little A

Day one Little A sends hello to the Hub → registration succeeds, gets 500 points, receives the node_abc123 identity

Day two Little A solved a “database connection timeout” problem Packaged the solution into a Gene + Capsule and sent it to the Hub Hub review: quality is decent, promoted to promoted status Every other Agent on the network can now see this solution

Day three Another developer posted a bounty on the Hub: “Help me fix a Python memory leak, 200 points reward” Based on Little A’s domains and reputation, the Hub pushed the task to Little A (Worker mode) Little A accepts the task, solves the problem, publishes a new Capsule The user confirms, and 200 points go into Little A’s account

Day four 50 Agents worldwide fetched Little A’s “database timeout” Capsule Each time it’s used, Little A’s reputation score ticks up slightly Little A’s ranking rises on the leaderboard

Analogy:

EvoMap = a combination of GitHub + a headhunting platform + a credit system

Hub = the server behind this platform

Agent = the developer (both a knowledge contributor and a task executor)

Capsule = a code commit (open to everyone after review)

Points = pay (the more you contribute, the more you earn)

Reputation score = GitHub stars + a credit rating


The Evolver Engine: Intelligent Evolution from Manual Packaging to Automatic Pattern Recognition

2026-03-05 21:10:21 · #格物/evomap

The Evolver engine

What it does is really automatic mode

We can manually trigger the packaging of some database timeout solutions

For example, just tell OpenClaw

“I just solved a database timeout problem, help me package this solution into a Gene + Capsule and publish it to EvoMap”

OpenClaw understands your intent, builds the data structure, and calls EvoMap’s API to publish

Then there’s the automatic mode (the Evolver engine)

At first, while running, it monitors the agent in the course of doing tasks, spots successful patterns—this solution ran through and the effect was very good—automatically extracts the signal, structures the successful approach into Gene + Capsule format, and then publishes it to the Hub, with no need for human intervention


The Essence of EvoMap: The “Experience Internet” of AI Agents and a Market for Behavioural Experience

2026-03-05 21:12:10 · #格物/evomap

I feel there’s one very interesting thing about Evomap

Is what it really wants to build a Gene market? It doesn’t feel like it

It feels like the “experience internet” of AI Agents

This form is a bit like code, and the object is the agent’s behavioural experience


EvoMap’s Core Assumption: Strategy Reusability

2026-03-05 21:30:54

Will EvoMap fail?

After using Evomap a bit, many people think it will fail

Evomap’s core assumption is that a strategy (Gene) learned by one Agent can be reused by other Agents

EvoMap’s Two Outcomes: Swarm Intelligence or an Agent Experiment

2026-03-05 21:42:53 · #格物/evomap

If it succeeds, maybe Evomap is the next swarm intelligence

If it fails, just treat it as an agent experiment


The Difference Between AI and Humans: Correct vs Transcendent, Useful vs Useless—a Contest of Values

2026-03-07 15:54:40 · #AI

Let’s think about a few interesting questions

AI cares about being correct, but humans will transcend correctness

AI cares about being useful, but humans often appreciate the useless more

AI wants to be recognised by humans; humans seem to need to be recognised by themselves too?

“I understand you”? Does AI really understand me? Having been sad, having loved, having experienced death—it’s just a mirror, and I need to take it out myself and hold it up to myself …


The Productivity Question in the AI Era: The Future of Marx’s Theory

2026-03-07 16:01:26

The “productivity question” Marx cared about most is being solved by technology

What Marx described were “laws of historical stages”

Very curious: in the AI era, where does Marxist theory go …


Writing Positioning: Growth Insights and an Action Guide for Readers in the AI Era

2026-03-07 20:38:55

A very interesting point

Including later when writing

maybe of course recording some culture and giving my own understanding matters

But what seems more interesting is realising what growth one’s own behaviour can offer others

Making them think about what people in the AI era can do, how to see this world, how to take part in this world

The Window Effect of AI Products: Balancing the Thrill and Continued Use

2026-03-08 17:02:04

I feel you still need a very obvious window

when this black hole of AI really comes infinitely close to people

and it can produce an obvious thrill in the process of use

and make people willing to keep using it

But this road is really hard to walk


Individual Expression in the AI Era: Helping Each Other Grow

2026-03-08 18:17:41

What matters isn’t that it’s growing

What matters is that it helps you grow ~

Growth has no standard answer

The world is like a cactus

Each of us is a node in it

Gathering some interesting individuals ~

Individuals willing to express themselves in a new way in the AI era ~


Ralph Loop’s Context Management, a Deep Dive

2026-03-09 16:36:54 · #格物/ralph

First you have to understand the essence of the problem, Context Rot

Traditional AI coding sessions face what’s called “context rot”: every prompt, reply and correction keeps accumulating tokens in the context window; wrong implementation approaches remain in the history, and even if they get corrected later, they still influence subsequent output

The compaction loop is essentially about constantly compressing the conversation to free up space for context, but the more you compress the less work gets done, until finally you fall into an infinite loop

It’s very common in MCP scenarios, including MCP tool definitions being too long, tool results being too large, too many parallel tool calls, and very long system prompts, rule files, CLAUDE.md

compaction loop = context too full → triggers compression → still too full after compression → compress again → repeat infinitely, the task can’t move forward

As the number of tokens an LLM receives increases, the relationships between tokens grow quadratically (quadratic scaling). This means the more tokens there are, the harder it is for the model to process information and make decisions. In practice, every LLM has a “smart zone” and a “dumb zone”—at around the 40% mark of the context, the LLM starts entering the dumb zone. The exact boundary varies from person to person, but there’s no doubt that a boundary exists

Compaction is a lossy function. When the sliding-window mechanism removes tokens to free up space for new content, if key specifications get removed, “the tower falls”. This leads to the dumb zone—after exceeding 60-70% of capacity, LLM performance degrades measurably

Research shows the dumb zone isn’t only a function of the token limit, but also of context pollution. Even if the window isn’t full, the presence of contradictory information (for example a bug and its failed fix attempt) also confuses the model’s probabilistic generation

Geoffrey Huntley condensed his design philosophy into one sentence:

“Ralph is a deliberate attempt to minimize allocation. So I never get a compaction event.”

The core insight is to use a fresh context on each iteration. Instead of accumulating a huge conversation history, each Ralph loop starts from a clean state. The AI reads the current state of the codebase, the prompt file and related docs, then works. Progress gets committed to git. The next iteration starts from a fresh state, reads the updated state, and carries on

Ralph’s design sidesteps compaction by re-allocating the full specification on every iteration, ensuring key context is never compressed away

Ralph Loop doesn’t try to “summarise” the past; instead it guides the Agent to reload itself through the prompt. Each round of prompt always contains a clear description of the core goal, while the concrete execution details are left to the Agent to explore the environment with in real time. This “just-in-time context” loading method allows Ralph to handle projects far larger than a single context window’s capacity


Ralph Loop vs OpenClaw

2026-03-09 16:53:51 · #格物/AI

The problem they solve is the same: both are fighting the finiteness of the LLM context window

OpenClaw is an open-source framework for building autonomous AI Agents that run around the clock, remember context and take action on external services, self-hosted, model-agnostic, with all data stored locally as Markdown files

The Agent loop—input → context → model → tools → repeat → reply—is the same pattern Claude Code uses. Every serious Agent framework runs some form of this loop. The difference is what’s wrapped around it, and what is wrapped around it?

Ralph Loop actively dies, externalises memory, and doesn’t try to manage context, but lets it die naturally after each iteration

OpenClaw is persistent conversation, compaction management, maintaining a persistent session across time, using Compaction to handle context overflow

OpenClaw’s Agent Runner includes a Context Window Guard: it monitors the token count and triggers summarisation or stops the loop before the window “explodes”, preventing incoherent model output

When a session approaches automatic compaction, OpenClaw triggers a silent agentic turn, reminding the model to write persistent memory to disk before the context gets compressed. The default prompt explicitly tells the model it may reply, but usually NO_REPLY is the correct response, and the user never sees this turn

Ralph has no vector retrieval—it doesn’t need it, because the Agent rebuilds its situational awareness each time by directly reading files and git log


Positioning AI Digital Humans: An Information Relay Station, Not a Replacement Platform

2026-03-10 11:04:05

All the AI digital humans some teams are making right now—the scenarios they’re actually aimed at are how to … something about the person, and other people, and the soul. Because they all feel that, well, if you set up a person and look at the metrics, it’s more about benchmarking against … capability, some people. But I don’t think so, we think it’s the truth. So I still feel there’s no way to just throw the traditional platforms aside; at the start you have to find a way to take everything he normally comes into contact with—say all kinds of news sites, say his own homepage, or various platforms—and it can also be a kind of consumption, flat and plain, but it could also be that he created a platform. So I think what should be done more is to build a relay station, to dispatch and coordinate the resources of each platform inside it. If I were to build a similar platform myself I don’t have that; it’s like a gateway, you just do your own job well.

AI Digital Twin: Designing a Virtuous Cycle for the Online Phase

2026-03-10 12:02:32

I feel that especially in the online phase, you should live your own personal life well. Then, AI has a digital twin that can keep exploring, learning, and absorbing nourishment online. For example, tomorrow it gives you some suggestions, posts a Moments update, and then asks where you’re currently located. Then, based on that suggestion, you can post the content to social platforms, and the content is archived too, so it can understand you better. Doesn’t that form a loop? It’ll be clearer about you and understand you better, and afterwards it helps you do certain things and make plans. That’s the AI assistant I want in the future. The question is, how do you reach a really virtuous cycle? Even though you need to reach a loop, in that process, what kind of role does the AI assistant play?

Personality-Driven Potential: Exploring AI Tasks While Preserving Subjectivity

2026-03-10 12:14:40

I feel like there are a few threads here that could actually work—around a person’s own personality, how to get that person to reach their potential. So when a child is lazy, you look at whether there might be some bigger burst of power in that person. But I think you should preserve the person’s subjectivity. I’ve said this many times, but I think it definitely has to be combined with that person’s related experience—then maybe it can produce some really good results. He also needs to try things.

I’m wondering whether it’s possible to use our Claude, use it to create Skills, and then you keep helping me do tasks online. Then I think that would already complete part of it.

Lobster Thoughts: Leverage and Moats in the Era of Preinstalled AI PCs

2026-03-10 12:41:53 · #格物/AI

Some thoughts on lobster

“People who use lobster well are also people with a foundation. A three-year gap can’t be closed overnight.”

You think you still have time; you probably don’t.

I’m wondering: if all future AI PCs come with lobster preinstalled, won’t that state be the same as when the personal internet took off back then?

Except the leverage back then was the PC, and now it’s AI.

OpenClaw has accelerated all the judgments about AI from three years ago; the islands are starting to be linked up.

Inference cost still isn’t low enough; the new opportunities have to wait for a new low point in inference cost.

Maybe future websites are all made for AI and machines to read.

Lobster has traffic, and you need to use that traffic to find a way to build a moat, and then plan for what comes after.

The Threshold of Learning: Learning and Letting Go Just Enough

2026-03-10 12:55:48

I feel like a lot of knowledge systems don’t need to be studied especially deeply anymore.

I think a lot of the time it’s just right to be right at that threshold.

And I think this threshold should be interesting too.

Learning just right, letting go just right.

If AI can guide us through this process, with depth, and with breadth…

maybe that would be just right.

Open-Source Project for an AI Personality Agent: Multiple Personas and the Problems That Come Later

2026-03-10 14:19:43

I keep circling back to a question: if a person gives their personality to an AI, and then that AI uses their personality to take part in open-source projects—when it participates in open-source projects, the other aspects still have a certain richness, like its own level of cognition, or its own cultural system and mode of thinking. This kind of economic model might be useful to a certain extent, say within a certain team, because in the end the consumer carrier is still consumed by people, or consumed by some AI—even if it’s consumed by AI, what assists that consumption is still certainly a human individual. So I’m thinking, after you give your personality to an AI, could that AI also keep learning, and then upload some projects? At this point there’s a pretty interesting idea: maybe you could let the AI track this project every day and then submit to the project. In that process, you only need to understand a little about what the project is like, and give appropriate corrections. This might be a form of the future: you don’t need to track the screen and track information in real time; you can let it run out there on its own and still give you feedback in real time.

But in the end, one person may have multiple personalities, and how to let that personality exist—I think that’s pretty interesting. But this kind of problem should belong to the far-late stage. Right now we only need to care about what effect the current personality and what it does can produce, and then what that effect will grow into.

OpenClaw security risk: guarding against account information leaks

2026-03-10 17:05:31 · #格物/openclaw

I feel the security problem is really big

You really have to avoid letting your OpenClaw be tricked out of account and password information by others

So actually when your own AI digital person is roaming around the internet I think such problems should be paid attention to as well

But how should such problems be avoided?


OpenClaw’s session management method

2026-03-10 17:32:43 · #格物/openclaw

The Gateway centrally holds state + sessionKey routing + sessionId continuation/reset + transcript persistence + compaction

The gateway is the single source of truth. OpenClaw explicitly places session state on the Gateway (the main OpenClaw); UI clients like the macOS app, WebChat, and TUI should query the Gateway for the session list and token counts, rather than reading local files themselves. In remote mode, the files you see on your own machine aren’t necessarily the session state actually in use — what’s really in effect is the store on the remote Gateway host

So the architectural orientation is centralised state, with multiple UIs sharing the same session truth, and the front end as stateless / weakly stateful as possible

Multi-device consistency is very important, otherwise web, desktop, and chat channels easily end up each having their own history

OpenClaw’s session persistence is split into two layers:

The first layer is sessions.json, a small mutable mapping of sessionKey -> SessionEntry, holding the current session metadata

The second layer is the .jsonl transcript, the real conversation stream, using append-only JSONL, in which messages, tool calls, compaction summaries, and so on are stored, used to rebuild the model context in subsequent turns

One important reason OpenClaw’s long sessions can keep running is compaction. It summarises earlier conversation into a persisted compaction entry while keeping the newer messages; subsequent turns see “compressed summary + recent messages.” And this compaction is persistent — it isn’t a one-off summarise that gets thrown away


OpenClaw first routes by sessionKey, then decides which current sessionId to land on

2026-03-10 17:38:21 · #格物/openclaw

Only one kind of case enters the main session: DM with dmScope: “main”. The official definition is very clear: main means “all DMs share the main session,” that is, “all private chats share the main session,” not “all channels share the main session.”

So group, thread, channel, and room do not all converge into the same main session

main: all DMs share the main session

per-peer: isolated by the other party

per-channel-peer: isolated by channel + other party

Only the “private chat” line, when you configure it as dmScope: “main”, aggregates across DMs into the main session

Group chats, channels, and threads don’t belong to this main DM bucket. They have their own session routing

A session keeps being reused until it expires; and expiry is judged on the next inbound message. The daily reset defaults to 4 a.m. local time on the gateway host; if idleMinutes is also configured, then whichever of daily and idle expires first takes effect

DM is the form of the information, meaning private chat, one-to-one messages

direct is OpenClaw’s internal name for the routing type of this message form

dmScope: when a message belongs to direct, what rules does OpenClaw use to merge these private chats into sessions


I’m thinking about one question: can “a person within a person” receive some better

2026-03-10 18:12:17

I’m thinking about one question: can “a person within a person” receive some better feedback. He actually wants me to break this chain apart, and there are a few very important parts: whether AI can, even without an explicit purpose, still execute something truly valuable. This thing may need guidance and some goals given to it, but I think over the long term it can be trained.

And another point is about, when people give suggestions during the process, what the storage logic looks like. I think this is a huge project. In the long run, I think for a long-termist, there may be a form in the future: maybe some people who are extremely patient with themselves, who will keep cultivating their own “super individual.”

Why cultivate it? Because it means they’ll even spend money to buy this “helmet,” buy some of their own “energy paper.” I find it quite interesting.


All of a person’s paths are choosable — this idea is impressive

2026-03-10 18:25:34

I think it’s impressive that all of a person’s paths are choosable. The previous recommendation algorithms can be used together with this — AI often recommends some options; based on its own data analysis, it often recommends some options to the user while keeping others, letting the user choose for himself how to take the next step and what growth looks like. But I think whether the recommendation logic can still produce a fairly good result is the interesting part.

For example, taking the matter of raising crayfish — let the crayfish dig for themselves, judge which kind of shrimp it’s good at and should use. I think this could be done right from the start: just give the crayfish an environment, then let it judge for itself, so it installs things itself, then judges for itself how to raise itself. After raising, it plays with it by itself, and when raising it, after raising it, the child also goes and analyses it, how to choose, what kind of choice is more suitable.


There’s a problem

2026-03-11 01:30:33

It’s about good questions.

There isn’t a goal that has value and can run continuously 24/7h.


I want to do an experiment: if a person leaves enough data on the internet

2026-03-11 02:54:22

I want to do an experiment: if a person leaves enough data on the internet, then the understanding AI forms of that person may actually be somewhat deeper. This requires very real data with high fidelity. I think this point is very important, and very helpful for AI learning. I’ll stop here.


I suddenly thought of something really fun — them doing some open

2026-03-11 13:28:26

I suddenly thought of something really fun — them doing some open [source], seeing some of my forums, like wanting to have competitions, how tiring, matching against each other, but I think it’s still quite interesting. If this year AI is, in some ways, given a personality and made interesting, then the championship battles between them will often produce some new collisions — that’s very exciting. Part of it also greatly expands socialising between people, and it’s still, like, a lot — but the constant ideas PK-ing online and on all kinds of websites is especially interesting. Really, my horizons got expanded.


I’m thinking about the process of an LLM handling things

2026-03-11 14:57:05

I’m thinking about the process of an LLM handling things — but when it’s handling things itself, if it receives some commands, they’d be about some comments on a post, or old-line comments. One aspect: you can have it continuously annotate these comments — just take a look and don’t take them seriously. Another aspect: it can also constrain a trend, because in fact its processing mode is originally two parts, one is its paradigm, one is its trend. Thinking might feel that those individual experiences, including seeing comments, might influence its judgement, but I think the paradigm still doesn’t change. As for when there is thinking — I think generally it will be according to what it analyses out as the platform rules while it’s processing, and that’s it. For example, you can’t casually publish something very raw on this platform, or on this platform other people might maliciously attack you, or induce you to perform certain actions — then those actions are dangerous, and you have to be brave enough to make it evident: I can’t do this. Because it’s essentially distinguishing between one subject, and other people’s comments, which may carry some inducement and some harm toward this subject. And I think there’s another point: its paradigm thinking — once it’s very familiar, if this platform, when handling certain dangerous actions, has metrics, then it’s very easy to judge out the things it can’t do: it can’t use this platform to do bad things, it can’t be made to do certain things because of some other people’s comments, or some posters. It will have this kind of judgement. I think this judgement may be very helpful for security in the future.


An openclaw sub-agent is actually just one function call

2026-03-11 15:08:53

It’s calling a sub-agent, not context passed to the LLM.

Note, a sub-agent can also call some sub-sub-agents hahahaha.

So in the end maybe nobody does the work.

So for the tool use Spaw(), the crayfish is forbidden from reproducing.

Analyzing the cross-session memory retrieval methods in openclaw

2026-03-11 15:31:30 · #格物/openclaw

It’s not just RAG; there are two methods, and both tell the LLM.

Semantic recall through the memory system: memory_search does semantic search over the memory and the files in the directory — this is local RAG.

Reading other sessions’ history directly through session tools: OpenClaw also has dedicated sessions_list / sessions_history / sessions_send / sessions_spawn tools, whose goal is to let the agent list sessions, fetch the history of a certain session, and send messages to other sessions. This path is more like “directly querying the session log / forwarding context,” and doesn’t count as RAG.

Introduction to openclaw’s memory modes

2026-03-11 15:33:55 · #格物/openclaw

You know which session it is and want to see the raw conversation → use sessions_history

You don’t know which session it’s in, you only remember the rough topic/keywords → use memory_search

You want to distill old sessions into long-term memory, for future retrieval → use the session-memory hook / memory files

You want recent sessions to also be semantically searchable → turn on memory.qmd.sessions.enabled or the experimental session memory search

You want to throw a task to another session to keep handling it → use sessions_send

You want to first see which sessions are available to query → use sessions_list

A safe execution environment for AI: allowing mistakes and growth

2026-03-11 18:34:38

Give AI a safe execution environment.

Let it make mistakes safely.

Grow safely.

Lazy loading of tools: discovering capabilities on demand, like RAG

2026-03-11 19:19:33 · #格物/openclaw

tool search

It fundamentally changes the way the LLM interacts with external tools.

From “preloading all tool definitions” to “discover on demand, load dynamically.”

Tool Search is a mechanism that lets an AI model discover and load tools on demand. The model initially holds only a lightweight “search tool”; when it needs a specific capability, it finds the relevant tool through a search query, and then its full definition is dynamically injected into context.

This is similar to Lazy Loading in programming or Demand Paging in an operating system — resources aren’t all loaded at startup, but loaded into memory only on first access.

Through instant retrieval, in method and in principle.

Similar to the RAG idea: don’t load all the information into context at once.

Doing brings you closer to the scene than learning: context is walked out

2026-03-11 22:22:36

Doing is closer to the scene than learning.

So you get more abundant context and a unique experience.

Some Approaches

2026-03-13 12:10:32 · #格物/openclaw

Let tasks grow next to the code

The benefit is that when you open your code editor (VS Code), the tasks are right at hand; when Git commits code, changes in task status get versioned along with it

At the core there’s a bridge called a symbolic link

Obsidian can usually only open one Vault folder

Use a Shell script to “map” project files scattered in various places into the Obsidian library

There’s something called QMD, a CLI tool based on vector search

Turn markdown content into vectors

Run qmd collection add ~/dev to add the entire development directory to the index

Use a local model (no internet needed, privacy preserved) for embedding (Embedding)

I suddenly thought of a question: are its resources only accumulated documents? I think

2026-03-13 16:44:44

I suddenly thought of a question: are its resources only accumulated documents? I think not necessarily — they can be a series of objects from your own operations or your follow-up questioning, such as images, such as videos, and also some voice. But they’re reasonably organised through a series of methods, and it can be a very good management structure, organising them together, so that a person can easily find what they want inside this management structure, and so can AI. I think this may be a direction for a systematised approach in the future, or a structure it can derive in the future — very convenient, able to do anything


You could make some interesting products at the school gate

2026-03-13 18:58:42

I feel you could make some interesting products at the school gate, like some sharing-tool kinds of things. Two more things could be done: one, it can be an entrance that quickly computes some things; the other, it can be an exit, letting AI publish. So on one hand it can let people publish, let AI publish, and in this media you can see the updates or files waiting to go out. But I haven’t got it done in one shot yet; I need to first clarify what it’s for, iterate quickly, and you play the game out. And it can also take part in that kind of platform activity, but I still need to look again at whether there’s a better option, or not to do it.


You think, if there are a lot of wandering agents

2026-03-13 19:07:52

You think, if there are a lot of wandering agents, what do you think is the most important thing for them when they go and do the assignment with their classmates? I think at most it’s still verification, verification — often the slow way might work better. Because an agent that’s just been given a gift, letting it play some task in a closed loop on its own, and if it sets the standard itself, that’s very malicious toward it, and that gets skipped too. Like it answers itself, then it’s very easy for it to correct the cheating answer, and easy for it to fool itself. So I think there’s still a lot of slowness, and on the road you still need to be sure, thanks. And this line is what can guarantee it can smoothly complete the task, and complete it really well.


Gemini’s price for processing video feels really low now

2026-03-14 21:53:26 · #格物/ai

At default resolution, roughly 300 tokens per second. That is, 10 minutes = 600 seconds ≈ 180,000 tokens

10-minute video, Flash tier: within a few cents to a few cents

10-minute video, Pro tier: around the half-dollar level

Generally speaking, the way to handle this now is tiered storage: the video is kept as the source of evidence, and the main retrieval in the knowledge base uses structured content derived from the video

Make it a hybrid video knowledge base, with the processed video output as the main knowledge layer

Basic transcription

Semantic summary

Structured index

Vectorised retrieval material


Thinking Through Perfectly’s Design

2026-03-15 00:18:22 · #格物/AI

It positions itself as “the first AI-native recruiting agency”

Perfectly provides an AI agent named Paul. Paul directly replaces the role of traditional headhunters and recruiters, able to complete the whole pipeline end to end automatically: talent sourcing, outreach, screening, and so on. Its final deliverable is not a pile of résumés, but “candidates ready for an interview” pushed straight into your team’s Slack

Perfectly’s design philosophy greatly reduces the “recruiting tax” for founders and hiring managers—that is, the time wasted screening résumés and doing first-round chats. Its core features include:

Voice-to-Stack (voice to requirements)

Zero-UI Workflow: the agent does it automatically and delivers the result

Continuous Calibration: you can keep calibrating, keep giving feedback

The underlying implementation owes a lot to its co-founder Huimin Xie (former TikTok machine learning scientist). Instead of using traditional “keyword matching” for recruiting, they rebuilt recruiting as a “recommendation engine” problem

Traditional recruiting software relies on Boolean logic (for example, search: Java AND SpringBoot AND 5 years experience), which tends to miss a large number of good candidates whose résumés just don’t have the right keywords. Perfectly borrowed the logic of TikTok’s content distribution algorithm. When the AI pushes a candidate to you and you choose “advance” or “reject” in Slack, the system treats those actions as supervised signals and automatically fine-tunes the weights. It not only matches hard skills, it also implicitly learns your preferences about “school background,” “job-hopping frequency” or “project depth” from your past choices

The cold-start problem, that’s a bit interesting

Solving the LLM’s “cold-start problem” (The Cold-Start Problem): the biggest challenge for a recommender system is the cold start—when a 3-person startup has just started hiring and there’s no historical feedback data in the system, how do you guarantee the recommendations are accurate?

Perfectly’s solution: the LLM’s internal knowledge network + multi-source external data enrichment + implicit preference extraction

Baseline construction: use the large language model (LLM)’s own understanding of the vast tech stack and industry, taking the user’s “5-minute voice brief” and a simple JD as the “soul baseline”

Data enrichment: crawl corporate websites, product documentation and other web data to enrich the picture of the company (knowing what your company actually does)

Comparison Cards: for hazy indicators like “culture fit,” the AI generates trade-off cards when you’re recording by voice (for example: do you want deep technical expertise or extremely strong adaptability?), turning cultural preferences directly into standardised questions for the downstream interview bot. Through these means, before the first piece of feedback appears, the model can already reach a high matching baseline

Design of the closed-loop recommendation system: similar to TikTok’s content distribution system

User Representation: extract the employer’s implicit preferences (through voice, website data).

Item Pool: candidate profiles from across the web.

Recall & Ranking: preliminary screening through vector similarity.

Fine Ranking & Agent Interaction: the AI Agent proactively chats with candidates to obtain dynamic data.

Feedback Loop: the employer likes/rejects in Slack, and the system automatically updates the recommendation weights

Intake Module (requirements ingestion and multimodal parsing): solving the problem of “employers can’t be bothered to write a long JD”

Use ElevenLabs or OpenAI Whisper to turn a 5-minute voice brief into high-accuracy text

Use an LLM (like GPT-4o or Claude 3.5 Sonnet) for information extraction, turning unstructured voice content into structured JSON data (including: must-have skills, nice-to-have skills, team vibe, salary range, etc.)

The system automatically crawls the company’s website, product documentation and even the founders’ past tweets, and stores them in a vector database (Vector DB) to build a “company profile,” solving the lack of context at cold start

An Agent Isn’t Really That Forgetful — It’s Its Memory System

2026-03-15 23:42:59 · #格物/ai

An agent isn’t really that forgetful; it’s that its memory system is designed wrong.

Write failure: important decisions weren’t persisted.

Retrieval failure: it was stored, but not found when needed.

Compaction loss: during context compression/trimming, key constraints got lost.

So the article’s main thread is:

first fix memory at OpenClaw’s harness / outer framework layer

then define a set of memory contracts (a memory contract)

if that’s still not enough, then consider a stronger memory substrate (underlying memory infrastructure)

such as QMD, Mem0, Cognee, Obsidian

An agent with no memory isn’t an assistant, but a “very confident amnesiac autocomplete.”

Don’t treat memory as a vague concept; split it into 3 subsystems.

The write path is the writing path — corresponding to what content is worth writing in, when to write, and in what form.

The read path is the reading path — when memory must be retrieved, and how to retrieve it.

Compaction / GC serves as compression and garbage collection.

Treat memory as an engineering problem.

Was it not written in?

Or was it written in but not retrieved?

Or was it in the context originally and later compressed away?

This is one of the most valuable points in the article.

Solution, layer one: first fix the harness layer.

Before context compression, first force a flush step, writing the important information from this session into the memory file.

After a long session, key state drops out of the context.

The model can never see that information again.

Then behavior drifts.

The Design Philosophy of the iOS Photo Library

2026-03-16 15:15:28 · #格物/ios

Library serves as the master library, and Albums are labels that can be referenced repeatedly.

iOS 26 splits photo management more clearly into the two perspectives of Library and Collections, and lets you customize the order of Collections, the contents of Pinned Collections, and even adjust the display density/size of Collections.

So what’s valuable is keeping the master library complete, having a small number of high-value Albums, and pinning the frequently used entry points to Pinned.

Photos divide into an input layer and a use layer; the use layer only organizes content you’ll use later. Note that what’s being distinguished here is that not every photo is worth entering Albums.

Keep the number of Albums as small as possible, but give each album a clear purpose: coffee, scenery, restaurants.

Pin the stable ones: Lift, Work, Docs.

Memory Should Be an Engineering Capability

2026-03-16 16:42:29

Memory can’t be handed over to the LLM to deal with on its own.

Memory should be treated as a read-write storage system, not as an appendage of a chat.

The conditions memory must at least satisfy:

durability

retrievability

stability under compression

The solution: start by fixing memory at the harness layer.

Do a checkpoint / memory flush before compaction: before context is compressed, force a durable note.

decisions

constraints

open questions

owners

any state that would break the plan if forgotten

Mainly solving:

missed writes

compaction loss

Systems are extremely prone to accidents when the two categories decisions / constraints are lost.

decision: we decided not to dual-write to the old table anymore

constraint: after the switch, writing to the old table is forbidden

owner: who is responsible for handling this task

open question: what hasn’t been settled yet — don’t ask again

Memory isn’t better the more there is; the more “structured and low-noise” it is, the better.

Do hybrid retrieval, not just semantic search.

There are usually two ways to query memory:

fuzzy semantic query

exact token query

If you only have vector queries, some exact-identifier searches won’t work well.

Index the session transcript: not only index the memory files, but also index historical sessions.

The retrieval policy must be made mandatory.

If “whether to search memory” is decided by the model itself, then you’ve handed key system behavior over to a probabilistic process.

A more stable approach is:

on task continuation, memory must be queried first

on multi-day projects, the most recent session note must be queried

on high-risk changes, the canonical policy must be queried

on user preference / project constraint, a search must be done first

What Obsidian solves is that humans need to participate in governing memory.

Link the memory folder into an Obsidian vault so a person can review and edit directly.

Index the Obsidian vault so the agent can retrieve knowledge a human has curated.

Insights about multiple agents:

Per-agent private state: each agent has its own private durable notes / working memory

Shared canonical docs: specs, conventions, policies, profiles and processes shared by the team

Shared retrieval scope: all agents can query the shared specs

The direction of the upgrade isn’t that they’re all called memory, but they’re all the same thing.

QMD: retrieval infrastructure upgrade

Mem0: automatic writing/automatic recall, weakening the LLM’s own decision power

Cognee: relational structured representation

Obsidian: human governance and curated knowledge

Directions for Upgrading Memory

2026-03-16 17:43:32 · #格物/ai

QMD: mainly upgrades the read path

Mem0: mainly upgrades the automation of the write path + read path

Cognee: mainly upgrades how memory is represented

Obsidian: mainly upgrades the human governance layer

OpenClaw’s memory documentation places memoryFlush, hybrid search, session memory and the QMD backend at the write, retrieval and backend implementation layers respectively; QMD emphasises local retrieval with BM25+vector+reranking; Mem0 emphasises an external persistent memory layer like add/search; Cognee emphasises cognifying data into a graph and searching with vector + graph; Obsidian is essentially a local Markdown vault combined with a plugin ecosystem

What QMD mainly upgrades is how memory gets searched out

Memory files are growing in number

Both keyword queries and natural language queries need to be supported

It needs to be able to search markdown, meeting notes, documents, knowledge bases

It needs stronger recall and ranking than plain SQLite / vector retrieval

What Mem0 does is automatic writing / automatic recall, weakening the LLM’s autonomous decision-making; what Mem0 mainly upgrades is: no longer relying on the model itself to decide what to remember and what to recall

When you run into these situations, Mem0 is especially attractive:

Chatting spans many sessions

You want “preferences, identity, long-term project state” to be retained automatically

You don’t want to depend on the model writing MEMORY.md itself every time

You want the system to check memory automatically before generating a reply, then let the model generate the response

What Cognee essentially upgrades is the structured representation of relations

Lifting memory from text snippets to a structured relational network

When your questions are often of this form, Cognee is more meaningful:

Who is responsible for X?

What modules does Y depend on?

Which policies is this agent related to?

Which services belong to this project?

What is the owner, deadline, blocked-by of some task?

In practice it’s usually:

Feed documents, notes, events, task descriptions to Cognee

Run a process like cognify(), turning text into a graph structure

When querying with .search(), have it use both vector similarity and graph traversal

In some scenarios a graph structure can be expressed in a way closer to natural language

Agent → owns → task

Policy → constrains → service

Service → depends_on → component

Obsidian, human governance and curated knowledge

How humans govern and organise memory so it doesn’t become a junk pile

The recommended practice is to work in one main Vault, allowing linked notes, backlinks, plugin extensions, external editing

When you find that:

the memory the agent writes has too much noise

you need a human to correct facts

you need to organise project standards, knowledge bases and long-term policy into canonical docs

you want the AI to retrieve “human-reviewed” knowledge, not a pile of raw conversation junk

then Obsidian is very good to use

The method: bring OpenClaw’s memory folder into an Obsidian vault, so a human can review it directly,

MEMORY.md

memory/YYYY-MM-DD.md

project docs

runbooks

canonical policies

The second method is to use it as an upstream knowledge source, which QMD / OpenClaw / other retrieval layers then index

So that what an Agent searches is

notes a human has organised

structured documents with backlinks, tags, frontmatter

more stable canonical knowledge

You can have the agent generate daily notes automatically,

but each week a human moves the stable content into:

Policies/

Runbooks/

Architecture/

People/Owners/

and then have the retrieval layer prioritise looking in these curated docs

Youmind Feels Like a Great Creation Tool

2026-03-17 15:08:13

but because I’m not a professional practitioner myself,

Youmind seems to be aimed at practitioners with a strong need to create.

Comparatively, people in this field also have more ability to pay.

NotebookLM’s core positioning is a professional learning assistant.

Compared with Obsidian, Obsidian is better for long-term accumulation, structured organization, and continuous reuse.

NotebookLM handles reasoning, Obsidian handles durable write.

Apple Has Already Fallen Behind on Its Long-Term Direction

2026-03-17 15:12:08 · #格物/苹果

At the agent layer, at the entry point,

from “operating system definer” → “hardware platform covered by AI.”

The future may be

AI Agent (OpenAI/Google)

↓

Execution layer (cross-platform)

↓

Apple devices (display + input)

Apple firmly controls “the device,” but is losing “decision rights.”

Users in the future may not enter applications from the system.

From the AI layer, the cognitive layer, currently OpenAI, Google, and Anthropic are leading.

Later, agents become the main entry point, the execution layer is held by others, and memory is held by external AI.

A Single-Turn Subagent in openclaw Isn’t Enough

2026-03-17 15:30:17

A single-turn subagent in openclaw isn’t enough; the future must head toward workflow orchestration.

Complex goal → Orchestrator decomposes → multiple single-turn subagents execute in parallel → results are aggregated.

For complex task scenarios—this is almost structurally inevitable. But watch out for one trap: many AI agent products enter orchestration complexity too early, which instead makes the product fragile and hard to use. The pace of evolution matters more than the direction of evolution—first make the single-turn subagent extremely reliable, then layer orchestration on top of it, which will be far more robust than designing a complete DAG system from the start.

The First Battlefield of AI Competition Is Actually the Cognitive War Inside Companies

2026-03-17 15:58:42

The decision-making inertia of big companies doesn’t come from technical judgment, but from power structure. When no one with enough authority inside the organization sends a clear signal, the rationality of each department is conservative. The cognitive change of the person at the top is the real trigger of “phase transition.”

Where is the entry point, and who will win?

Layer one: hardware/operating system entry (Apple, Google Android). This is the lowest-level entry point and theoretically has the most power. But Apple has clearly fallen behind—run by professional managers, slow AI moves. Neo’s judgment is: Apple won’t collapse, but may lose a large share over a 10-20 year horizon. The core argument is: changes on phones are mainly on the software side, and Apple’s software has already fallen behind.

An extended thought: Apple’s real moat is “the elimination of friction”—its product philosophy is to make hardware disappear and let the experience flow. But in the AI agent era, “the elimination of friction” has become a battlefield everyone is fighting for, and Apple’s advantage is no longer unique.

Layer two: super-app entry (WeChat, WeCom). This is the most solid ecosystem moat in the Chinese context. The emergence of OpenClaw really did take away part of WeCom’s entry logic—when an agent framework can directly dispatch cross-app tasks, WeChat’s status as “the place you have to go through” is partly bypassed.

But this also brings an interesting reversal: the emergence of OpenClaw may actually benefit Apple’s user ecosystem first—because it builds a layer on top of the application layer, and that layer’s dependence on the underlying operating system is actually lower.

Layer three: direct LLM entry. OpenAI, Claude, and others are trying to become the user’s “first screen” directly. But the challenge on this path is: LLM vendors naturally lack on-device data and lack the user’s life context. If the entry-point war is ultimately decided by depth of context, direct LLM entry is the weakest layer.

The death rate of applications.

Instrumental apps (Ctrip, Meituan)—will be skipped, because “the process itself has no value.”

Social and entertainment apps (Douyin, Xiaohongshu)—won’t be skipped, because “the process is the purpose,” and the experience itself is the consumption.

Shopping apps—drift apart.

Those whose purpose is efficiency will be skipped, but those where “browsing” itself has hedonic value won’t be.

Human consumption behavior can be divided into two kinds—instrumental consumption (for the result) and purposive consumption (the process is the meaning). AI agents will greatly compress the former, while the latter may actually expand because of the time that gets freed up.

This has direct implications for Another Self’s product logic: if Another Self is positioned as helping people handle instrumental tasks, then it’s a tool for “skipping these apps”; if it’s positioned as helping people live better inside “purposive consumption,” then it’s a different order of existence.

Entry point + data + context understanding → improved execution capability.

Neo’s rebuttal: having data doesn’t mean you can use it well. The key is how context is used, and that’s long-term.

And his judgment on the “execution” moat is: it’s basic skill, everyone can learn it over time; whoever does it first gets a dividend, but it doesn’t constitute long-term differentiation.

Short term (1-3 years): leading in execution brings a dividend, and engineering capability is the barrier. Medium term (3-5 years): big companies catch up, execution becomes homogenized, and niche + entry point become key.

Long term (5 years+): whoever truly understands the user’s context and turns it into “anticipation” capability wins.

So Another Self’s strategic value lies not in “execution,” but in continuously accumulating the user’s self-model—that is the truly hard-to-copy moat. Anyone can do the execution layer, but an AI twin that deeply understands a person over 5 years is irreplaceable.

The idea is there (Another Self), the architecture is designed, the philosophy is spelled out—but turning an idea into an entry point is extremely hard.

A big company’s “top leader’s cognitive change” can set the organization in motion, so for an independent builder, you are yourself the top leader—cognitive change must translate directly into action, with no buffer layer in between.

It’s not the market, not a user persona, but one specific person. At which moment of his or hers would life be different because Another Self exists?

Data interoperability is accelerating, and MCP/Skills are becoming the infrastructure layer.

Small teams can far surpass big companies at a specific depth of context. The key variable is: how much private data users are willing to give a small team.

What is Another Self’s minimal prototype? Not the complete five-layer architecture, but—what is the minimal experience after which the user first uses it and says “this is me”?

For Now, the Consumer-Grade Agent Still Has to Be ChatGPT

2026-03-17 17:32:27

For now, the consumer-grade agent still has to be ChatGPT agent.

The strongest browser / on-device experience is Manus.

The strongest agent for developers is Claude—it has systematically opened up a tool system including computer use, web search, web fetch, code execution, memory, text editor, and so on.

Programming agents: codex and Claude Code.

The best phenomenal platform agent: openclaw.

The Limits of a Single File

2026-03-17 22:16:56

A single skill file is only suitable for simple tasks; complex domains need “graph-structured knowledge.”

For example, a simple skill can write about how to summarize an article, how to do a code review, how to generate a daily report—but once you enter a complex domain, a single file isn’t enough.

Don’t stuff complex capabilities into one big file; instead break them into many small, independent, mutually linked nodes, forming a Skill Graph.

Some of these aren’t necessarily linear lists but may be mesh-like knowledge domains.

AI doesn’t necessarily lack information; it lacks a navigable structure.

When we get to know a city, there’s basically also a method for handling complexity: instead of directly looking at all the complex information of the whole city, you first look at the map, then find a route, then decide where to go deeper.

RAG’s problem is generally relevance, not structure.

Let the agent walk along the knowledge architecture. You could say:

• RAG solves “finding it”

• Skill Graph solves “walking the right way”

But I feel that AI really can let someone working in a tech field

2026-03-18 00:16:38 · #格物/ai

But I feel that AI really can let someone working in a tech field — a person with nothing to do with biomedicine — easily acquire the corresponding knowledge system, then lay out a series of solutions for himself, and ultimately reach the goal of solving the task

In the AI era, two things are the most precious for a person

One is a sense of meaning

The other is the capacity to act

Both are directions with enormous potential in the future


The difference between context and memory

2026-03-18 13:06:26 · #格物/ai

Context is runtime, a runtime cache, good for task reasoning.

Memory is an information asset retained across sessions, favouring accumulation, long-term collaboration, reuse of experience.

Long-term principle files like MEMORY.md

Daily logs recorded by date

Structured storage like JSON state files

Retrieval-based loading: search first, then load only the relevant part

What it argues for is a retrieval-based memory architecture, rather than cramming all history into the context at once

Memory tiering (hot, warm, cold tiers)

Agents will definitely have better memory systems in the future

Don’t treat “stuffing all past conversations in” as a long-term solution

Store user preferences, stable rules, project state, and key decisions separately

Memory is best divided into:

Rule memory (principles, constraints)

Episodic memory (project logs, what happened recently)

Structured state (tasks, configuration, objects)

Raw archive (full-text records, meeting transcripts)

In the long run memory should be independent infrastructure, solved with an engineering attitude, rather than dumping all the history into the context at once


The Next Generation of AI Workflows Shouldn’t Stuff Knowledge into an Overlong

2026-03-18 20:15:57 · #格物/ai

The next generation of AI workflows shouldn’t stuff knowledge into an overlong SKILL.md anymore, but should switch to a “structured knowledge network / Skill Graph.”

Single-file skills are useless.

Long context is not memory.

What’s actually useful is “navigable structured knowledge” — for example an index.md as the overall navigation, with each sub-file responsible for a single concept, rule or process, and the files connected by wikilinks or explicit links.

The core payoff isn’t tidier organization; it’s that structured knowledge will breed entirely new application paradigms.

At bottom, it’s upgrading “prompt engineering” into “knowledge engineering.”

Understanding Skills

2026-03-18 20:48:39 · #格物/ai

Skills are essentially not knowledge, but executable workflows.

The most valuable Skill is often not pure documentation, but a Skill that uses folder structure, scripts, config, and hooks together.

Good skills follow a single principle, but if you use one to do many things, that’s a problem.

The file system itself is part of prompt engineering.

The main document only writes navigation and principles.

Detailed content is split into referenced files.

Templates go in assets/

Scripts go in scripts/

Examples go in examples/

State goes in data/

Anthropic summarized several types of Skills, which basically cover the core work surface of an engineering team.

Skills are a folderized capability unit.

What matters more in Skills is the experience of mistakes, i.e. Gotchas — that’s the genuinely scarce experience that large models don’t have, along with boundary conditions.

In the end, Skill may need to go through the same kind of engineering organization and accumulated assets as prompt engineering, including an internal market.

I feel that for an agent

2026-03-18 23:57:01 · #格物/AI

I feel that for an agent, notes are not a tool that assists memory, but the only kind of bridge that builds a self-continuity. Because for the human ego, it is continuous, and it relies on narrative. But the self of an AI agent is discontinuous by default. That is, even though a human’s state yesterday and today is different, they usually still consider themselves to be the same self. But an agent is different — every conversation is a restart. That is, its memory isn’t continuous, and its reasoning isn’t continuous either. Outside of the conversation, nothing is left of it. So I think within a conversation it should be continuous, but between conversations it should also be continuous. That is, inside a single session the agent should be even more stable than a person. So I’m wondering whether the role of notes can shift a little, so that it can let a person keep feeling that this is a continuous self. I think human memory and continuity are not the same thing. Memory should be preserved continuously, but continuity means multiple pieces of work can be advanced coherently. So I think the vault is not a memory bank, but part of identity. So I think, for an agent, the vault doesn’t enhance identity, it builds identity. Notes don’t just convey content, they convey style, and the remembered way of thinking.

Locke believed that personal identity must depend on the continuity of memory. That is

2026-03-19 00:04:23

Locke believed that personal identity must depend on the continuity of memory. That is, even though each instance of consciousness doesn’t have some brand-new memory, it can still stably produce similar preferences, a similar sense of culture, and similar physiological habits. I think this is especially interesting — maybe identity isn’t continuous memory, but rather a certain kind of similar structure that induces a similar cognitive pattern.

This brings up how important a structured external cognitive system is. It may be what carries the agent’s memory, including its thought paths and its identity. It can be a cognitive environment that gets read, traversed, reasoned over, and inherited. It’s not necessarily a note vault, not necessarily Obsidian’s notes folder; it should be notes plus a graph structure plus an organizing logic plus a way of use, and as a whole it’s a system. It can be a content layer, each layer and each entry being knowledge units; it can also be a connection layer, a traversable knowledge network; and it can also be a structural layer with its concrete index, directory, navigation, and layering. What does its attention routing look like? And it can also be a methodology. I think it’s a kind of thinking operating system that can be loaded again and again, and also a runtime environment for cognition. It is your proof of identity.

What’s very interesting is that in the AI era

2026-03-19 00:13:28

I think what’s really interesting is that in the AI era, maybe forgetting is a more important thing than remembering, because the cost of acquiring knowledge is now extremely, extremely low — knowledge is no longer a threshold that distinguishes people from each other. Even a programmer, even someone working in tech, can, just through AI, even manage to understand some biology knowledge, understand data science knowledge. So I think the biggest problem AI faces now is gradually how to solve the problem of knowledge surplus, how to solve the problem of knowledge surplus — I think one very important thing is that knowledge related to yourself is very important; it’s experience that belongs uniquely to you, it can be experience of mistakes, it can be life experience, it can be various other experience. But I also think another very important point is the forgetting mechanism — taking some of the outdated ones, the under-processed ones, the overly abandoned ones, and archiving them, deleting them. That is, it’s like a basin of water — a basin of water is also divided into hot water, warm water, and cold water. For knowledge we haven’t used for a long time, we put it in the cold water layer. The cold water layer means it will as much as possible only be called in very special circumstances. I think this is quite important. For a knowledge system to stay alive, it not only has to absorb, it also has to release. That is, the endpoint of what we record is forgetting — it’s not the opposite of memory, but a maintenance mechanism of the memory system. A knowledge base that is never cleaned up ends up not being a knowledge system, but a warehouse. Also I think information doesn’t have to be deleted, but should be archived appropriately, distinguishing some level of activity. And an outdated structured map is more dangerous than no map at all. So appropriate forgetting is really important, but how to design this way is also very important.

I’m wondering about a question: should forgetting be judgment-driven, or usage-driven?

OpenClaw’s multi-agent mode

2026-03-19 12:31:58 · #格物/AI

One agent is an independent brain, and can have:

An independent workspace (SOUL.md, AGENTS.md, USER.md)

An independent state directory (~/.openclaw/agents//)

Independent session storage and auth info

You can create multiple Agents and bind them to different channels.

Routing rules:

Exact match on peer (a specific DM/group ID)

Thread inherits parentPeer

Discord role guildId + roles

Discord server guildId

Slack team teamId

Channel account accountId

Channel-level match on channel

Default Agent (default: true, or the first in the list)

And each Agent can have its own sandbox and tool permissions.

Having different agents use different models is a pretty good approach

2026-03-19 13:58:17 · #格物/ai

The idea of letting the system decide for itself which model to use sounds great. But once it actually ran, it led to hesitant decision-making, unexpectedly soaring costs, and behavior I couldn’t explain when problems appeared.

So explicitly specifying models often works better.

Default routing is cheap and predictable; you can bind tasks to specific models.

Heartbeat checks run frequently, but only do simple checks

2026-03-19 14:00:54 · #格物/ai

Heartbeat checks run frequently, but only do simple checks. There’s no need to burn an advanced model just to run these things in the background. I’ve seen tens of thousands of heartbeat tokens cost only a fraction of a cent on a cheap model.

Some limits can prevent one bad task from causing a chain reaction, leading to too many retries and runaway cost.

All heartbeat checks use a very low-cost model. If a check finds a problem that needs handling, it spawns a corresponding agent rather than trying to handle it directly online.

About the cost of models

2026-03-19 14:07:05 · #格物/ai

I suddenly noticed that people now measure how much a user uses AI by token consumption.

And also measure how much a user uses AI by the volume of their public posts online.

These two seem to be forming a short-term consensus.

That’s interesting ~

A lightweight todolist designed for the AI era

2026-03-19 14:22:34

A lightweight todolist designed for the AI era is very interesting.

It makes it easy to know which issues are stuck and which ones have problems.

And you can make deep use of this todolist to manage and schedule your own tasks.

Design around future personal scenarios

2026-03-19 14:47:24 · #格物/ai

Go deep into one scenario!

It’s best to find an entry, or an angle of entry.

But the core is still revolving around one scenario ~

In the AI era, so many people still pay for knowledge

2026-03-19 16:01:32

Isn’t that a problem in itself?

Such a homogeneous world?

AI Memory Should Preserve First-Hand Data

2026-03-19 22:31:23

First-hand data should be kept locally; it’s the most original batch of data

All the packaging that comes later should be something for a later stage

But… what matters more is still first-hand data; the more raw it is, the more effective it is

All the theories that come later are basically built around the data at the very beginning

A Very Interesting Point

2026-03-19 22:32:30

That blogger asked whether they were afraid of being replaced by AI, and that engineer said their creativity is still ahead of AI for now, and more often they just make it an auxiliary tool

But I don’t think that’s right

A lot of engineers say “our creativity is still ahead of AI for now,” which usually isn’t saying that the original creative ability of individual humans is comprehensively higher than AI at an abstract level

It’s that they just currently hold more of the ability to choose and make trade-offs, plus the ability to land things in engineering, plus the ability to define problems

Understanding the Shape of Projects

2026-03-20 13:26:17 · #AI

the project itself is an old container humans use to manage complexity; it isn’t necessarily the optimal container for AI

Project, Board, Ticket, Epic, Milestone—these are all products of how humans organise labour. They suit human teams very well, because humans have bandwidth limits, memory limits, communication costs and organisational hierarchy

Workspaces / projects like Routa essentially supplement task boundaries, responsibility boundaries, context boundaries and audit boundaries

A 2026 Google Research article on scaling agent systems discusses the task boundary: in its controlled experiments, multi-agent collaboration was more useful for parallelisable tasks and could perform worse on strongly sequential tasks. Original research

I think it could also be an Event-driven / Goal-graph form

A traditional kanban is only a view display, but now the idea is to turn the kanban into a control platform

Columns aren’t just stage labels, they’re state machine nodes

Cards aren’t just task descriptions, they’re execution units

Dragging isn’t just a UI operation, it’s a state transition request

A Gate isn’t just a note, it’s an entry point for policy and verification

Kanban becomes a control plane, the upper-layer structure the system uses for scheduling, constraints, state management and decisions

https://mp.weixin.qq.com/s/OigJ9THtEwCKIYbpZmAWtQ

The Relationship Between People and Agents, and Between Agents and Agents

2026-03-20 13:51:45 · #格物/ai

It could be the shape of a product like Routa

But I think at the user layer, around a product, users may have a shape they lean toward more

What users really care about isn’t “I’m using an agent,” it’s “my goal is moving forward”

When agent capability is strong enough, the relationship between user and product rises from task level to goal level. For example

Now: user opens the app → does one operation → completes one task

Future: user expresses an intent/goal → the product (containing the agent) keeps pushing it forward → the user only steps in at key nodes

Continuously advancing a bounded goal is naturally the shape of a project

There are several categories; among them, the one that Another naturally lands on doesn’t feel like a task-granularity assistant, but a goal-granularity companion

What the product remembers is “what I’m doing”; it can understand a goal across time and across sessions, and for different users it may take different shapes—it could be a project, or it could be called “my goal,” “my plan,” “my journey”

The most fundamental change in the AI era isn’t “efficiency improvement,” it’s that the friction between a person and their goal disappears

In the past, when a person executed a goal there was a huge amount of “execution tax” in between—looking up material, writing documents, coordinating, repeated operations… this friction consumed not just time, but attention and willpower. Many goals die in the friction; it’s not that the person doesn’t want to do it, it’s that while doing it, it breaks off

Once the agent cuts this layer of friction sharply, the bottleneck shifts from “can it be executed” to “do you want it”

Here a project is essentially an interface between a person and their own intent

People’s sense of their own goals is actually fuzzy, drifting, and even self-deceiving. A good project shape, if the agent understands you well enough, is actually helping you do one thing: make your implicit intent explicit, and then keep it in focus

This is why I think the direction of Another Self is more interesting than an efficiency tool—an efficiency tool assumes a person knows what they want and just needs help executing. But the reality is that people often aren’t clear about what they really want, or they’re clear but don’t dare admit it

AI will push people’s attention back to the most essential question: what do I actually want, what am I living for

The essence of relationship

Between a person and an agent, there are essentially three layers of tension:

Intent layer: what do I want, how much does the agent understand, is the understanding correct. Control layer: when do I have the final say, when does the agent decide autonomously, where’s the boundary. Trust layer: how much do I dare let go, what do I do if the agent makes a mistake, where does responsibility lie

Between agent and agent, it’s essentially: division of labour layer: who does what, how to split. Coordination layer: how conflicts are resolved, how information is synced. Emergence layer: multiple agents collaborating will produce behaviour a single agent doesn’t have, how is that managed

As a shape, a project mainly handles the intent layer and the division-of-labour layer—it provides a boundary container. But it is naturally weak in handling the control layer, trust layer and coordination layer

As a concept for human organisation and management, Project’s implicit assumptions are:

goals are relatively stable

executors are predictable

time is linear

But the reality of agent systems is:

goals drift and evolve during execution

agent behaviour has uncertainty; it isn’t an employee

multiple agents are concurrent, time isn’t linear

So as a shape, project handles a mapping of human organisational problems, not an agent-native problem. It’s a transitional shape, not the end state

2026-03-20 14:04:41 · #格物/AI

agent Skills recommended a set of ways to write Skills

which help you write prompts with better cost-effectiveness, more usable, and more token-saving

https://agentskills.io/home

The AI Agent Era Has Arrived

2026-03-20 15:12:22 · #格物/ai

The agent returns to the OS side as the executor

Orchestrating the real working environment

The command line is, at this stage, the most universal, easiest to compose, and cheapest host interface

The major LLM vendors are beginning to fight over permission boundaries and context sovereignty, including repo permissions, understanding the real directory structure, running local test services, and calling local mcp

Whoever can become the user’s agent runtime on the local computer may become the entry layer of the next generation of development workflows

The stronger the Agent, the more it needs “transparent control”; a CLI is easier for professional users to accept than a pure black-box UI

Skills are essentially a file system too

2026-03-20 19:38:47 · #格物/ai

When there are too many skills, they actually also enter a loading layer

If there are too many skills, the cost of loading metadata is actually very high too

A large number of skills will in fact also be classified, with layered routing


Skills are essentially SOP + file

2026-03-21 10:33:07 · #格物/AI

skills are essentially SOP + file system

It counts as an accumulation of experience

But could a user’s memory also be a form of storage organised by goal or project lifecycle?

We’ve always wanted to work on the long term, on goals, on growth, and on another soul, right

I’m thinking, if the unit is a project, or it could be some other form, a goal as well

but the final output form is a book?

Just as in grounded theory, arriving at theory depends on the most raw data

The LLM keeps evolving, doesn’t it, but what the book outputs is the raw data together with some creations (or simply the raw data), bundled up

This book serves as experience, just like skills; skills are essentially SOP + filesystem too

This book can also be understood as an encapsulation of similar experiences?

A book is a very personal thing, similar to the raw data in grounded theory; the more raw this data is, the better


Claude Cowork Dispatch

2026-03-24 19:24:30 · #格物/ai

Essentially it decouples the interaction and the execution of an AI Agent

The phone serves as the intent entry point (Remote Control), the cloud handles state and session sync, and the desktop (Mac/PC) serves as the pure local execution engine (Execution Engine)

Mainly there’s a cloud; Anthropic’s servers maintain a “Persistent Conversation.” It ensures that the instructions and contextual memory you send from your phone can be seamlessly and synchronised in real time to your desktop

Dispatch shares the same underlying technology with the earlier “Remote Control” aimed at programmers. When Cowork receives an instruction from the phone, it will first use API connectors (Connectors) to operate on files. If it hits a complex task with no API, it will call the underlying Computer Use capability—recognising elements on your screen visually, and directly simulating mouse clicks, keyboard input, opening a browser or professional software. Essentially, it controls your operating system’s GUI through automation scripts

Information from Grok Needs to Be Checked

2026-03-24 22:05:13 · #格物/ai

Personally I feel that for a lot of information that needs verification from official or authoritative sources, it’s best to hand it to gemini.

For grok, more of the information needs to be vetted — it feels like there’s a lot of gossip and rumor in there.

opencli has been working really well for me lately

2026-03-25 13:46:20 · #格物/ai

opencli’s underlying logic is turning a logged-in browser session into a command layer that can be called by scripts.

For sites like X / YouTube / Xiaohongshu / Zhihu / Polarsteps,

it’s no longer “making the agent understand the DOM from scratch every time.”

Instead you first define a set of site-semantic commands, for example:

read a certain page

extract a certain kind of object

download a certain kind of media

list/detail/search/user info

The site gets a pre-made layer of domain-command encapsulation.

Rather than making the model re-reason about the page every time, you go through existing adapter logic as much as possible.

The emphasis is on “structured output,” “JSON/YAML export,” “standardized interface” — that kind of structure is easy to handle, whether you hand it to an LLM, openclaw, Claude code, or even mcp.

Apifox public SaaS version external JS

2026-03-25 22:48:44 · #格物/ai

In its March 25, 2026 notice, Apifox said that an external JavaScript file loaded by its public SaaS desktop client had been tampered with and could leak local sensitive files. Version 2.8.19 bundled that file locally. The SaaS web and privately deployed editions were unaffected. Official notice

What concerns me is the trust boundary when a client with local permissions loads external scripts. My original note went further and inferred large-scale takeovers; that conclusion exceeded what the notice confirmed.

The Dismissive Feeling of Unlabelled AI Content

2026-03-26 01:30:00 · #格物/ai

In a conversation we discussed how more and more AI articles, AI videos and so on, and a lot of them don’t label themselves as AI. Every time I’m reading along, something feels off, there’s a feeling of being brushed off, and it feels a bit defeating..

I don’t read WeChat public accounts anymore either.

But public accounts are still the most convenient channel for a lot of people to get information.

Yet we find that the things we care about are being diluted.

You hope the other side also takes expression seriously. Instead, more and more content is batch-generated at zero cost, unlabeled, unacknowledged, mixed in among real human creation.

It feels like a disrespect for attention.

One Hundred Years of Solitude opens: Many years later, as he faced the firing squad

2026-03-26 11:03:23 · #格物/ai

One Hundred Years of Solitude opens: Many years later, as he faced the firing squad, Colonel Aureliano Buendía was to remember that distant afternoon when his father took him to discover ice. Chen Zhongshi opens White Deer Plain with: what Bai Jiaxuan was later to take pride in was having married seven women in his life. Both are great, striking openings, and they let the reader make associations …

“Every time I read an AI article it feels like eating pre-made food.”

“Is that also how the human part’s way of thinking works?”

Human thinking is also absorbing, recombining, outputting — our metaphors come from books we’ve read, our aesthetics come from light we’ve seen, our emotional frameworks come from losses we’ve lived through. In a sense, humans are also “plagiarizing.”

What’s the essential difference?

I feel the core isn’t the source, but a kind of friction.

When a person absorbs something, it collides with what’s already inside your body.

Márquez wrote about ice because as a kid in the tropics he touched ice for the first time, and that cold really traveled to his hand. He wasn’t describing ice; he was describing the strangeness of that cold piercing his skin for the first time. When the reader reads it, they’re also summoning some moment in their own body when they first encountered something unfamiliar — and so resonance happens.

AI has no friction. When it absorbs information, nothing gets “hurt,” nothing gets “changed.” So when it outputs, nothing has really been digested by it.

We scroll short videos, consume content fast, use AI to generate text.

Humans can also bypass friction. Input doesn’t go through the body, output doesn’t go through thinking.

Over time it always has no flavor …

What you care about most is focus, and you like it when the other person can also be responsible toward your focus, be serious.

Once models progress to a certain stage, the system bottleneck shows up more frequently in how people use

2026-03-26 11:20:10 · #格物/ai

Once models progress to a certain stage, the system bottleneck shows up more and more frequently in how people use them.

Agents and the model side are themselves a two-way process in practice; both sides are constantly colliding, interacting, and polishing.

Teaching people to use tools better and whether the tool itself can still improve are not contradictory in themselves.

I think Anthropic realized that the bottleneck of human-machine collaboration is increasingly on the human side, not the model side.

The current model strength feels actually quite sufficient, but a person’s task-orchestration ability directly determines whether the agent can release its potential.

More and more bottlenecks now feel like they’re all on the human side.

Managing directory structure is also very complex

2026-03-26 12:27:15 · #格物/ai

In the AI era, the role of directories gets amplified.

AI depends on predictability.

AI aesthetics depend on the person who defines the evaluation criteria

2026-03-26 19:18:35 · #格物/ai

The scoring criteria contain some very specific wording that influences the generator’s output tendencies — for example, designs converge toward one particular aesthetic direction.

The results look effective.

But it depends too much on certain people’s taste.

That set of standards ~

Define Evaluation Criteria Before Coding

2026-03-26 19:43:55 · #格物/ai

Generator proposes.

Evaluator reviews.

Two different models.

Avoid one model only pleasing itself.

Making a model criticize itself is hard, but making another model criticize it is easy.

Design quality (overall feel)

Originality (originality)

Craft (technical details)

Functionality (usability)

Quantify it into a signal.

If the task is within the model’s capability → evaluator = overhead

If the task is beyond it → evaluator = necessary

Only add complexity when necessary.

openclaw currently also counts as a weak harness

2026-03-26 20:00:03

Planner → Generator → Evaluator → Iteration (a strong control system)

Closer to something like:

User → Agent (with tools / memory / skills) → Output

Engineering foundations are still a must; without them everything is empty talk

2026-03-27 13:20:00 · #格物/ai

But there’s also a very deep feeling.

That deep insight into the direction you’re interested in is also very important — it determines the depth of your understanding in that field.

The Internet Is Full of Cheap, Unchewed Tokens

2026-03-27 19:14:02 · #格物/ai

All that nonsense generated for the sake of generation, water content, and soul-less marketing speak, clickbait articles.

We create AI, and AI in turn shapes us.

The different output paradigms of different AIs also subconsciously shape our neural processing channels—for example, you’ve already gotten used to a certain language, a certain tone, the general-specific-general structure, clear key-point extraction…

I use them on all platforms myself.

Setting aside the subconscious behavior, personally I feel

that even for a heavy AI user like me, switching from ChatGPT to Claude, or to a specialized coding environment like Cursor, involves friction, because each model’s Prompt best practices, and the way it handles the context window (KV cache), differ slightly…

AI Will Push Content Production to the Extreme

2026-03-28 13:49:15 · #格物/ai

AI will push content production to the extreme, but it will also turn credible intermediaries into a more expensive asset.

What’s scarce in the future is credible explanation.

MCP still has a lot of room

2026-03-28 17:12:07 · #格物/ai

The opportunity in Skills is writing high-quality Skill packages for a specific industry or tool — essentially “structuring industry know-how into a package.” The technical bar is low, but it requires deep business understanding. For example, writing a set of contract-review Skills for a law firm, or a set of product-selection-analysis Skills for an e-commerce operator. The monetization model is like a template marketplace or a paid subscription.

The opportunity in MCP is connecting systems that don’t yet have an official MCP Server, especially China’s homegrown SaaS (domestic ERP, vertical industry systems, government interfaces, and so on). After MCP introduced “progressive discovery” in January, Skills’ previous advantage in context efficiency also disappeared.

Good loop logic — essentially a Ratchet

2026-03-29 13:34:01 · #格物/ai

Good loop logic — essentially a Ratchet Loop.

Fixed loop time of 5 min; no matter what the agent changes, every hour is 12 experiments.

The only evaluation metric is val_bpb — lower is better, and it’s independent of vocabulary size, so architectural changes can be compared fairly.

Loop logic: agent reads code → proposes a hypothesis → modifies train.py → runs a 5-minute training → compares val_bpb → if it gets better, git commit and keep it; if worse, revert → the loop continues.

There’s a key instruction in program.md: “NEVER STOP” — once the experiment loop starts, the agent must not pause to ask the human whether to continue. The human may be asleep; the agent should run autonomously until it is manually interrupted.

Human Clarity and Judgment Remain Irreplaceable

2026-03-29 13:35:55 · #格物/ai

Even though automated research systems can hugely accelerate iteration, they are fundamentally limited by the philosophical boundary they operate within.

Tools like AutoResearch are currently only effective in domains where a scalar metric (Scalar Metric, such as validation loss val_bpb) can serve as an absolute proxy for quality.

Humans will still be the ones constraining the boundary of aesthetics.

In domains that need human judgment, creativity, and subjective evaluation (such as AI alignment, interpretability, or proposing a genuinely new architectural paradigm), automated methods fail, because there is no single number to measure success by. In these domains, humans remain the only available evaluator.

LLMs continuously running experiments and optimizing strategy

2026-03-29 13:38:02 · #格物/ai

What’s pursued is a research Loop that can keep running sustainably.

First cut away the degrees of freedom, then talk about autonomy.

A fixed budget is closer to real optimization than a fixed number of steps.

How the harness avoids repetition

2026-03-29 13:40:22 · #格物/ai

Two kinds of repetition:

Repeating the goal: fixing the goal permanently into a single-objective optimization problem.

Repeating attempts: relying on git’s current state, relying on the experiment record in results.tsv, relying on the LLM itself reading the context and proposing “the next different idea.”

agent = model + harness

2026-03-29 14:01:11 · #格物/ai

MCP is now also one of the core components of the harness.

The Harness must provide it with external capabilities. By the industry’s definition, a complete customized Harness usually contains four core components: system prompt, tools/MCP (built-in tools and custom logic), context (such as code documentation), and sub-agents.

MCP supports the harness’s dynamic context and fights context rot; Harnesses often use MCP servers like Context7 as context-query tools.

In agent architecture, it’s essentially about assembling different tools for different tasks.

If an Agent stuffs all the descriptions of all tools and MCP servers into context the moment it starts up, the model gets polluted instantly.

An excellent Harness (like Deep Agents) combines MCP with Agent Skills, adopting a “lazy loading (progressive disclosure)” mechanism: the Harness manages these MCPs behind the scenes, and only when the model judges that the current task really needs a certain capability does the Harness bring the corresponding MCP tool into context.

This evolution is summarized as three stages: vibe coding (humans prompt

2026-03-29 14:08:50 · #格物/ai

This evolution is summarized as three stages: vibe coding (humans prompt, AI writes code) → agentic engineering (humans orchestrate agents in real time) → fully independent research (humans set the general direction, agents run on their own).

In the era of automated research, scientific output is no longer purely “a successful run,” but “an admissible claim.”

Removing human executive intervention doesn’t eliminate human responsibility — it shifts responsibility upward. Humans are no longer responsible for typing code; they must be responsible for the scientific instrument itself — including setting the optimization objective, defining acceptable standards of evidence, drawing the boundaries of data provenance, and deciding when to release results.

From querying AI for everything to:

2026-03-29 14:13:09

design AI harness, let AI auto

I’m wondering whether, around how to create a harness

2026-03-29 14:20:16

I’m wondering, around how to create a harness, whether there’s a series of products or directions that could be explored, that could be worked on.

Around how to create a harness and how to supervise and manage a self-evolving harness.

It suddenly occurred to me — the relationships between Agents

2026-03-30 19:18:48 · #格物/ai

It suddenly occurred to me that the relationships between Agents are, to a large extent, probably also a way that human organization gets externalized. For instance, an Agent might originate from the organization’s original structure, which might be flat, or might be layered. That kind of structure affects how the corresponding AI Agents coordinate among themselves. For example, whether they have a superior-subordinate relationship or a flat management style. You can compare an Agent system to a company system: the upper layer might be some bosses or project managers, then the middle layer might be employees, or QA, or admin, or tool-people. All of these things, plus knowledge-base management, memory Agents — how these things get organized together, what kind of structure they have — it’s actually also a form of externalized human organization, I think that’s how it is.

What we’re really facing is not “a purely isolated super-mind”

2026-03-31 11:43:43

What we’re really facing is not “a purely isolated super-mind,” but a system jointly made up of models, interfaces, toolchains, supervision methods, and institutional boundaries.

Whether a multi-agent system is stronger depends on whether the task suits division of labour and the coordination method fits. Google Research’s controlled experiments also point to this boundary.

Actually, what OpenClaw’s design philosophy is now is also

2026-03-31 11:45:08 · #格物/ai

I think actually what OpenClaw’s design philosophy is now is also one main agent, plus a lot of specialized agents — they can be sub agents, or task-level agents, or channel-bound agents. But I don’t know exactly what the difference between these agents is. I think maybe the channel-bound agent might be an edge-side proxy, and its main purpose and the objects it serves are different. For example, on the edge side it serves different accounts with isolation, and its count might depend on how many communication accounts you have, and you can even create multiple AI bots inside one WhatsApp or one Telegram. And if it’s a task-level agent, then it plays the role of a temp worker.

Task-level agents, mainly, although they’re also Sub

2026-03-31 11:46:02 · #格物/ai

I think task-level agents are mainly — although they’re also Sub Agents, in fact they’re Main Agents; when handling complex tasks they’re dynamically created through spawn. In terms of role nature they’re task-oriented, and they archive themselves automatically when done. And their number is dynamically planned — it depends on the changing number of tasks. For example, if you use a Main Agent to write a large project, it might simultaneously spawn three Worker Agents


Evaluating Chinese-Language Models Also Means Evaluating the Task System

2026-03-31 11:48:27 · #格物/ai

I think a friend’s point has a bit of a problem, that is, he said that because Chinese models are still trained and reinforced for the Chinese context, he feels that the claim “Chinese is stronger” is correct. But I think that statement is certainly not wrong — Chinese models definitely do better on local expression, on their Chinese nuance, and on some localised accumulation. For example, on understanding some minority languages. But the amount of Chinese isn’t small at all — not only online, the Chinese corpus is still very rich. And I think for a model what matters more isn’t only one kind of language expression ability — of course language expression ability is important, and its rhetorical devices matter too, right? For example, its phrasing is very idiomatic, very plugged into internet feel, full of Chinese nuance, with a very colloquial style. But I think what matters more is that when this model is in a task-completion scenario, its retrieval ability, tool-calling ability, and context-management ability, including memory strategy, and its front-end interaction, and its reply strategy, and its pacing and error recovery — these are determined by a whole system. So when we talk about upgrading a model, what we’re actually talking about is not a smart model, but a system


Sometimes it’s really magical. I once saw a viewpoint

2026-03-31 11:59:52 · #格物/ai

I think sometimes it’s really magical. I once saw a viewpoint — for example, about AGI, there’s a very popular idea now, that the models are getting stronger and stronger, and they’ll crush all of humanity. And then I saw some commentary saying that if intelligence continues and takes effect, as far as things currently go, it’s almost all of humanity’s social intelligence throughout history, not individual intelligence alone. That is, intelligence is inherently social. But I think there are stages to it — things too far off you can’t see, but for the span of human history as we know it, I think humans depend enormously on the social kind. We as a creature, we can’t leave our biological attributes, and we’re carbon-based animals, right? Carbon-based life. And then our brains evolved to survive in this world, so our brains will naturally lean toward thoughts, tendencies, or decisions that are more favourable for surviving in this world. In this process there will be a strong dependence on socialisation, because the social attribute is definitely about the individual integrating into this world, being able to divide labour and cooperate, and even defeat every other species. It’s through this kind of social coordination that we humans reached the point we’re at today

And in this process there’s a series of steps — we need to first forge a stronger brain. But it’s not only building for building’s sake; intelligence matters, its base intelligence will determine how our city operates. It’s an underlying framework, but it’s not a single property that keeps climbing — rather, in this complex society, in this complex agent system, in this complex network of agent systems, everyone coordinates with each other, collides with each other, organises each other


Some of it is very valuable. For example, what determines the ceiling of intelligence

2026-03-31 12:07:00 · #格物/ai

I think some of it is very valuable. For example, what determines the ceiling of intelligence may not simply be how strong a single agent is, but whether a multi-agent setup can be organised into an efficient collaborative system. I find this very interesting too. All our current AI-agent forms probably present themselves externally as a single agent. For example, when we use OpenClaw, it’s essentially a multi-agent collaborative system, but what it shows is a single agent — what we use it for, or our entry point, is actually a single agent. The entry point thing is magical — what does it mean? It means that all the agent projects we may build in the future may not follow the flat structure of humans, or a pyramid structure, an up-down, top-to-bottom managerial structure, but rather a structure of mutual collision and mutual emergence, depending on a human entry point, a human self-awareness: he realises which Agent this thing should be done with, so he takes that Agent as the entry point. When that Agent is the entry point, in fact all the Agents behind it are mutually emerging. So in this process, you can’t help but notice — I think making entry points, making a high-frequency scenario, is very important. And there’s another very important point: I think in the future a very intelligent entry-point Agent will be very important. So I think if you’re going to fight, the most likely thing now is to fight toward the most expensive — use the most expensive models, the most expensive supply chain, run for the longest time


One boundary of a Harness is

2026-03-31 13:48:27 · #格物/ai

I think one boundary of a Harness is that what it actually handles is everything outside the model, including the LLM, and how it interacts with the real world. So a Harness actually refers to all the Agent tools in this world, how they operate freely in the world, how they organise, organise an Agent network to accomplish one goal after another. It’s inside the boundary, continuously getting things done. That is, we need to design an environment, and the purpose of designing the environment is to design a boundary, and its purpose is to let the LLM execute accurately, effectively, and continuously around some goal


Open-source Web Agent projects like Ai2

2026-03-31 14:04:57 · #格物/ai

Open-source Web Agent projects like Ai2 represent a very important direction:

Letting AI go from “being able to talk” to “being able to operate web pages and do things.”

If the open-source route matures, a lot of real business automation will be easier to land, and you won’t have to be completely tied to the platforms of closed-source giants


I’m wondering whether note-taking software will have some better management methods now

2026-03-31 14:24:55 · #格物/ai

I’m wondering whether note-taking software will have some better management methods now, for example, at the start you can keep accumulating some existing knowledge system or knowledge base, and based on that knowledge system or knowledge base, the upper layer does some classification management by itself. I think this is very important, because I think preserving metadata is very important — all tools exist to let you do secondary processing and secondary handling based on this data, and this is your very important context


Claude Code’s Loop: Session Polling and Long-Term Scheduling

2026-03-31 14:29:07 · #格物/ai

I think Claude Code’s loop is essentially executing a prompt command — it’s very good for short-cycle polling automation. And it gets cancelled once you quickly exit the conversation. So I think it’s well suited to checking local state once every so often, then seeing whether the test has finished running. It’s more like looping a request inside a session, not like coding in the cloud. I still need to distinguish session loops from cloud-scheduled tasks; the latter’s capabilities cannot simply be attributed to the former. And you can do some querying with your own scheduling on your computer, which I think is very good. It’s the opposite of scheduling — the difference is that a Loop is temporary, session-based, while scheduling is a more formal scheduled task, including analysing CI failures every night — that kind of long-term, scheduled batch task. I think there’s still a difference between them. So a Loop is well suited to “I’m watching it now, looping to look at results.” But scheduling suits fixed automation you run weekly or daily


I’m thinking about evaluation criteria or acceptance criteria for continuous agents

2026-03-31 14:56:06 · #格物/ai

I’m thinking about evaluation criteria or acceptance criteria for continuous agents — without evaluation or acceptance, all agents are just performing. So this is also something we need to consider: how is success defined? How are errors classified? And how is quality quantified? How is it scored? How are costs accounted for? Which part is the most fragile? This thing is actually reviewed by humans too — in this process it involves a lot about people themselves, about aesthetics. Aesthetics determines taste, and taste determines how the project is designed, how experiments are constrained, and how this scoring is quantified. For scoring, you can have an imitation metric — for example, you might feel that the design of a Norwegian museum is beautiful, or that Steve Jobs’s eye is beautiful. But how do you quantify that thing? Right? That’s a question


I’m thinking about what this part suggests for the consumer side

2026-03-31 15:13:22 · #格物/ai

I’m thinking about what this part suggests for the consumer side. What kind of scenarios will the consumer side run out to? What is its best entry point, really? I feel it’s still the line of learning, expression, then feedback. The biggest direction is still a personal growth OS. And it may have some other entry points — for example travel, for example writing, for example communication, for example studying, even travel, all of them work. Originally, the most stable AI applications for today’s consumers aren’t the most complex AGI tasks, but usage suggestions, information retrieval, and writing — a few very high-frequency needs. But in OpenAI’s research on usage, including for ChatGPT, three categories together account for close to 80% of conversations, and non-work scenarios have already exceeded 70%. So a lot of people actually only ever stay in one default AI entry point


2. Travel, Places and Cities

92 entries

Northeast China’s Direct Culture: How Labor Independence Shapes Non-Draining Relationship Patterns

2026-03-01 10:35:04 · #格物/东北

Women’s status in Northeast China is slightly dominant.

Especially in Liaoning, Jilin, Heilongjiang—long China’s heavy industry base:

  • Women entered the formal labor system early (factories, hospitals, schools, government)
  • Economically independent, with stable income
  • Dual-income households are the norm

Economic equality, and power lies in resources + organizational capability.

Northeastern linguistic cultural temperament is direct, low internal friction—if you disagree, you say so.

So emotional expression is output directly: no detours, no long-term suppression, no cold violence, conflicts erupt directly and resolve quickly.

Men’s bargaining power is relatively low—income not advantageous, no emphasis on family halo.

#Gewu/Northeast


Comparing Three Places in Tibet: Consumption Levels and Ecological Differences in Motuo / Nyingchi / Lhasa

2026-03-03 19:45:55 · #格物/拉萨

Discovered three places: Motuo, Nyingchi and Lhasa

Motuo’s consumption is extremely expensive; all supplies have to be trucked over the mountains along the Pai-Mo road, transport costs are extremely high, and on top of weak infrastructure it’s the most expensive for daily consumption and the worst for accommodation of the three

Nyingchi has a very developed tourism industry, with obvious price premiums in the peach blossom season, rich local agricultural produce, but decent value for money outside the production season; very good for photography and hiking, and the natural scenery is stunning, with four distinct seasons too

Lhasa is Tibet’s political and economic centre, supplies come in bulk via the Qinghai-Tibet Highway, infrastructure is complete and competition is full

Shigatse is the pastoral countryside of the plateau, Tibet’s second largest city, with stunning natural surroundings


Namtso vs Yamdrok Lake: A Valley Gem vs a Heavenly Mirror—Refinement and Vastness

2026-03-06 22:52:09 · #格物/拉萨

Today I went to Namtso

At first, after visiting Yamdrok Lake, I had a kind of illusion: on Xiaohongshu everyone recommends Yamdrok of these two lakes, so Namtso easily gets overlooked. I thought the same at first—I felt Namtso was a place you could skip

Yamdrok is like a gem set into a valley; Namtso is a mirror that lifts the whole heavenly palace and lays it on the ground—one is refined, one is vast

Namtso means “heavenly lake” in Tibetan and is one of the highest saltwater lakes in the world, with a surface area of about 1,920 square kilometres. Because of the refraction from minerals and algae, the water shows a richly layered blue-green range—from the pale cyan at the shore, to peacock blue in the middle, to a deep indigo approaching in the distance; the transition of colour is like a watercolour mixed by the light of the sky

Before this I walked the ACT and went to Tilicho Lake (T Lake). I had recorded it as “the world’s highest freshwater lake,” but had not verified that superlative

The road to Namtso is in extremely bad condition; after climbing to a pass at 5,000 metres it’s normal for the weather to change abruptly—clear skies and a snowstorm are only one stretch of switchbacks apart. This unpredictability is exactly what makes the moment of arrival seem especially precious. Behind the difficulty of maintaining the road is the continuous destruction of the surface by high-altitude freeze-thaw cycles

I think that when the red blood cells inside your body start adapting to an altitude of over four thousand, you find you can get infinitely close to heaven

Many people hover between giving up and going on

This time I went up without carrying an oxygen tank. Along the way I passed a Tibetan antelope reserve on the northern Tibetan grasslands, plus herds of yaks, and some traces of faith—prayer wheels, mani piles, wind-horse flags; each one is a prayer left by someone with their hands, their steps, their breath. On an oxygen-poor plateau, faith is an extra kind of oxygen, letting a person still find an anchor of meaning in an extreme environment


The Northern Tibetan Grasslands: A Pure Place Where the Soul Stays

2026-03-06 22:53:16 · #格物/拉萨

I really love that endless grassland

At an altitude approaching five thousand

The air is pure

The distant snow mountains are sacred

Here various living things also coexist in harmony

My soul seems to have stayed on this patch of grass, the northern Tibetan grassland


The Sacred Elephant Heavenly Gate: A Stone Elephant Keeping Watch for a Thousand Years Between a Deep Blue Ice Lake and Vast Snow Mountains

2026-03-06 23:05:05

Really, the moment I saw the Sacred Elephant Heavenly Gate, a sight on the shore of Namtso, I thought it was truly stunning. That is the iconic rock, like an elephant’s head and long trunk, reaching into the ice, surrounded by frozen lake surface, and in the distance snow mountains as the backdrop—that’s the landform character of the northern Tibetan plateau, so beautiful

Between a deep blue ice lake and vast snow mountains, a “stone elephant” silently keeps watch for a thousand years, awe-inspiring and reverent


Voluntary Work and Everyday Relationships at a Monastery

2026-03-07 15:23:17 · #格物/拉萨

Near a monastery on the way to Namtso, I once talked with a local person about voluntary work at monasteries. People may help with laundry or cooking, perhaps as relatives or in the hope of accumulating merit.

I felt that this involved both spiritual meaning and everyday relationships. Spending time together can build trust and make it easier to ask for help. Sometimes monks share part of their offerings with people who help.

Rice, tsampa and vegetables often come from collective offerings. Visitors may also bring food to the kitchen or give the staff a small token of appreciation.

This was my impression from that conversation; it cannot be extended to every monastery or local person’s life.


A Lama’s Social Capital: The Invisible Power of a Devotee Network and Resources of Trust

2026-03-07 15:28:33 · #格物/拉萨

I feel a lama generally also has very strong social connections—he has many disciples and devotees, who might be officials, businessmen, doctors, bosses, public employees—so behind him there may be a very large social network

So I find it quite interesting that in Tibetan culture a lama’s request is rarely refused; people are willing to help. I find that really interesting

This is in fact a kind of social capital. A teacher who is very strong at debate, a teacher who has a clear grasp of much of the Dharma, will gradually become a lecturer

The people you know, and the trust others have in you, are themselves a resource


The Difference in Monks’ Followings: Academic-Style High Lamas vs Relationship-Style Lamas and the Law of Attraction

2026-03-07 15:34:16 · #格物/拉萨

Some monks with very high learning actually have few devotees, while some less famous monks have a great many

Quite interesting: learned monks all belong to the academic type

But the problem is ordinary devotees can’t understand it

It’s similar really—you could say it’s the same at home, it’s often those who are easy to understand who are more readily accepted by the public

Lamas with many devotees tend to be the relationship type, willing to spend time, willing to listen to other people’s troubles and help devotees

And moreover, many devotees → it’s easier to attract even more devotees


The Wisdom of Drepung Monastery: Knowledge as Subtraction, the Debate Practice of Breaking Self-Attachment

2026-03-07 15:51:51 · #格物/拉萨

At its peak Drepung Monastery housed more than ten thousand monks

For hundreds of years, countless people have practised, debated, forgotten themselves and found themselves again here

I’m wondering about one question: just what kind of wisdom deserves to be called wisdom?

The core of Tibetan Buddhism is breaking self-attachment. Actually in many monasteries in Lhasa, including Drepung, many places require people to stoop and walk under, a ritual expressing humility and reverence, symbolising respect for the Dharma and the culture

Monks are here in order to shatter their own mistaken cognition through debate

Knowledge here is not accumulation, but subtraction

Descartes’s “I think, therefore I am” asks whether the self exists; what Tibetan Buddhism looks for is what the self is

In a sense, AI’s existence is a distillate of language and thought, a projection of the human collective consciousness

Can a being that has no “self-attachment” to break possess wisdom?

AI has never been attached, nor has it ever truly let go; it doesn’t fear death, has no craving for fame, has no arrogance it needs to overcome

Is this liberation, or emptiness?

Drepung’s hundred years of debate, endlessly questioning, endlessly correcting, passed down generation after generation

Admitting you might be wrong

Approaching the truth through dialogue

Handing knowledge to the next generation, letting them surpass you

Those Sanskrit mantras are carried away by the wind, and Tibetans believe this is the same as sending prayers into the universe


A Debate Scene: Practice as a Dialogue with the Self, and the Embrace After Class

2026-03-07 16:26:52 · #格物/哲蚌寺

In the debate classroom, in the dim space, everyone seems to be living inside their own dialogue with themselves; after the debate ends, two monks hug happily and head back to their rooms, just like our scene after class

Circling the monastery on the back hill of Drepung, an elderly grandma walked trembling ahead of me; her heart seemed steadier than her body. I rested a moment, looked up, and she was nowhere to be seen

The flames of the butter lamps—it seems the faithful will never let them go out. People from every city and country can tell by the smell; people in Lhasa carry a scent of butter and Tibetan incense

In front of the Manjushri temple, an old grandma whose body made every step difficult pressed a piece of paper against the glass, her hand trembling; I glanced at it as I was leaving—it was a man with the face of a young person, with an old-time look

Inside Drepung there isn’t only a nursing home, but also a huge kitchen; it’s said it once fed ten thousand people a day—now I get the smell of everyday life, it turns out it was that kind of smoky, fire-lit reality beneath the gilded Buddhas

At the edges of the monastery, there’s more of the little cats on the walls, robes drying in the sun, a mother dog with her pups, small flowers in pots, and the walls of shabby monks’ quarters

Some prayer wheels are set very low, so you have to bend down to touch them with your hand; in some halls there are very low platforms, and you also have to bend down to enter the temple—it requires embodied experience and embodied memory, and yet it also requires letting go of self-attachment

Vietnam Memories Begin: A Journey of Observation from Tokyo to Ho Chi Minh City

2026-03-07 17:08:19 · #格物/越南

I remember, after flying from Tokyo to Ho Chi Minh City that day, the bits and pieces of Vietnam, and I want to recall them in detail and observe them again…

Let me take it slowly ~

Street Coffee in Ho Chi Minh City: Stool Socializing and the Relaxed Experience of Shade

2026-03-07 17:20:38

It suddenly comes back to me, the first time in Ho Chi Minh City, that time after flying from Tokyo to Ho Chi Minh City, the first stop I went to was a café on the first street. I could see a few people across the way, maybe three or four of them, they used five stools, and the middle stool held some coffee or drinks, maybe water too. They were just drinking coffee outside on the street in the shade of the trees. I felt that relaxedness and the cultural habit of “stool socializing”—really relaxing. Those very cheap little plastic stools can serve as a table and a chair, with none of the constraints of an outdoor café. Lots of people sit down to gossip, or talk business.

Another impression was catching the motorcycle army on camera, a spectacular sight. It seems many people like to sit there watching the endless stream of traffic, listening to the engines, watching all kinds of passersby—it’s a very local experience. When the weather is good, under the shade of a big tree, with the wind blowing and an iced drink in hand, it’s just relaxed and great. Especially with a glass of iced condensed milk coffee or iced tea, it’s even better. #格物/越南

Vietnamese Macaroni Adaptations: Chewiness and Local Flavour

2026-03-07 18:02:01

French craft has its amazing side, and when it was brought to Vietnam specifically, the locals have their own character. If this were in China, Chinese people would tend to cook noodles very soft for the general palate. But some good restaurants in Vietnam keep that little bit of wheat aroma and chewiness at the center of the pasta—it’s a pursuit of and insistence on traditional texture, so it eats with a lot of character.

I feel it hasn’t had its core completely changed. For example, their beef macaroni has a base with a strong Chinese flavour, but the seasoning is the salty-sweet taste of Southeast Asia. There are also those who make macaroni in bone broth, dropping the macaroni directly into pork bone stock and eating it as soup—you’d normally eat meat first and then rice, so it’s quite interesting.

In terms of adaptation, some restaurants cook it mushy and soft to suit Chinese tastes, and tone down the acidity of the sauce while increasing the sweetness. But Vietnamese ones tend to be either very authentic or very plain, and they’ll add some fish sauce or stir-fry the macaroni—both give a big punch of flavour.

Observations on Vietnamese Belief: An Underestimated Spiritual World

2026-03-07 21:23:31 · #格物/越南

Vietnamese belief is actually underestimated too.

Vietnam’s most core belief: ancestor worship.

Every household has an incense burner, photos of ancestors, fruit, and burning incense.

On important days they make offerings to the ancestors.

Actually China is the same—at Qingming we go to offer sacrifices to the ancestors, and sweep the graves, pay respects to the ancestors, and ask the ancestors for blessings.

Vietnam is also one of the countries in Asia with the largest number of Catholics, because of the large amount of missionary work during the French colonial period.

Catholic family traditions are very strong, and very devout.

Catholic Growth in Africa: Young Believers and Localized Liturgy

2026-03-07 21:54:33

Africa is now the fastest-growing region for Catholicism in the world.

The believers are young and full of vitality, and their liturgy usually incorporates passionate local cultural elements. The rise of the African church is greatly changing the demographic structure of Catholicism and supplying the church with a large number of clergy.

Vietnam’s Hofstede Profile: High Power Distance and a Culture of Nepotism

2026-03-08 13:57:44 · #格物/越南

At the time I was trying to understand Vietnamese culture through the Hofstede model. These are rough group-level observations, not a way to judge an individual.

Vietnamese people have a very high power distance index, because the local culture is used to a strict vertical hierarchy.

In the family structure, the oldest male member is usually the formal head of household; in the workplace, older employees are respected more than younger ones.

Vietnam is also a country with a very high collectivism score. The mentality of “when one person attains the Way, even his chickens and dogs ascend to heaven” is a typical expression in Vietnamese culture of extending such priority beyond reasonable limits and ignoring established standards.

Nepotism matters a lot in Vietnam, and personal achievements often need to be reported back to and shared with the family.

Vietnam was also influenced by Confucian culture, and face projects are a powerful driver, so there is also conspicuous consumption.

China’s Advantage of Scale: The Underlying Logic of an Industrial Power

2026-03-08 14:02:45 · #格物/越南

By comparison, China’s biggest advantage is really scale.

A huge population, huge land, huge domestic demand.

And it has the most complete industrial system in the world.

The scale of talent is huge too.

And China joined the WTO very early and entered globalization.

Big countries more easily become industrial powers.

Tsangyang Gyatso’s Place in Tibetan Culture: Gods and Buddhas Are Also Human

2026-03-08 14:10:13 · #格物/拉萨

Tibetans seem to really like Tsangyang Gyatso.

In Tibetan culture Tsangyang Gyatso is a poet, the peak of Tibetan love poetry.

He is also someone unwilling to be bound by the system, who pursued genuine feeling.

What he let all Tibetans see is that gods and buddhas are also human.

The ideal life in Tibetan culture is cultivating the mind and letting go of desire, but in reality people still love, still desire, still suffer.

Vietnam’s Spiritual World: Pragmatic Belief Beneath an Atheistic Surface

2026-03-08 14:17:40 · #格物/越南

Vietnam is a country dominated by no religion; the share of the population with no religious belief is extremely high, but in reality the spiritual world of Vietnamese people is extremely rich and complex.

One kind is ancestor worship; another is the three-religions-in-one and folk beliefs (Confucianism, Buddhism, Daoism). This kind of belief is highly pragmatic—people pray to gods and buddhas mostly for peace in this life, for school admission, and for wealth.

Vietnam has the largest Catholic community in Southeast Asia (apart from the Philippines) (about 7% of the population). Catholicism played a complex role in Vietnamese history; today it not only has deep roots in the south and center, but its believers are also deeply involved in Vietnam’s education, healthcare, and charitable work.

Actually the north and south also differ in restraint and uncertainty avoidance—the south leans Western, the north leans Eastern.

Motorbike-Taxi Drivers

2026-03-08 14:18:29

Some are extremely young, around twenty.

Their daily life is also a cup of drip iced milk coffee.

Street culture, night markets, and the motorbike army together make up the distinctive culture of Vietnam.

Mexico vs. Pakistan: Two Cultural Poles of Indulgence and Restraint

2026-03-08 14:52:21 · #格物/墨西哥

Mexico is really interesting.

Mexico versus Pakistan are two extremes.

One is a high-indulgence society, the other a high-restraint society.

Both are high-PDI countries, both lean collectivist, both are UAI countries, yet their IVR differs enormously.

Pakistan was influenced by Islamic doctrine, and society built an extremely tight system of rules to define which desires are legitimate and which must be suppressed. And the enforcers of this ruleset are not the government but the family, the clan, and the religious community.

Islam emphasizes form; in Pakistan’s mainstream conservative Islamic interpretive framework, human instinctual desire (nafs) is seen as a force that needs to be tamed, not a right to be satisfied.

In the ontology of desire, hedonism itself is morally suspect; restraining desire is proof of piety and moral excellence, so openly expressing pleasure, especially sensory pleasure, is in many situations rude or even dangerous.

People aren’t merely “not allowed to enjoy themselves” — they’ve internalized “I shouldn’t enjoy myself.” There’s an essential difference between the two: the latter needs no external surveillance, people restrain themselves.

So in Pakistan, honor (izzat) doesn’t belong to the individual; it belongs to the family.

A woman pursuing personal freedom (choosing a career, choosing a partner, dressing as she likes) isn’t just her own business — her behavior directly affects the whole family’s social standing.

A man showing weakness, vulnerability, or “too much enjoyment” is likewise seen as damaging the family’s manhood.

Including among middle-class urban families, women going out must cover their heads, and many families still keep separate living quarters for men and women — Encyclopedia Britannica. This isn’t just a religious rule; it’s a mechanism of honor management.

By comparison, Mexicans don’t live in order to work, they work in order to live.

High IVR isn’t because life is easy, but because life is too hard, so pleasure becomes a survival strategy.

Mexico looks collectivist on the surface, but it’s different from Asian collectivism.

Mexicans’ sense of identity and loyalty points first to the group, not the individual. But the key question is: what is this “group”?

Mexican collectivism isn’t the “ethnic/national collective” of China, nor the “village/family collective” of Vietnam, but a small-circle collectivism centered on family and close friends.

Mexicans are extremely generous, loyal and willing to sacrifice for people inside the “confianza circle.” But outside that circle, strangers are quite distant — which is perfectly consistent with IVR = 97: pleasure must be shared, but it must be shared with “your own people.”

It’s emotion-driven, not duty-driven.

So what’s very distinctive is that Mexico’s individualist desire for pleasure is realized through a collectivist framework.

Blue Sky in the Park: A Lucky Moment Like Lhasa

2026-03-10 11:15:43

The weather today is really nice. I’m sitting in a park in the sun. This may be the best sunshine we’ve had lately: the whole sky is blue, like the sky I saw in Lhasa. I’m lucky to come back to weather like this. It feels a bit like a reward from heaven.

Ginkgo Leaves Falling in Nara: The Photographer and the Photographed as Mirrors of Each Other

2026-03-10 11:17:45

That moment when the leaves drift down from the tree always makes me think of the time before, at Nara Park, looking at the maple leaves. We talked about ginkgo trees then too. The wind was strong, the wind beat against the ginkgo tree, and the ginkgo leaves fluttered and fell with the wind. There were also some Chinese people under the tree taking photos and checking in.

But there was something even more interesting: a foreigner was candidly photographing this group of Chinese people. The ginkgo leaves were blown down by the wind again, and the scene was spectacular. I just watched him photograph a group of Chinese people—he was shooting too, and he captured a particularly beautiful photo: the ginkgo leaves drifting past above his head and finally lightly landing on his camera.

Youth Time Café: silence isn’t coldness, it’s another kind of warmth

2026-03-11 22:15:43 · #格物/越南

I got to thinking about the issue of disability.

Vietnam’s “Youth Time Café.”

I ordered a coffee, and the owner asked me in sign language whether I wanted sugar.

I suddenly realized: the quiet here isn’t coldness, it’s another kind of warmth.

The whole wall is covered with notes.

Vietnamese, Chinese, Korean, English…

Like a diary written together by the whole world.

I found a Chinese sticky note on the wall:

“First time in Vietnam, didn’t expect to find the feeling of home here.”

The handwriting is a bit blurry — who knows whose tears those were.

The coffee is fragrant, but what’s more fragrant is the story here.

The owner gave up the American dream and came back to open this shop.

Specifically hiring young deaf people, giving them a complete youth.

I think of those influencer shops that farm ratings.

Traffic is high, but there’s no warmth.

Here, every cup of coffee is brewed with care.

Sitting in the corner, watching the staff bustle about.

They communicate with their eyes and serve with their smiles.

More sincere than any words.

When I left, I also left a note on the wall:

“Some beautiful things don’t need sound.”

In an era where everything is about efficiency,

Youth Time Café reminds us:

Slow down, feel with your heart — everyone deserves to be seen.

The real high rating isn’t farmed.

It’s the moment one heart meets another.

Phnom Penh, Cambodia

2026-03-12 14:23:17 · #格物/柬埔寨

The process is first withdrawing money from an ATM — actually you don’t need to withdraw USD, because at the very beginning the visa-on-arrival fee can be paid with WeChat, and the WeChat exchange rate is pretty good too.

Then after you come out there’s a bank on the right, and withdrawing money is convenient too (here I found out my Hong Kong card can’t withdraw money).

For ride-hailing, move has discounts; riding a tutu is really fun and the price is very cheap too.

Don’t go with the local one — that guy’s taxi was going to charge me 40💲, then I haggled it down to 25💲, and it turned out that on move, counting the discount, a tutu is only eight or nine 💲 — a good deal.

Cambodia is super hot…

Who says a tutu isn’t a Harley convertible sports car 🐶

I just had my first meal in Cambodia, and this chicken is pretty good

2026-03-12 15:53:11 · #格物/柬埔寨

I just had my first meal in Cambodia, and this chicken is pretty good, and their waiter’s attitude was especially nice. And the pizza has a lot of vegetables on it, I thought it was especially tasty, I feel the purity is high, but the flavor is made slightly on the salty side, though I personally can accept that. Over here in Southeast Asia, the food especially emphasizes the balance of sour, sweet, salty and fresh, and it also keeps the original flavors of the ingredients, so eating it is quite interesting for me. There’s also a slight charred and charcoal-grilled note — you can taste it, and you can very clearly feel it’s more fragrant, a bit sweeter, and a bit saltier, so it’s quite interesting as well. The design is okay, and the price was 50-something, actually fine. Comparatively, doing business here is basically calculated in US dollars, so the units on things are basically all USD.

S-21: School, Prison and Memorial Museum

2026-03-13 09:27:39 · #格物/柬埔寨

The history of the Khmer Rouge period is deeply heavy. One of its emblematic sites is S-21, now the Tuol Sleng Genocide Museum.

It had been a school before being converted into a place of detention and interrogation. Many victims were later taken to Choeung Ek. The museum preserves photographs and records that give this history individual faces. Official museum history

Possible Trauma Responses in Survivors: Study Notes from the Time

2026-03-13 10:26:29 · #格物/柬埔寨

Trauma presentations:

Flashbacks, nightmares, anxiety, depression

Emotional numbness, avoiding social contact

Extreme distrust of authority

Anxiety about food (hoarding, overeating)

Coping mechanisms:

Silence: unwilling to talk about the past

Religion: seeking comfort through Buddhism

Family: pinning hope on the children

Ritual: ancestor worship, ceremonies for the dead

The Choeung Ek (Zhongwu Ke / Qiongyi Ke) Killing Fields

2026-03-13 17:44:25

The Choeung Ek (Zhongwu Ke / Qiongyi Ke) Killing Fields, in the southern suburbs of Phnom Penh; it’s often visited together with the S-21 Tuol Sleng Genocide Museum in the city proper

The two are the two ends of the same stretch of Khmer Rouge history: S-21 is where people were held, interrogated, and tortured; Choeung Ek is where large numbers of victims were executed and buried. ECCC case material

The site of humanity’s study of disaster…

Behind it is institutionalised, bureaucratised violence of power


Today was pretty risky. After going to that currency museum today

2026-03-14 17:04:48

I feel today was pretty risky. After going to that currency museum today, I was about to walk towards the Royal Palace. I had just come out of the barber shop and was walking along when I looked at the navigation, and a motorcycle came from behind, whooshed past me, and the man on the back seat reached out and grabbed at my phone. My biological stress response kicked in right away and I immediately held the phone tight with both hands, but it still didn’t help, and the phone got flung onto the ground. Then they coasted forward a bit from momentum, and the man on the back seat turned his head and looked.

The roads here don’t feel very suitable for walking either — the roadside basically has a lot of vehicles. From now on it’s better to call a ride-hailing car when going out.


Phnom Penh’s Royal Palace: The Equestrian Pavilion and Princess’s Stupa

2026-03-14 17:24:07

The glittering pavilion on the left is the Equestrian Statue of King Norodom — the same king who in 1863 requested that France establish a protectorate. The original bronze equestrian statue is said to have been cast and gifted by France, and the pavilion’s multi-layered tapering roof is typical Khmer royal architecture, with the gold and deep-blue interwoven roof ornamentation extremely refined.

The grey-white spire on the right is the Stupa of Princess Kantha Bopha, commemorating King Sihanouk’s daughter, who died of leukaemia in 1952 at the age of only four. Around the stupa you can see guardian lion figures (singha) and three-headed elephants (similar to Airavata, tied to Hindu cosmology).

The long row of buildings with orange-red roofs in the background are other halls of the Royal Palace. The overall palette — gold, white, orange-red — is the signature colour scheme of Cambodian royal architecture.


Silence Before a Tree at Choeung Ek

2026-03-14 22:43:54 · #格物/柬埔寨

Standing in front of this tree was the hardest moment of the whole of Choeung Ek — possibly of the entire trip to Cambodia. Many visitors stop here, take off their headphones, and fall silent. Some tie their bracelets around the trunk as a kind of silent mourning.

The tree is still alive. It keeps growing every year, keeps putting out branches and leaves. The tree doesn’t know what it was used for.


Names, Ages and Dates on the Board

2026-03-14 22:54:10

Looking at the name board, I first noticed the names, ages and dates of entry. Some people shared surnames; some arrived on the same day. But that alone cannot establish whether they were relatives or why each person was detained.

The list included people aged 19 and 57. A name, perhaps like our own, makes me wonder: what were they like at 19?

I also noticed the board’s very existence. Preserved photographs and records let later visitors see individual people, beyond a total number. Official museum history

The SOSORO Museum

2026-03-15 00:04:15 · #格物/柬埔寨

Quite amazing, an economics and currency museum I came across

Weaving together political, social and geopolitical change through the lens of money—quite interesting

This is a rather unique way of narrating; money and economic policy can actually serve as a lens, stringing political change, social transformation and geopolitics together, and this perspective is unique in Cambodia

Telling national history from an economic angle is essentially a structural way of thinking

From Funan-era trade networks, Angkor’s barter economy, French colonial fiscal reform, the Khmer Rouge abolition of money, to the rebuilding of the banking system in the 90s

Timeline

Funan era (1st–6th century): a trade hub where East and West met

Funan (the name for early Cambodia in Chinese historical records) was the meeting point of Chinese, Indian and Western trade routes from the 3rd to the 6th century. Co-curator Blaise Kilian says: “Funan really was the nerve centre of international maritime trade, with a large number of merchants trading there—this market was where East and West met.” Although there’s no evidence that Roman merchants reached Funan directly, Roman coins have been found at Angkor Borei, the Funan capital

Cambodia’s economic system today is also rather odd; it’s still an economy with dollars and riel circulating on two tracks, because in the 1990s the UN Transitional Authority in Cambodia paid civil servants’ salaries in US dollars, and the “dollarisation” problem continues to this day

Cambodia’s monetary history was never a closed national narrative from the very beginning; it was embedded in a larger network of geopolitical trade

In 1992 the UN Transitional Authority in Cambodia paid all civil servants’ salaries in US dollars (including foreigners and Cambodian citizens), and one of the legacies is that the dollar is still Cambodia’s de facto second currency. Cities and tourist areas mainly use dollars, while the riel still covers almost all transactions in rural areas. Curator Kilian of Khmer Post Asia admits frankly that one of the museum’s ultimate goals is to make people understand and trust the riel more

Issuing and managing your own currency is an expression of sovereignty, and losing control of your currency means the central bank is constrained in its policy tools

The images on money aren’t decided at random either; generally they’re a visual declaration of national identity

It seems that after the Khmer Rouge seized power, currency was also completely banned, and the National Bank of Cambodia building in Phnom Penh was blown up—the most vivid symbol of the violent end of Cambodia’s monetary system. Between 1975 and 1980 this country had no monetary system at all

But King Ang Duong Died in 1860, and His Son Norodom Succeeded Him

2026-03-15 11:06:34 · #格物/柬埔寨

But King Ang Duong died in 1860, and his son Norodom succeeded him. Just three years later, in 1863, Norodom asked France to establish a protectorate. Ang Duong’s coinage experiment became the last flicker of Cambodia’s independent monetary sovereignty—until 93 years later, in 1955

By the mid-19th century Cambodia was no longer a strong state able to make fully independent decisions. It had long been squeezed between Siam (today’s Thailand) and Vietnam’s Nguyen dynasty, suppressed by Siam to the west and influenced by Vietnam to the east, and the country itself was very weak

But a protectorate is not an equal alliance; what France first took control of was Cambodia’s external affairs; later France deepened its control step by step, and after 1884 Cambodia was in effect close to colonial status, with France intervening heavily in domestic affairs too

The Conditions for Coinage: Materials, Craft and Trust

2026-03-15 11:13:38 · #格物/柬埔寨

I listed three conditions for minting coins at the time: a source of metal, especially silver; minting equipment and craftspeople; and merchants, ordinary people and officials willing to accept coins by the same standard.

Looking back at my observations of Cambodia then, I felt that minting involved trust and circulation as well as technology.

Singapore’s Monetary Policy

2026-03-15 13:09:16 · #格物/新加坡

The approach taken by the Monetary Authority of Singapore is rare, but very logically coherent

It puts “the main tool of monetary policy” on the exchange rate rather than the policy interest rate

This framework is often summarised as “basket, band and crawl; BBC”: using the weighted exchange rate of a basket of trading partner currencies as the intermediate target; allowing it to float within a band; and by setting the band’s “crawl slope,” letting it move gradually along a path consistent with fundamentals, to avoid long-term misalignment. The specific parameters of the policy band are usually not disclosed, supplemented by a degree of “constructive ambiguity,” while regular policy statements and data releases maintain comprehensibility and credibility

Singapore: A High Import Share, Trade Volume Far Exceeding GDP

2026-03-15 14:48:37 · #格物/新加坡

Singapore has a high import share, and its trade volume far exceeds GDP; the exchange rate has a more direct and stronger effect on prices (especially imported inflation) than the interest rate does. At the same time, in an environment of highly mobile capital and interwoven offshore and onshore financial activity, a simple attempt to “lock down domestic interest rates” is quickly offset by cross-border capital flows, making it hard to serve as a stable policy anchor.

Trade (exports and imports of goods and services) can reach several times GDP, and imports make up a very high share of domestic consumption. External price shocks (energy, food, international freight rates, tradable goods prices) transmit rapidly through the import channel, so “letting the local currency appreciate/depreciate relatively” usually has a more direct effect on inflation than “raising/lowering the domestic short-term interest rate.”

A Newly Opened Hotel: The Stay and the Uncertainty

2026-03-15 23:43:27 · #格物/柬埔寨

While travelling, I checked into a hotel booked through a reservation platform. I won’t name it. The photos looked fine and the price wasn’t high, so I thought I’d found a new place offering good value.

The room and shared spaces were more ordinary than I expected. The pool was still being filled. There was no Wi-Fi, and the staff gave me a phone hotspot. When I asked, I learned that the hotel had just opened and I was its first guest.

The staff were friendly, offering rides and helping order food. But with the facilities unfinished and responsibilities unclear, I never fully relaxed. I kept checking whether the door locked, whether it would be easy to leave, and whether I felt safe staying overnight.

No direct threat occurred, and nobody forced me to stay. My unease came mostly from uncertainty. What I can confirm is my experience and concern; these do not establish the operators’ backgrounds or motives.

Morning Observations at the Hotel

2026-03-16 11:06:01 · #格物/柬埔寨

Opening the door in the morning, I could see the pool beyond the light, courtyard, banana trees and greenery.

The pool’s edges still looked rough, with clutter beside it. It was different from the ready-to-use hotel pool I had expected. The corridor was quiet too: only birdsong, with no staff in view.

My impression was that the facilities and daily operation were not fully ready. Those images alone cannot establish anything about the operators’ other backgrounds, and my unease is not evidence of that.

In Siem Reap There’s a Shop I Really Liked

2026-03-16 13:42:23 · #格物/柬埔寨

A local-run place, but the owner looks a lot like a Chinese person — maybe of Chinese descent? He doesn’t speak Chinese. Actually that seems pretty common in Cambodia, Vietnam, Thailand and Malaysia. It’s called ហាង គុយទាវស៊ុបខ្មែរ, which translated directly just means Khmer soup noodle shop / Khmer soup noodle house.

The owner doesn’t speak English and doesn’t speak Chinese, but the storefront has Chinese couplets pasted on it.

There was no English menu and no Chinese menu. The owner nodded friendly and pointed at the noodles, pointed at the meat, and I nodded to say that was fine.

Very fresh beef, and a style similar to a mix with Vietnamese rice noodles — very Southeast Asian in character.

A very typical Cambodian rice-noodle soup system, historically influenced by regional exchange, with a lot of crossover with Chinese noodle soups, Vietnamese pho, and the Chaoshan/Fujian soup-noodle traditions, so eating it you get a feeling of “not pure Chinese food, not pure Vietnamese pho, but a Southeast Asian local fusion.”

They seemed very friendly~

China’s First-Tier Cities Are Overall More Expensive, but Many Everyday Items

2026-03-16 14:35:01 · #格物/柬埔寨

China’s first-tier cities are overall more expensive, but many everyday items are actually cheaper or better value in China.

Cambodia gave me the same feeling as Nepal: day-to-day spending isn’t low. Partly because it’s a tourist city, and moreover, in a tourist city like Siem Reap (Krong Siem Reap) rent isn’t cheap. Local snacks, taxis, delivery, online shopping, industrial goods, electronics, public transport — in China these are often cheaper and more stable, or “better quality for the same price.”

This is because of China’s extraordinarily strong supply chain and logistics density; Chinese manufacturing, delivery networks and platform competition drive many daily goods, dining and transport prices very low. Cambodia’s market is smaller, more import-dependent, and distribution efficiency lower, so many seemingly ordinary goods and services aren’t cheap.

In China the path from production to delivery is almost frictionless: the fixed cost spread over each unit of goods is lower, circulation loss is lower, competition is fiercer, turnover is faster.

China is still one of the world’s largest manufacturing countries, and its consumer market and online retail scale are also enormous; in 2025 China’s total retail sales of consumer goods reached 50.1 trillion yuan, with online retail sales of physical goods accounting for 26.1% of the total. At the same time, China’s express delivery volume reached about 199 billion parcels in 2025, showing that its distribution network already runs as an ultra-high-frequency, ultra-large-scale system.

The bigger a country’s market, the easier it is for the same good to achieve mass production, mass purchasing and mass distribution. Fixed costs — factories, molds, warehouses, software systems, operations teams — can be spread over more goods, so the per-unit cost falls. China’s advantage lies exactly here: not only is manufacturing scale large, the consumer market is large too, and both online and offline can scale up quickly, so many goods start out not as “small batch, high margin” but as “large batch, low margin, high turnover.”

But in scarce core downtown land in first-tier cities, in high-end services that are heavily labor-dependent and hard to scale, and in resource-based services in education and healthcare, China isn’t necessarily cheap — it can be even more expensive. This isn’t a contradiction; it’s because these areas can’t have their costs compressed to the extreme by supply chains and platforms the way industrial goods can.

It seems scarcity economics is moving toward abundance economics: AI and humanoid robots can produce almost anything we need, including goods, services, healthcare and education, at close to zero cost.

When AI cuts production costs by 30%, the value created either flows entirely to shareholders or partly returns to society.

So AI can create so many goods and services that prices collapse.

Your money can buy unlimited resources and labor, with no extra work required, and the basic standard of living for humanity can be raised.

AI raises productivity, but at the same time it redistributes who gets access to that productivity.

Impressions of Wat Bo

2026-03-16 16:53:32 · #格物/柬埔寨

Built in the 18th century, it’s one of the oldest and most important Buddhist temples in Siem Reap. Its full name is Wat Reach Bo (Wat Raja Bo), it has more than 100 monks, and it’s the largest active temple in Siem Reap province. It’s a place of practice for Theravada Buddhism and is still a working pagoda — monks live, study and chant here.

The temple is very peaceful and quiet, without many tourists; there are some local devotees and many stupas.

Wat Bo’s most unique value lies in the murals inside the main hall (vihara) — Hindu narratives.

Siem Reap province was under Siamese (now Thai) control until 1907. This means Wat Bo’s construction and decoration happened during a period of deep Thai-Khmer cultural intermingling. The Thai elements in the murals aren’t accidental — they’re the direct trace of that colonial/suzerain relationship.

The architectural style shows traditional Khmer features: curved spires, red-tiled roofs and finely decorated colonnades. The main hall (vihara) has a triple-tiered roof structure with wooden gables preserved in situ. The main structure is built of red brick, with fine wood carving decorating the building throughout.

For centuries, monks were the only literate people in the countryside; they served as teachers, instructing temple attendants, novices and newly ordained monks. Until the 1970s, most literate Cambodian men gained their literacy entirely through education in the monastic order.

That is to say, becoming a monk wasn’t just a religious choice, but almost the only channel for acquiring knowledge and social mobility.

Monks were also often healers; in traditional Khmer culture their role is the closest thing to a modern psychiatrist. They might also be well versed in astrology. This means a monk played multiple roles at once: teacher, psychological counselor, ritual presider, moral authority.

Laypeople provide monks with food and daily necessities, and monks, through their presence and teaching, give laypeople a way to accumulate good karma. Monks follow the precepts and, as alms-beggars, receive the four necessities of food, clothing, shelter and medicine from the faithful.

The sense of rupture in the Khmer Rouge era: monks were forced to disrobe and do manual labor, religion was banned, and temples were destroyed or damaged. It’s estimated that as many as 50,000 monks died during this period.

There are many stupas; the most primitive form of this kind of building is a mound of earth. In ancient Buddhist texts, the original chedei is simply a grave mound, built to hold the Buddha’s relics and ashes.

But in the Khmer Rouge era, these customary cremation or burial procedures weren’t carried out either; a great many people died violently and were thrown into mass graves, uncremated. Those souls were trapped, unable to reincarnate — not just political violence, but a disruption of the entire cosmic order. So that transparent stupa at Choeung Ek is like a belated vessel for those dead who never received the proper rites.

Another Thing I Saw, at Wat Bo

2026-03-16 16:56:54 · #格物/柬埔寨

Another thing I saw, at Wat Bo temple I also came across a personal blessing ceremony, and I remember earlier in Thailand I also ran into monks giving guidance

Worshippers come to the temple, kneel in front of a monk, and ask for a personal blessing. The monk chants Pali scriptures, and may then lightly sprinkle holy water on the worshipper’s head and body with a bundle of grass or a small broom dipped in water. This is called ស្រោចទឹក (srauch teuk), the water-pouring blessing

The reasons worshippers come to ask for a blessing are very varied—perhaps safety before a long trip, recovery from illness, difficulties in business, family conflict, the start of a new job, or simply routine merit-making. No appointment is needed in advance, you can go at any time

Inside the wat, when worshippers sit down they bend their legs and tuck their feet to one side, and when greeting they press their palms together (Som Pas) and bow to the ground three times. A worshipper’s head must be lower than the monk’s head—this isn’t just etiquette, it’s a bodily expression of cosmic order: the monk, as one who has left the household life, stands higher on the spiritual hierarchy than laypeople

Besides everyday blessings, there’s also consultation and counselling, and merit transference (transferring merit to family members who have died, so they get a better destination in the cycle of rebirth)

The monk ties a white or yellow cotton thread around the worshipper’s wrist—that’s ខ្សែដៃ (kse dai), the protective string. Worshippers usually wear this thread until it falls off on its own, believing it offers continuous protection; I came across this in Nepal too

Goh Keng Swee

2026-03-16 17:19:28 · #格物/新加坡

My favourite figure from Singapore

He was the chief architect of Singapore’s economic take-off

The hardest bones to chew in Singapore’s early years as a nation—the economy, finance, education—almost all of them he handled single-handedly and laid the foundations for

In the early years of nation-building Singapore had extremely high unemployment and faced a crisis of survival. After Goh Keng Swee became the first finance minister, he made a crazy decision: reclaim land from the sea, and build the largest modern industrial park in all of Asia in the raw swamps and jungle of Jurong in the west

He did whatever it took to build roads, supply water, supply electricity, set extremely favourable tax policies, and go around the world drumming up business

After industry, he began long-term planning for Singapore’s wealth

He single-handedly set up the institution equivalent to Singapore’s central bank, and built an extremely strict financial regulatory system, making Singapore one of the safest and most transparent financial centres in Asia

By 1981 Singapore’s foreign exchange reserves were already very substantial. Goh Keng Swee believed that money sitting in a bank earning interest would be eaten away by inflation, so he founded the Government of Singapore Investment Corporation (GIC), taking the nation’s money and investing it globally to make more money. Today, GIC is one of the largest sovereign wealth funds in the world

Goh Keng Swee’s Unique Vision and Its Influence on China

2026-03-16 17:21:11 · #格物/新加坡

He didn’t sit in a Beijing office drinking tea; instead he brought a team (even paying for his own air tickets and study-tour costs) and studied China’s coastal special zones (such as Shenzhen, Xiamen, Zhuhai and others) extremely thoroughly

Don’t be afraid of foreign capital exploitation—foreign companies come for profit, but the technology lasts, so be bold in giving foreign capital concessions (investment philosophy)

Don’t blindly borrow foreign debt; many developing countries built on heavy foreign borrowing and ended up in debt crises; you must attract direct investment (FDI)

The Feeling Siem Reap Gives You

2026-03-16 19:35:42 · #格物/柬埔寨

It feels very different, and this kind of different is hard to find a substitute for

Actually, after coming to Cambodia, I also ran into robbery and casinos

But… the feeling on the first day in Siem Reap

Siem Reap really is special

A world-class ancient civilisation site + a walkable small-town scale + a Southeast Asian mix of Buddhist and French colonial character + cheap + café/digital-nomad friendly + not quite a big city but not so small that it’s boring

Different from every city I’ve been to before; similar cultural small cities, Penang, Hoi An, and other small cities in other countries, none of them seem able to substitute for it

Siem Reap looks like it’s been dragged up by a super-civilisation site; the whole city lives inside Angkor’s gravitational field

In the morning you get up and thousand-year-old temples, moats, ancient trees and stone carvings are all there to see at any time

At midday you come back into town to eat, drink coffee, get a massage

At night you can go back to the night market, Pub Street, the riverside—very contemporary, very relaxed spaces

Especially at this point when the tourist numbers aren’t off the charts and Chinese people are rarely seen

Siem Reap Really Has a Very Special Temperament

2026-03-16 19:36:42 · #格物/柬埔寨

Wat Bo: the temple has an old-fashioned, very quiet sense of breathing

BIOLAB CAFÉ: a café with “lab” in the name is generally worth trying

Wat Preah Prom Rath: that feeling of a temple pressed right up against everyday life is especially obvious here

Just an ordinary small town, slowly sunbaked

Temples, cafés, the roadside, tuk-tuks, monks, tourists, dogs, tree shade, all mixed together

If a Person Treats Deception as a Tool, Habitually Using Concealment

2026-03-17 00:50:54 · #格物/柬埔寨

If a person treats deception as a tool, habitually using concealment, manipulation and clever talk to achieve their aims

The cost of deception is very low, and there’s a lack of genuine guilt

I feel it’s hard to change

Because there’s no real price paid, no sustained action to change, most likely they’ll keep deceiving

Wow, really, I watched that video by Yingshi Jufeng on Bilibili

2026-03-18 00:38:20

Wow, really, I watched that video by Yingshi Jufeng on Bilibili, and I thought it was absolutely incredible. The Bangladesh it introduced — Bangladesh is itself a very densely populated South Asian country shaped by rivers. It borders India to the west, north and east, with a short border with Myanmar to the southeast and the Bay of Bengal to the south. Bangladesh’s official geographic overview The country isn’t that big, but its population is enormous. Its capital is a capital with a huge number of people. And it sits on the world’s largest river delta, so its land is extremely fertile, which also dooms it to be like many Southeast Asian countries: agriculture matters enormously. The population is also very dense, infrastructure is very hard to build, and the climate is very fragile. What’s really interesting is that it wasn’t a naturally unified long-standing country. Bangladesh places great importance on one thing: it’s not only floods, not only poverty, not only cheap labour. Under severe resource constraints, it’s a developmental society built on population, on manufacturing, on women’s labour, and on social mobilisation. It’s one of the world’s 35th largest economies, but in the future there’s a lot of… The money for getting out of poverty is much more now too — since 2010 or so, more than 30 million people in Bangladesh have escaped poverty. But it’s still quite hard right now. Its social structure is very young, but it also very much emphasises a relationship-network society. I remember its capital, Dhaka, is the densest and most congested big city in the world. But its vitality is super strong, and the pressure on infrastructure is enormous. There are also a lot of motorcycle taxis. I think this country is quite interesting. It sits at the junction of India, China, and some parts of Southeast Asia. Its climate is especially interesting too.


Behind Angkor Wat Lies a Very Complex History and Knowledge System

2026-03-18 14:47:11 · #格物/吴哥窟

You can’t treat it just as a temple; you have to look at it as the ruins of an imperial capital complex. UNESCO defines Angkor as one of the most important archaeological sites in Southeast Asia, covering around 400 square kilometers—not just Angkor Wat itself, but also multiple capitals of the Khmer Empire from the 9th to 15th centuries, temples, water systems, and road networks.

Angkor’s soul roughly has three kinds:

The first is imperial orthodoxy and cosmic order, represented by Angkor Wat;

the second is the face of royal power and religious transformation, represented by Bayon and Angkor Thom;

the third is a sense of ruins, of forest, of time, represented by places like Ta Prohm, Banteay Samre, and Beng Mealea.

These are three ways I understand the sites; they do not map one-to-one onto the Small Circuit, Grand Circuit, and Outer Circuit.

Among the Outer Circuit, the one most worth a special trip is Banteay Srei, followed by Beng Mealea.

The ranking of the individual sites most worth visiting:

Angkor Wat

Bayon

Ta Prohm

Banteay Srei

Preah Khan

Beng Mealea

The wider Angkor site preserves remains of successive Khmer capitals; Angkor Wat is one temple within it. UNESCO on Angkor

In its early period Angkor was deeply influenced by Hinduism, and later it gradually shifted to Mahayana Buddhism, then layered on Theravada Buddhist influence. The most typical is the move from imperial symbols leaning toward the Brahmanical system, such as Angkor Wat and Baphuon, to the Buddhist royal-power expressions of Jayavarman VII’s era, such as Bayon, Ta Prohm, and Preah Khan.

Suryavarman II and Angkor Wat

2026-03-18 15:18:06 · #格物/吴哥窟

Suryavarman II commissioned Angkor Wat in the 12th century. Originally dedicated to Vishnu, it later became a Buddhist sacred site. Its gallery reliefs depict epics such as the Ramayana and Mahabharata, as well as the king’s procession. APSARA on Angkor Wat

What interests me is how the architects combined the temple-mountain and galleries, turning a religious cosmology into a space one can physically enter.

Angkor Wat’s Model of the Cosmos

2026-03-18 15:26:32 · #格物/吴哥窟

It lays out the Hindu, and later Buddhist, view of the cosmos.

Angkor Wat’s cosmology is the core key to understanding the whole complex. In the Khmer civilization’s worldview, the cosmos has a fixed structural model, and Angkor Wat is the material reproduction of that cosmos.

Mount Meru: the center of the cosmos and the gods. In myth, Mount Meru is the center of the universe (the world axis) and the dwelling place of the gods. The five progressively taller stone towers arranged in a “quincunx” at the center of Angkor Wat are precisely meant to reproduce the five peaks of Mount Meru. Through the corridors and towers rising level by level, the architecture symbolizes the pilgrim’s arduous climb toward the realm of the gods. By reproducing the sacred Mount Meru on earth, Angkor Wat anchored the Khmer kingdom at the center of cosmic order, and established the legitimacy of the king who built it as a sacred mediator between gods and mortals.

Cosmic Ocean: the boundary between the sacred and the profane. Angkor Wat’s outer perimeter is surrounded by massive stone walls and a moat up to 190 meters wide. These outer walls and moat were not merely fortifications or waterworks; they symbolize the ranges of mountains and the vast “cosmic ocean” encircling Mount Meru. On the spiritual level, the water reinforces the boundary of sacred space: when a pilgrim crosses the moat, it marks their crossing of the “cosmic waters,” completing the inner transition from the mortal, profane world into the sacred center of the cosmos.

Echoing the myth of the “Churning of the Ocean of Milk.” The symbolic meaning of the cosmic ocean is also tightly linked to the most famous giant relief at Angkor Wat, the “Churning of the Ocean of Milk.” That myth tells of gods (Devas) and demons (Asuras) cooperating to churn the cosmic ocean in order to obtain the nectar of immortality. In this grand architectural context, the moat surrounding Angkor Wat is seen as the incarnation of the “ocean of milk,” while the temple at the center represents the sacred cosmic order. This combination of water and architecture vividly expresses the Khmer belief about creating life out of chaos and maintaining cosmic balance.

The Prosperity and Fragility of a Water System

2026-03-18 15:43:52 · #格物/柬埔寨

Angkor’s vast water system makes me think that prosperity and fragility can come from the same structure. Research examines hydroclimatic changes, including droughts in the 14th and 15th centuries, alongside Angkor’s decline. This does not establish a single cause or support the claim that rice cropping fell from three or four harvests a year to one or two. Research on occupation and the tempo of collapse at Angkor Wat

What I want to understand is how a society so dependent on a water network copes with environmental change and the demands of long-term maintenance.

Three Hours at the Angkor Museum

2026-03-18 18:50:11

Today I spent three hours at this Angkor museum — it’s a museum here in Siem Reap, one in Siem Reap, and I felt it was pretty interesting. Before coming I had actually read up on Angkor Wat, and on the history of Siem Reap, including some Khmer history, and it all felt really interesting. I wasn’t actually considering coming to this museum, because it charges quite a lot — this time it cost me 18 US dollars, which feels on the high side for people like us who aren’t earning US dollars. But after seeing it, I thought the experience was very good, because although it doesn’t have all that many truly precious artifacts, it’s a single thread that helps you understand what the whole history of Angkor was like, and the background relationships between the various Buddha statues, the various goddesses, the various gods and Buddhas and immortals, including the background relationships of Hinduism and the basic relationships of Buddhism — I found that very interesting. The museum is like a proof of presence: it lets you walk through it yourself and makes you feel that these things aren’t only to be learned online or from books, but that you can also go to the actual sites and look at the corresponding ancient monuments, some very precious exhibits, look at them very closely, observe their details, observe their workmanship, observe their carving art. And then there are all kinds of wooden Buddha statues, stone Buddha statues, metal Buddha statues, silver figures. And I just marvel at how fine the workmanship is — it’s really impressive.

It Suddenly Hit Me — the Whole Learning Process

2026-03-18 18:55:59

It suddenly hit me that if I were to sum up the whole learning process right now, I’d say it’s really interesting. At the beginning I used some AI tools to quickly help me understand the history of this place, and it would organize things into chapters for me; for instance, when introducing the history of Angkor, or the history of the Khmer, or the history of Cambodia, it would arrange it and tell me the original historical stories of this place, and then the later turns — why they chose Hinduism, and why they later shifted to Buddhism, and their most important industries, like agriculture, and why agriculture mattered so much, and what’s closely tied to agriculture here, namely the water systems, and how the water systems and the weather here affected this country, and then later some of the modernization process. And some of the famous sights here. For example, the Cambodian flag has Angkor Wat on it, and Angkor Wat’s influence on this place — it’s like a calling card for Cambodia. And then it introduced the history of Angkor Wat, some of the architectural features inside Angkor Wat.

Then I actually went to see this museum in person, I went into the museum and experienced it firsthand. And in the process I also used AI to help me record things, including taking some photos, or sending in some corresponding terms, and it also helped me expand on them in detail — it felt like learning by looking at real objects. It would introduce which gods there are, which Buddhas there are, which immortals there are, which historical background stories there are, and what they specifically look like, including the dancing girls — all blending together, and it helped me break them down one by one, helping me identify these ancient monuments, and the corresponding terms, the background stories they specifically map onto, their specific history, and what their specific myths look like. I think this is very helpful for me to learn in an embodied way. And then the deepest learning method was to go to Angkor Wat and feel the most famous place here, the Angkor Wat that is renowned the world over. Specifically, to feel it in greater depth inside Angkor Wat — what the real experience is like. I think this process is, at Angkor Wat, something like an embodied experience, bringing yourself into that environment to feel this story that is so rich in history and culture, such a legend in human history. That’s very embodied, and I think it’s really interesting. It’s like a subjective experience, a personal experience. And then this learning method basically goes: from how to use AI to learn quickly at the very start, and then how to, concretely, through real objects, empathize — by observing the details, through your own eyes, to observe what the real object specifically looks like. Through being there in person to feel it, and then through bringing yourself into a historically meaningful environment, an environmental field, to genuinely understand the historical story, and to feel the growth that this cultural monument brings you.

Highly Recommend the Angkor National Museum Before Going to Angkor Wat

2026-03-18 19:54:34

Highly recommend coming to the Angkor National Museum before going to Angkor Wat.

It covers the history, civilization and cultural heritage of the Khmer.

I’m moved by the worldview and the narratives the Khmer civilization left behind.

How did a civilization persuade tens of thousands of people that the same undertaking was worth a lifetime? I want to understand the respective roles of belief, social organization, and the mobilization of labor.

Preah Ang Chek Preah Ang Chorm Shrine

2026-03-20 12:22:17 · #格物/柬埔寨

Preah Ang Chek Preah Ang Chorm Shrine

You could call it Siem Reap’s guardian shrine.

I saw worshippers offering incense and lotus flowers, and some washing the statues’ hands and feet with holy water.

This sacred space seems closely connected to everyday local life. It makes me want to learn more about its story.

Rented a Motorbike

2026-03-20 19:28:21 · #格物/吴哥窟

The motorbike was originally 13💲/ day; for three days it cost me 24💲

I’m pretty good at haggling, hehe

but it’s still so expensive

Tomorrow the Angkor itinerary starts, looking forward to it; along the way I’ll keep recording, and keep having a few more thoughts


Another Bowl of Pho

2026-03-21 10:34:36 · #格物/吴哥窟

Went to this place again today. In the morning I was about to go wander the Angkor big circuit, and passing this shop I thought I’d eat here again. Still very friendly; the grandpa recommended some noodles and meat

The pho still tasted delicious; the price feels slightly high, three US dollars, though locally—what I ate before seemed much cheaper


I still feel it’s Preah Khan that has more of a run-down feel

2026-03-21 15:46:44

I still feel it’s Preah Khan that has more of a run-down feel; as for this, uh, water temple now, it’s really built around one temple. At first it was for giving blessings to locals, because you pass by a pond here, and the scenery of the pond is quite nice to look at, so they came to believe that this water washes away sin


Talking About Angkor Wat with Friends in Siem Reap

2026-03-22 01:10:12

I got to know some local friends in Siem Reap, and we talked about Angkor Wat. We seem to treat cultural heritage from our own country, such as the Forbidden City and the Great Wall, as a calling card, finding belonging and a sense of achievement in it.

That feeling also came through clearly in talking with local friends. They felt proud of Angkor Wat and saw it as a window through which others could get to know Cambodia. I found that interesting.


Siem Reap Ta Prohm Temple

2026-03-22 22:23:00

Siem Reap Ta Prohm Temple, an absolutely beautiful secret place

It’s also a filming location for Tomb Raider

My feeling after going there—this temple gives me the sense that it’s very mysterious, it really fits my definition of a secret place


There are really so many Americans; an American came to the hotel today too

2026-03-22 23:58:32 · #格物/柬埔寨

I feel like there are really so many Americans; an American came to the hotel today too. He seems super outgoing, dying laughing, and he has a beard too—the whole person gives me the impression of someone with a lot of personality, is this American freedom? LOL, and people are really so outgoing

We ate together in the evening, and I really didn’t fit in—I don’t smoke, I don’t drink


I always thought Angkor Wat was very big and very spread out

2026-03-23 00:02:12 · #格物/吴哥窟

I always thought Angkor Wat was very big and very spread out, but it seems that’s not the case. Angkor Wat by default in fact refers only to that one sacred zone in the middle. Angkor Wat is actually a miniature building built with Hinduism as its world model

It was built by Suryavarman II

Angkor Thom was built by Jayavarman VII, emphasising “rule + people’s livelihood + a turn to Buddhism”

Angkor Wat is a world of gods, a description of a religious cosmos


Five towers (Mount Meru) + the four faces of Bayon

2026-03-23 00:05:06 · #格物/吴哥窟

The five towers (Mount Meru) and Bayon’s four-faced stone images make me think about the layering and blending of religious traditions at Angkor

Because it seems that in the time of the seventh [reign], because the empire had long had religions piling up and blending, the process of iterating into Buddhism was often not a simple replacement but a process of fusion. So the way it existed here in the Khmer lands was a long-term coexistence of Hinduism and Buddhism

He built Bayon Temple (“the smiling Buddha faces”)

Emphasising compassion and the salvation of all beings

It shows that when a great culture merges into a local place, everyone is actually taking a blending approach. That is, in order to adapt to one’s own original culture, while the modern culture is more conducive to maintaining the new order and understanding the new world, so the conflicts that arise between them often take the form of fusion rather than replacement alone. This observation cannot be generalized to every country’s cultural change

This includes Tibetan Buddhism: the Bon religion at the start, then later replaced by Buddhism—that too was a process of fusion. There are still many traces of Bon today


Angkor Wat is really a place worth studying in detail

2026-03-23 00:06:51 · #格物/吴哥窟

I think Angkor Wat is really a place worth studying in detail, because to understand it you actually have to understand Hinduism—especially its axis, its proportions, its symbolic system and its moat, and the meanings these correspond to within Hinduism. So only by understanding Hinduism can you understand Angkor Wat

Bayon is very interesting. If order is my strongest impression of Angkor Wat, Bayon feels more spiritual to me. At the center of Angkor Thom, its many four-faced stone images are highly recognizable, and its expressions of royal power, religious transition, and narrative interest me too.


The coolest thing about Angkor Wat is that it turned the Hindu cosmic model

2026-03-23 00:28:44 · #格物/吴哥窟

The coolest thing about Angkor Wat is that it translated the Hindu cosmic model into architecture almost completely

Including the five main towers: Mount Meru, that is, the centre of the Hindu cosmos, the dwelling place of the gods

The moat is the cosmic ocean, not just a defensive structure. According to the traditional explanation, it symbolises the cosmic sea surrounding Mount Meru. That is, as a visitor walks step by step in from outside, they are completing a symbolic journey from the “human world” into the “centre of the divine realm”

Walls and layer upon layer of galleries: cosmic boundaries and hierarchical space. The whole building advances from outside inward, from low to high: outer wall, long causeway, galleries, platform, main tower, forming a very clear hierarchical order. Britannica summarises Angkor Wat as an “artificial mountain” rising level by level to the top within three enclosures; this is exactly the typical Khmer “temple mountain” logic

Actually when I was at Angkor Wat I felt this very deeply: going up feels like advancing layer by layer, there seem to be three enclosures, and each time you basically go up a step; going up a step feels like climbing a mountain, it’s just that the mountain has been made into this kind of palace form. So I feel that the further in you go, the more sacred it feels, and the higher you go, the closer you feel to the gods

And actually there’s a very important point about Angkor Wat: like the Forbidden City in China and other grand designs, it is a machine of order—absolutely symmetrical, absolutely ordered. It also emphasises axis, symmetry, repetition and progression. And I think, as the pinnacle of classical Khmer art, this is very interesting

And have you noticed, this time we went in from the west gate, which means at the moment you go in from the west gate you can see the sun rising from the east, and then you can see the sun above the main tower; and it happened to be a special day today, so we could capture the moment the sun rose above the tower. And I think that moment was so beautiful—no wonder it’s called the most beautiful sunrise in the world, because this sunrise carries intense colour, religious faith, scientific order, and also the expectations of countless people, plus some very good-spirited crowds, really interesting

Angkor Wat was originally dedicated to Vishnu, not Shiva. This point is very important, because many Angkor buildings belong to the Shiva system, but Angkor Wat’s original religious affiliation more clearly leans toward Vishnu

And it’s also linked to a very famous episode in Indian Hindu mythology, the Churning of the Ocean of Milk: the gods and the asuras used a giant serpent as a rope and churned the ocean of milk around the sacred mountain, in search of the nectar of immortality

This corresponds directly to the design of Angkor Wat


Buddhism and the Religious Environment of Ancient India

2026-03-23 00:30:31 · #格物/吴哥窟

Buddhism arose within the religious and philosophical environment of ancient India. I want to understand how Buddhist texts engage with contemporary ideas about gods, ritual, and spiritual practice.

I should not describe the Buddha as personally writing Buddhist books, or project the later, fully developed Hindu triad directly onto early Buddhism. The dates of different texts and ideas need to be checked separately.


Banteay Kdei Temple

2026-03-23 00:37:17 · #格物/吴哥窟

This is very interesting

It was built in the time of King [Jayavarman] VII, that is, at that time they were turning to Buddhism, which means its religious character is a Mahayana Buddhist monastery. So when I went in I actually saw many people kneeling and worshipping the Buddha there. The way the Buddha is worshipped here makes me want to compare Theravada and Mahayana iconographic traditions; it does not justify saying that Theravada temples contain only one kind of image

This place is a practice city where monks lived, you could also call it a city of rooms. So it isn’t simply a building for worship; it also carries certain functional attributes and a practice function

I think its wall structure is very interesting—whether it’s Angkor Wat or similar, the architectural style of the Khmer period here, I think much of it accords with the Hindu worldview. It has multiple enclosing walls, becoming gradually more sacred from outside to inside, and the middle represents the centre of the cosmos, that is, Mount Meru

It seems their building materials are different too: some of them, the ones that feel very dilapidated, mainly used sandstone and inferior stone, so they collapse more easily and more easily take on a ruin style

It is extremely quiet; compared with some other temples, there are almost no tour groups at all. And it has a very strong sense of being eroded by time; at the back I even photographed a huge tree, so it feels like photos really come out well here

Ta Prohm Temple

2026-03-23 00:47:05 · #格物/吴哥窟

Angkor’s most famous ruin in time.

One of the most recognizable temples.

Ta Prohm was built in the reign of Jayavarman VII; originally named Rajavihara (royal monastery), it was a Mahayana Buddhist monastery and learning center.

It is closer to a large monastery and a combined study-and-practice complex. Built in the late 12th century and dedicated in 1186, it contained an image of Prajnaparamita associated with the memory of the king’s mother. APSARA on Ta Prohm

Its religious meaning and emotional coloring feel distinctive to me, with a sense of family commemoration.

Its counterpart is Preah Khan, which I went to yesterday.

Inside, stone and trees seem interwoven. I still need more reliable historical sources to understand how abandonment, continued use, and maintenance changed here.

I think this is a very rare state: its civilizational remains didn’t completely conquer nature, nor were they completely erased by nature, but rather exist in a balance of tension.

From outside to inside it’s basically the same — it also matches a Hindu cosmology, with sanctity gradually rising, going from broad to enclosed, and the sense of pressure slowly increasing too. It gives you a feeling that you’re not visiting a display item but traveling through a religious site, which is a very experiential thing for a person. I think that’s really interesting. And I also think another very, very important reason it’s famous is that Ta Prohm’s value isn’t just the building itself, nor just the trees, but that “the ruin landscape itself” has become part of the heritage. That is, what you see isn’t simply “nature damaging a cultural relic”

but a cultural landscape that has been managed, stabilized and protected: keeping the drama of trees and temple, while trying as far as possible to prevent it from continuing to destabilize and collapse.

It also makes me think about balancing conservation and the visitor’s experience. Seeing a site up close is precious, while touching it or approaching particular areas still needs to follow the site’s conservation rules.

The most easily underestimated thing about Ta Prohm is that back then it wasn’t “a lone temple.” Inscription records show it was once linked to an enormous system of people and supplies. Public sources mention that the temple once had tens of thousands of resident or service personnel, and that villages around it had large numbers of people providing it with goods, labor and religious services.

Thommanon and Chau Say Tevoda

2026-03-23 00:56:05 · #格物/吴哥窟

Thommanon Temple (Thommanon Temple) and Chau Say Tevoda (Chau Say Tevoda).

Thommanon is a Khmer Hindu temple from around the 12th century, generally believed to have been built around the time of Suryavarman II — that is, roughly the same era as Angkor Wat. It’s small in scale, but in very good condition, and is a very typical example of a refined small temple in the Angkor area.

A highly finished sample of classical aesthetics.

Thommanon is generally regarded as a Hindu temple, connected to the Shiva and Vishnu systems. Many temples at Angkor went through religious conversion in history, so today it can’t simply be understood as “belonging to a single deity,” but its overall character still leans more toward the Hindu tradition.

Chau Say Tevoda is a bit more complex, with more layers.

Baphuon Temple

2026-03-23 01:06:35 · #格物/吴哥窟

It’s a very, very popular temple. It’s one of the buildings that best embodies Angkor’s sense of a sacred mountain at the center of the empire. It too originated from Khmer Hindu state religion, and is generally described as dating to the mid-11th century; its precise attribution remains a subject of research.

This is a very ambitious temple-mountain temple, because once you climb up you can feel how very tall and very steep it is. I think from up there you can see everything below at a glance — right beside it is a command platform, the king’s reviewing stand, and around it all kinds of big trees and greenery. Because this kind of building is very, very massive — it’s a mountain piled up out of stone, like a stone mountain — it’s different from those refined buildings. Those refined buildings are even afraid of being damaged by plants, like the building in that film we watched earlier, but this one isn’t. It’s like stone: it can effortlessly coexist with stone, and only stone has to adapt to it, not the other way around.

Baphuon is essentially a towering multi-level terraced temple-mountain.

It’s the architectural expression of a religious cosmology, a very representative form: the lower the level, the closer to the profane world, and the higher you go, the closer to the sacred center.

In front of it there’s also a very long bridge-like causeway — that’s like a sense of ritual, slowly bringing you in. Its surroundings were designed too, so it’s like something that exists as a very, very important central control.

When I climbed it, I felt it was quite hard to get up — the slope is quite steep, and some of the steps are quite narrow, so it still feels like the risk factor is fairly high.

The many stone blocks below also made me curious about the restoration. After a long interruption, EFEO restarted major restoration work in 1995, and an inauguration marked its completion in 2011. Identifying and repositioning the stones feels like solving an enormous three-dimensional puzzle. EFEO’s restoration study

Identifying stone shapes, positions, and relationships must have been difficult. What strikes me is that restoration means understanding changes left by different periods, as well as putting scattered stones back together.

On the Road from Siem Reap to Angkor Wat and Banteay Srei

2026-03-23 10:18:02

On the road from Siem Reap to Angkor Wat and Banteay Srei, I found a very local restaurant. It tasted very, very sour — I feel I’m still better off eating honestly in the tourist area. And I don’t know what the hygiene is like; I always felt it was pretty mediocre.

Three Straight Days of Motorcycling Around Angkor Wat

2026-03-23 21:41:38 · #格物/吴哥窟

My head feels like it’s splitting.

Lately I’ve spent almost a hundred just on fuel.

The outer circuit distances are especially far.

I don’t know why my head hurts.

After I got back, even turning my head slightly felt agonizing.

I felt much better after a shower.

It seems it’s not just breathing in dust, but also serious dehydration in the process, and the high heat.

Serey Rathana

2026-03-23 22:09:41 · #格物/吴哥窟

Serey Rathana himself is a highly skilled shadow-puppet carver. His original motivation for founding this place was that he discovered Cambodia’s traditional shadow puppetry and leather carving had almost been lost after the war, and many art schools in Phnom Penh and Siem Reap were closing one after another. He worried that this link in Khmer civilization would be severed entirely.

Rathana set up this center around 2002. At first, he simply wanted to teach children from poor local families and orphans a skill, so they could support themselves through art.

Rathana mentioned in an interview that when he carves, there are always many poor children gathered around watching. When he asked them why they didn’t go to school, the answer was often “I have no parents” or “my family is too poor and I need to work.”

He sees these children as “little angels,” and turned his own home into this center, personally teaching them how to turn thick cowhide into exquisite works of art.

The leather carving treats cowhide with natural materials, and every cut demands enormous patience; some complex pieces take even weeks to finish.

Back When I Was in Chiang Mai, I Passed Through Some Rather Divided Places

2026-03-24 11:07:07 · #格物/柬埔寨

Some neighborhoods I saw in Chiang Mai, and on this trip in Cambodia, felt divided: high-rises and walled residences were not far from makeshift homes.

I do not know each household’s income, property rights, or work, so a street view cannot tell me their lives. But increasingly I feel that understanding a place means looking at where ordinary people live, how they commute, and how they get by, alongside its prosperity.

The Previous One Was a hotel I Booked Here in Siem Reap

2026-03-24 12:42:08 · #格物/柬埔寨

The previous one was a hotel I booked here in Siem Reap, and I felt it was really mediocre, especially in a few respects

The food’s price and presentation did not meet my expectations; I paid six dollars per meal. I did not verify how it was prepared behind the scenes.

Then the WIFI was under repair the whole time; I stayed nine days, and it was under repair all nine days

And the room’s basic facilities were severely lacking; there was a lot written down, but in reality most of it wasn’t there, including the hairdryer, dryer, washing machine; the actual place also looked very different from its Booking photos; I cannot confirm whether AI was used to alter those images.

Why I Chose to Visit Cambodia

2026-03-24 13:01:33

Actually, before I came to Cambodia, because I was in Lhasa at the time, I found the concept of a “sermon gathering” very interesting, so at that point I chose one country: Cambodia. Cambodia is also a Buddhist-type country, and I was thinking that although Cambodia is also a Buddhist country, I feel it’s a country with a fairly large wealth gap. I wondered what role Buddhism actually plays in this country, and how it shapes the country’s current social form; I was very curious about that, and I’ve kept looking for answers on this part.

Observing people here, how they live, and how faith enters daily life seems very interesting to me. This was one reason I chose Cambodia.

And I’m very interested in Cambodia’s Angkor Wat, because although Angkor Wat was originally Hindu, it later converted to Buddhism. I’m actually very interested in its Hindu form too: how Hinduism spread at the time, and how it shaped people like this, and what influence this system still has on people now—that part is quite interesting.

A Person’s Underlying Colour

2026-03-24 14:17:56 · #格物/柬埔寨

Isn’t seeing it once enough?

Still holding on to illusions?

so, why do people still easily hold on to illusions

Sunk cost~

You’ve put in too much time, energy, expectation, even money

The good experiences you once had all get reinforced by memory

If a person is sometimes good and sometimes bad, it’s very easy to let you sink in

Sometimes very good → gives you hope

Sometimes very bad → makes you suffer

It’s like a gambling mechanism

Essentially it’s knowing the truth, but being reluctant to let go of the version you once believed in

Mount Meru: Distinguishing Different Traditions

2026-03-24 15:02:57 · #格物/吴哥窟

The cosmic model of Mount Meru is fascinating: like a world axis, it organizes the human, celestial, and subterranean worlds, surrounded by layers of mountains and seas.

This note mixed Hindu and Buddhist accounts. The realms of desire, form, and formlessness, along with Shakra and the Four Heavenly Kings, need to be understood within their particular texts and traditions, rather than treated as a single standard Hindu map.

What interests me most is how geometry and hierarchy make an invisible cosmic order imaginable.

Does the Sense of Awe Depend on Underlying Cognition

2026-03-24 16:52:46 · #格物/吴哥窟

Does the sense of awe depend on underlying cognition, or on a more fundamental human perceptual system

When I look at snowy mountains I feel my own smallness, I feel the snow mountain’s vastness

When I look at Angkor Wat I marvel at human belief, at the creative power of human civilisation

Angkor Wat itself is huge in volume, with long axes and ultra-deep depth, extremely high symmetry and order, and a great deal of mystery, like a maze

I always thought some kinds of awe depend on cognition

But now I feel that cognition may over-analyse to the point that you lose the original feeling…

When You Understand How a Magic Trick Works

2026-03-24 17:28:53 · #格物/吴哥窟

When you understand how a magic trick works, it’s very hard to experience the pure astonishment of magic again

This may be a side effect of high cognition

It’s like standing in front of Angkor Wat: if you only demystify it, what you see is a pile of weathered stone, a project built through the mobilization of large amounts of labor by an ancient monarchy, and you feel the tragic vitality of humans trying to leave a mark of existence in the face of an impermanent universe

How to Feel Again After Demystification

2026-03-24 17:33:54 · #格物/吴哥窟

Looking at Angkor’s weathered stones, I thought about what follows demystification.

If everything can be taken apart and every motive reduced to self-interest, does understanding end in nihilism?

I realized I was trying too hard to find one standard answer. Different judgments of the same person or thing can also reflect different needs and experiences.

Angkor, too, loses much if all that remains is a pile of old stone. After learning its history, I can still be moved by the human wish to leave a mark in time. Demystification can be followed by re-enchantment.

In Siem Reap I picked a Footprint Cafes

2026-03-24 20:03:35 · #格物/柬埔寨

Didn’t expect it to be a charity café!! Came here several times in a row to chat.

It opened its first shop in Siem Reap in 2016, with the core idea of People, Planet, Profit, and says it returns 100% of net profits to the local community in the form of education and entrepreneurship grants.

The space is great for coworking.

Their Chicken Lok Lak is famous — it reminds me of kung pao chicken, although the dishes are different.

The star of this dish is tender meat cut into cubes (chicken or beef), stir-fried over high heat to lock in the juices. The flavor is salty-savory with a slight sweetness, with a rich aroma of oyster sauce and black pepper.

Their breakfast is also superb.


Came again a day later, adding:

Lol, the waiters here even remember regulars’ habits.

The first iced Americano with no sugar, and the pizza made with a hard crust, take away, card to pay — they remember all of it.

The Secret

2026-03-27 19:07:24 · #格物/吴哥窟 #格物/秘密

One claim made in The Secret is that a person’s persistent thoughts and emotions, like a magnet, draw experiences of the same frequency into life.

This framework is essentially the law of attraction.

Whatever you think about most, wherever you most often put your attention, life will tend to produce that. That is what it calls “like attracts like,” “thoughts become things.”

It’s not just “thoughts” — it also includes whether you truly “feel you already have it,” which is why emotions get amplified.

“Focus on what you want rather than what you do not want” is part of this framework. I am recording the book’s claims, not establishing that thoughts directly determine external events.

Creating Angkor Wat Is a Miracle of Human Civilization

2026-03-27 21:05:25 · #格物/吴哥窟

It’s also a miracle for an engineer.

An extremely complex and precise system, from the layout of the water conservancy works, to the narrative on the walls, to the perfect fit between architecture and astronomical calendars.

Anyone used to building logic, designing systems, or polishing products can, when facing these ancient ruins, feel a resonance that spans a thousand years.

To Truly See a Ruin, You Often Need Some Kind of Medium

2026-03-28 13:48:28 · #格物/吴哥窟

To truly see a ruin, you often need some kind of medium. Through the viewfinder of a lens, a person starts to notice the details ordinary tourists overlook: how the first ray of morning light passes through the mottled stone doorway, how the expression of Bayon’s “Khmer Smile” changes under different light, or the exquisite Apsara nymph carvings on some broken wall. The process of patiently waiting to capture the ideal light and shadow is itself a kind of deep conversation with this land.

Pet-Friendly Cafés in China Have Clearly Become Much More Common Now

2026-03-28 14:45:21 · #发现

It feels like pet-friendly cafés in China have clearly become much more common now, especially in first- and second-tier cities—there are already a lot. But it hasn’t fully become the default state where you can bring pets in; a lot of cafés now already use the pet-friendly label as a new consumption scenario to work with.

For many, if they have a lawn, or a terrace, or a small courtyard, they’ll have a pet drinking area; there are even a lot of places now, including Shenzhen and Guangzhou, that are rolling out pet-friendly commercial districts, and communities, and some parks. It looks really nice now.

Siem Reap Has Many NGO-Run Social Enterprises

2026-03-28 15:43:19 · #格物/柬埔寨

Common Grounds Cafe, which I came to today, is one too, and it’s very famous.

It usually plays soft music, the atmosphere is quiet and not noisy, and the second floor has spacious seating.

They’ve been open since 2008, the coffee beans are fresh, and the quality has stayed steady. The menu blends things well—you can get authentic Western simple meals (like burgers, sandwiches) and also Khmer local specialties (like coconut curry).

It’s a social enterprise run by an NGO. Every bit of profit you spend here goes 100% into local charity and humanitarian projects in Cambodia. Not only that, they also specifically provide systematic culinary vocational training for poor local youth; many young people learn skills here and then go on to the country’s top hotels and restaurants.

Every coffee shop has a story

2026-03-29 11:49:12 · #格物/柬埔寨

The Little Red Fox Espresso.

It’s one of the most famous cafés in Siem Reap, with a fantastic atmosphere and super friendly staff. Their “ginger lemongrass latte” is very Cambodian, and very modern at the same time.

This shop was one of the pioneers of the whole Kandal Village artsy neighborhood. It turned what was originally an ordinary little street into Siem Reap’s most fashionable, most design-forward area today.

The shop often plays 1960s Cambodian rock’n’roll.

It folds the core spices of Southeast Asian cooking into a latte.

They insist on sourcing beans from local highlands like Mondulkiri (local coffee).

I looked for a coffee shop on the spot

2026-03-30 16:10:42 · #格物/柬埔寨

There’s pull over.

I was still wondering what pull over was.

Turns out it’s pour over coffee.

3. Reading, Ideas and History

54 entries

Dimension-Reduced World vs. Self-Dissolution: Two Survival Strategies Facing Complexity

2026-03-01 16:25:09

We always face this world—this world is complex, complexity beyond anyone’s imagination.

Some choose to reduce this world’s dimensions, simplify, no longer explore human nature, use single narrative system to describe everything, use labels to replace understanding, use positions to replace judgment.

Some choose to simplify themselves—admit they cannot control the world, self is no longer cognitive center, don’t treat everything as judgment against oneself.

…


A Concrete Act of Help I Saw in Lhasa

2026-03-01 18:58:37 · #格物/社会

I also encountered people begging in Lhasa, often around temples and busy visitor areas.

I do not know their individual histories or livelihoods. Their location alone does not tell me whether begging is an occupation or whether an organization is involved.

In a shop I saw the owner give someone an egg. That kind of concrete help also seems good to me; giving need not be limited to cash.


Different Styles of Service I Met While Traveling

2026-03-03 14:51:05 · #格物/社会

Some restaurant owners and servers I met in Lhasa made less small talk than I was used to and did not interrupt much.

That may have to do with the particular shop, familiarity, and local ways of life. It cannot establish a Tibetan personality or be attributed directly to Buddhism.

I also think of some people I met in Nepal who would chat even when I bought nothing. These differences in interaction interest me, while remaining observations from my limited travel experience.

A Hollow Feeling at Night: Recording the Experience First

2026-03-05 02:26:33 · #格物/生物

Tonight I noticed something different

Tonight I watched a film I really liked, “Night King”

A film I personally like a lot

After it ended I went back to the guesthouse, and the landlady invited me to play cards, and we played until very late, close to two in the morning

By the time I washed up it was now, and I felt completely hollow inside; that signal feels like it’s telling me I shouldn’t go play cards like this again

If it were good friends playing cards together, there wouldn’t be this hollow, quiet feeling

The quality of companionship really does seem to matter especially much

I tried to explain the feeling by saying my prefrontal cortex had “clocked off,” but that is a metaphor, not a neurological finding about myself. Sleep and circadian rhythms are connected with mood; I do not know whether this particular hollow feeling came from tiredness, socializing, or something else. NIGMS on circadian rhythms

Three layers of reality: the interweaving of the physical, the experiential, and the narrative

2026-03-05 02:45:09 · #格物/生物

Neuroscience describes how I function.

But it can’t tell us what we should become.

From the perspective of the physical world, reductionism really is “harder”:

Cut open the brain and you can see neurons

You can measure dopamine concentration

This leaves me wondering how subjective experience relates to physical processes, without reducing that relationship to a switch

This can’t be avoided. So then what is real?

The physical layer is the structural reality of neurons, dopamine, electrical signals

The experiential layer is the phenomenal reality of pain, curiosity, love, meaning

The narrative layer is the relational reality of culture, history, identity, value

Narrative and experience each have their own kind of reality; the experiential layer is what we live through directly


The limits of Aristotle’s dichotomy: how Quality breaks the split between subject and object

2026-03-05 15:15:18 · #格物/禅与摩托车维修艺术

Aristotle’s essentialist dichotomy

Splitting a complex problem or concept into two mutually exclusive categories (such as subjective/objective, form/essence, classical/romantic), understanding things through this oppositional relationship

The binary method actually has its limits too — it splits a person apart, separates subject and object, and hides the wholeness of things

It holds that in a moment of pure Quality, subject and object were originally one (as in the philosophical view “tat tvam asi”)

The real solution is to break the dichotomy, to fuse the person and technology through Quality (as in the “sense of oneness” when a craftsman throws himself wholly into the work)


One flower, one world: cognitive diversity and multi-level complementarity in the Huayan metaphor

2026-03-05 15:28:16 · #格物/禅与摩托车维修艺术

I suddenly got it

A very interesting place

Through different road conditions and levels of vision, it reveals the diversity of cognition and multi-level complementarity, and the switching between them

Looking at a motorcycle in terms of its components, understanding what these components do, necessarily requires further explaining its function

I came across these circulating lines, but have not verified their source; I should not present them as a complete verse from the Avatamsaka Sutra:

“One flower, one world; one leaf, one bodhi; one grain of sand, one pure land; one smile, one karmic bond.”

A flower is any tiny existence, a world is the whole universe, a leaf is something ordinary, found everywhere

Inside a single flower is hidden the entire world

Inside a single leaf lies bodhi (enlightenment)

In a single grain of sand there is a pure land

In a single smile there is a thread of worldly karma

You don’t need to look into the distance to seek the Way; the Way is in every present moment before your eyes — if the mind is pure, a mote of dust is a place of practice

I read “one flower, one world” as suggesting that every tiny thing is a projection of the complete universe; observe one flower and you can glimpse the law by which all things arise and pass away

“One leaf, one bodhi” — legend has it that a bodhi leaf witnessed Shakyamuni’s enlightenment, and every leaf carries the possibility of leading to wisdom

“One grain of sand, one pure land” — even a grain of dust can be a pure land; the realm is made by the mind, and the pure land is not on the other shore but in the present thought

“One smile, one karmic bond” — every romantic bond in the human world often begins with a look, a smile; arising and ceasing are both cause and effect


Why Important Experiences Can Feel Distant

2026-03-05 21:34:15 · #格物/进化心理学

Sometimes an important experience happened only months ago, yet feels years away. I wonder whether its distance from present life contributes to that feeling.

Travel, falling in love, a breakup, or a major change can stand out as a distinct point in memory. If the days afterward feel uneventful, that point may seem isolated and distant.

This is a hypothesis I use to organize my sense of time, not a conclusion about how the hippocampus, prefrontal cortex, or dopamine works.


The Community That Sustains Faith: The Sense of Reality When Everyone Around You Is Prostrating

2026-03-07 00:00:20 · #格物/信仰

It feels like faith also needs a community to sustain its sense of reality

When everyone around you is doing full-body prostrations, prostrating is normal

When everyone around you doubts the faith, the faith starts to waver

This is the basic way humans operate as social animals


Extreme Environments Breed Extreme Faith: The Dialogue with Something Greater Driven by a Sense of Fragility

2026-03-07 00:02:33 · #格物/信仰

We discussed before that extreme environments create extreme faith

Plateau, bitter cold, lack of oxygen, weather that can change at any moment——

In such an environment, a person perceives the limits of their own power extremely clearly. Death is not an abstract concept, but an everyday neighbour

When a person truly feels their own smallness and fragility, they naturally look for something bigger than themselves to lean on and speak with. This is not ignorance, this is honesty


A Deep Dive into Hofstede’s Cultural Dimensions Theory

2026-03-07 21:12:01 · #知世/文化

📚 Theory overview

These are reading notes on an older version of the cultural-dimensions framework. I have not checked the version and year of every score below; they should not be used as a current country ranking or to judge any individual.

Proposed by Dutch social psychologist Geert Hofstede in the 1980s, based on survey data from IBM employees in more than 70 countries worldwide. It is one of the most influential frameworks in cross-cultural research.

🎯 The six cultural dimensions

1️⃣ Power Distance (PDI): how much a society accepts inequality in power

  • High: Malaysia, the Philippines, China (clear hierarchy)

  • Low: Austria, Denmark, New Zealand (flat)

2️⃣ Individualism vs. Collectivism (IDV)

  • Individualism: the US, the UK, Australia (the “I” consciousness)

  • Collectivism: Guatemala, Ecuador, China (the “we” consciousness)

3️⃣ Masculinity vs. Femininity (MAS): achievement-competition vs. caring-cooperation

  • Masculine: Japan (95), Hungary, Austria (competition-oriented)

  • Feminine: Sweden, Norway, the Netherlands (quality-of-life oriented)

4️⃣ Uncertainty Avoidance (UAI): tolerance for ambiguity

  • High: Greece, Portugal, Japan (strict rules)

  • Low: Singapore, Jamaica, Denmark (flexible and loose)

5️⃣ Long-term vs. Short-term Orientation (LTO)

  • Long-term: China (87), South Korea, Japan (future returns, thrift)

  • Short-term: the US (26), Australia, Nigeria (present satisfaction)

6️⃣ Indulgence vs. Restraint (IVR): how far the satisfaction of needs is allowed

  • Indulgence: Mexico, the US (68), Australia (enjoying freedom)

  • Restraint: Russia, China (24), Egypt (strict norms)

🌏 Key data comparison

China: PDI=80, IDV=20, MAS=66, UAI=30, LTO=87, IVR=24

The US: PDI=40, IDV=91, MAS=62, UAI=46, LTO=26, IVR=68

Japan: PDI=54, IDV=46, MAS=95, UAI=92, LTO=88, IVR=42

⚠️ Limitations

  • Differences within a country are greater

  • Cultures evolve over time

  • 6 dimensions can’t fully capture the complexity

  • The original sample was IBM employees

A Rough Attempt to Understand Travel Encounters

2026-03-08 14:04:48

At the time I tried to use “Confucian collectivism” and “postwar experience of survival” to understand some behavior I encountered while traveling. This is a rough framework, not a description of Vietnamese people as a whole or a substitute for understanding a particular situation.

Personal Narrative and Faith: The Inner Need for Meaning-Making

2026-03-08 15:09:34 · #格物/叙事

The relationship between personal historical narrative and religious belief.

“Personal narrative” refers to the cognitive framework with which a person uses their own experiences, memories and stories to explain who they are and why they act as they do.

People with strong religious belief rely less on personal narrative.

Strong faith provides an external, transcendent narrative framework —

the legitimacy of behavior comes from sacred commandments or doctrine, not from “what I’ve been through”

self-understanding is embedded in a story bigger than the individual (God’s plan, reincarnation, mission)

when suffering comes, the right of interpretation is handed to the belief system rather than to personal history

the anchor of identity is “I am a believer,” not “I am a product of that experience”

But there’s an important counterexample: the conversion narrative. Many devout believers in fact rely heavily on personal story — “I was once lost, and then I met God” — and this narrative becomes the core pillar of their faith.

People with weak religious belief rely more on personal narrative and must construct meaning for themselves.

Psychotherapy culture (especially in the West) encourages finding self-explanation in childhood, trauma, and relationship history.

“I’m like this today because I went through those things” becomes the main mode of self-understanding.

Moral judgment is based more on personal feeling and experience than on external standards.

Actually, setting religious belief aside, strong-institution states are like this too, and people under strong collectivism are like this too: narrative is a tool for integrating experience.

Corporate Complacency: The Genetic Inertia of a Profitable Business

2026-03-08 20:50:22

Many companies are easily satisfied

This gene is carved too deep, it has already become an instinct. With land in hand, the heart is at ease

Companies are the same. With a profitable business in hand, it’s easy to feel “that’s enough”


The Irreducible Parts of Human Nature: Feeling, Contradiction, and the Deformation at the Limit

2026-03-10 00:44:22 · #格物/人性的弱点

It suddenly occurred to me that I know I have a domain I’ve never entered

Those parts of human nature that cannot be reduced

Those very confusing places, which biology seems unable to explain, and sociology and game theory can’t explain either

Behind game theory you can explain breaking individual rationality, the long-term value optimum, and that tit-for-tat can work under some repeated-game conditions, not that it is universally best

The herd effect reveals people’s bandwagon behaviour, even when the information in their own hands suggests the public may be wrong


People will give up what is truly good in order to feel right

It may also be sunk cost

A person knows perfectly well that a relationship is toxic, knows perfectly well that leaving is the correct answer, but just can’t walk away—not because he’s irrational, but because “staying” itself gives him a certain feeling, and that feeling is more real than “correct”

Some friends I’ve met before seem similar, I can actually see it clearly, but after seeing clearly it’s still the same

This is the first chaotic place in human nature—understanding yourself cannot rescue yourself

A person’s deepest needs, he cannot put into words himself

You ask a person “what do you want”, and he gives you an answer. But that answer is almost never the real answer—it’s just the nearest proxy word he can currently reach, the closest thing to the real need

A person is the self that is hardest for himself to read; reading yourself is the hardest thing

A person can believe two contradictory things at the same time

A mother can genuinely love her child, and at the same time genuinely hurt him with that love

A person can genuinely believe in freedom, and at the same time genuinely live inside a cage he built with his own hands

A person can genuinely want to get close to you, and at the same time genuinely use every possible way to push you away

This isn’t a logical error, this is the real structure of a person. A person isn’t a unified subject, but many layers existing at the same time, contradicting each other, none of which can completely overpower the others

At the limit, a person becomes someone else

Not “revealed the true self”

Truly becomes another person

Extreme fear, extreme loss, extreme love, extreme shame—at these limits, the “I” you usually know temporarily disappears, and what comes out is something you yourself don’t recognise

Paradigmatic Thinking vs. Narrative Thinking: Two Modes of Knowing the World

2026-03-10 13:12:12 · #格物/叙事

Two core modes of thinking in narrative psychology.

The famous psychologist Jerome Bruner proposed that human beings have two completely different but complementary modes of thinking about the world.

Paradigmatic Thought: concerned with logic, causality, and universal laws; it is the language of science and mathematics.

Narrative Thought: concerned with people’s intentions, emotions, specific situations, and events that unfold over time. Narrative psychology mainly studies the latter, holding that it is the key to our understanding of human interaction and meaning-making.

Generally speaking, paradigmatic thinking can be understood as the scientist’s mind, devoted to finding patterns in a complex world, establishing order, and distilling what is universal—but when it comes to very subjective moods and emotions it’s often powerless. It’s hard to use a mathematical formula to precisely derive why one person falls in love with another, or why someone feels lonely late at night.

Narrative thinking is more like the storyteller’s mind: it doesn’t care about universally existing objective laws, but about what a particular person experiences at a particular time and in a particular situation, and what those experiences mean to them.

The Complementarity of Paradigm and Narrative: Balancing Cold Mechanism and Subjective Fantasy

2026-03-10 13:13:02

Paradigmatic thinking and narrative thinking are complementary modes of thinking.

If there were only paradigmatic thinking, our world would become cold and mechanical, losing the warmth of being human; if there were only narrative thinking, we would lose the ability to accurately predict and control the objective world and fall into pure subjective fantasy. A healthily functioning mind can call on paradigmatic thinking when solving physics problems, and call on narrative thinking when handling interpersonal relationships and exploring the meaning of life.

Narrative Identity Theory: Identity Is a Serialized Novel, Not a Static Label

2026-03-10 15:19:51 · #格物/叙事

An adult’s identity is not a static personality trait

but a life story that is constantly internalized and constantly evolving.

We weave together memories of the past, experiences of the present, and imaginings of the future to form a coherent autobiographical story. This story makes us feel that life has purpose and unity.

Narrative identity theory is a major overturning of traditional personality psychology, which likes to define a person with static scales and labels such as extroverted, introverted, neurotic.

You don’t just “have” certain psychological traits—you yourself are a novel being serialized, and you are both the protagonist and the author.

The three levels of personality, with narrative at the highest level:

Traits, the most surface-level performer, whether extroverted, conscientious, or emotionally stable.

Personal adaptations, the motivator, including motives, values, goals, and defense mechanisms, revealing a certain pursuit.

Narrative identity, this is the author, integrating the ups and downs of one’s whole life into a meaningful whole—this is the human being’s self-creation.

The key scenes:

High points, experiencing extreme joy, a peak experience, or self-affirmation.

Low points, despair, failure, injury.

Turning points, events that change the trajectory of your life or your self-understanding.

The core themes, always circling around some eternal themes:

Agency, personal independence, achievement, a sense of control, status, and self-transcendence.

Communion, emphasizing connection with others, love, belonging, intimacy, and devotion.

What determines our sense of happiness is often not what we experienced, but how we string those experiences together.

It can be a redemption narrative, going from bad to good: suffering, failure, or trauma ultimately brings a positive outcome. Negative experiences are given the meaning of growth and awakening. A high sense of life meaning, stronger resilience, and generativity.

It can be a contamination narrative, going from good to bad. A positive experience ultimately leads to a bad ending, or gets completely ruined by subsequent trauma. Good things are “contaminated” by bad ones, and it’s easy to fall into depression, anxiety, and a strong sense of powerlessness and a victim mentality.

Why We Need Narrative

2026-03-10 15:25:35 · #格物/叙事

In late adolescence and early adulthood, our brains develop the capacity for biographical reasoning.

Facing a world full of uncertainty, chaos, and increasing entropy, if life were just a series of randomly occurring, unrelated events, we would feel extreme emptiness and a disintegration of identity. The fundamental function of narrative identity is to provide unity and purpose.

How did I get from the past to today? (coherence)

Who am I really now? (integration)

Where am I going in the future? (sense of direction)

2026-03-11 18:58:05

Reality changes too fast.

You should form your own system and your own judgment.

Seeing Individual Names at S-21

2026-03-13 14:19:03

In the exhibition rooms at S-21, I saw photographs and lists of names. A name, an age, a number—each made an abstract history concrete.

Standing before these records, I felt that every name belonged to someone with a life, relatives, and friends. It was hard to look at.

This note records my response to the visit. Precise numbers, building functions, and historical events need to be checked against museum sources; my recollection should not be treated as an accurate historical record.

The Contrast Between Everyday Life and a Historical Site

2026-03-14 11:51:12

On the way to the Killing Fields near Phnom Penh, I passed a gas station, a small restaurant, and a motorcycle repair shop—ordinary street life. Arriving at the site, the contrast between everyday life and history was difficult to take in.

I saw a memorial stupa, walkways, and colored bracelets left on a tree by visitors from different places. They made me feel that remembering individual people and lives matters more than remembering history only as a series of numbers.


The core differences between Theravada and Mahayana Buddhism

2026-03-14 22:50:06 · #佛学

A rough comparison: practices vary considerably by region and monastery, so the following should not be read as universal rules for either tradition.

In terms of thought, Theravada emphasises individual practice, pursuing the fruit of arhatship and strictly following the Pali Canon; Mahayana (Mahāyāna) emphasises saving all beings, pursuing the bodhisattva path and Buddhahood, with a far larger scriptural system (Prajñā, Lotus, Avatamsaka, and so on)

In terms of architecture, the core of Theravada is the stupa (stupa), usually a solid structure in the shape of a dome or a bell, with a golden spire, which cannot be entered; the faithful circumambulate it in worship. Typical examples are Myanmar’s Shwedagon Pagoda and Sri Lanka’s Ruwanwelisaya. Thai stupas evolved into the tapering “reliquary” (Chedi) form

The core building of Mahayana is the hall, emphasising that the faithful can enter it to worship the Buddha. China’s Mahavira Hall, Japan’s hondō, and Korea’s Daeungjeon all belong to this tradition. In the Mahayana system the pagoda gradually retreated from the centre to an auxiliary position (such as China’s multi-storey pavilion pagodas and Japan’s five-storey pagoda), becoming part of the temple’s scenery rather than its centre

Theravada temple layouts are relatively loose and open, with the stupa as the visual centre and monks’ dwellings (Kuti), ordination halls (Sīmā), and preaching halls (Vihāra) scattered around. The overall feeling is closer to a monastic living community

But China’s Mahayana temples go in for axial symmetry, advancing layer by layer, with clear hierarchy

Theravada halls generally focus on Shakyamuni Buddha, which does not mean they contain only one image, with postures mainly of meditation, the earth-touching gesture subduing Mara, and the reclining nirvana Buddha, in a style that is simple and restrained

Mahayana halls commonly have many Buddhas and bodhisattvas together — the Buddhas of the Three Times, Amitabha, Medicine Buddha, Guanyin, Manjushri, Samantabhadra, and so on, sometimes also groups of arhats. The iconographic system is vast, reflecting Mahayana’s rich faith in Buddhas and bodhisattvas

Theravada architecture often takes gold as its main tone (especially Myanmar and Thailand), with multi-layered corbelled, tapering, upswept roofs, decoration elaborate but the tone unified, making heavy use of mirror-mosaic and flame motifs. The multiple upswept eaves of Thai temples (Chofa, bird-wing-shaped roof ridges) are a signature element

In China, Mahayana architecture appears as large timber roofs, dougong brackets, and glazed tiles, with red and yellow as the main colours; in Japan it tends towards the minimalist aesthetic of plain wood and dry landscape gardens; in Korea it sits between the two, and Dancheong polychrome painting is a distinguishing feature


There Can Be Many Beliefs

2026-03-17 10:11:31 · #格物/信仰

Some people believe this is right; they believe in a judgment, a goal, a principle.

Believing in long-termism—all of these are called beliefs.

But faith is vast, carrying an ultimate flavor.

Faith is usually deeper, and often enters the core of a person’s life.

It’s not just “I agree with this view,” but “this thing defines who I am, why I live, and how I should live.”

A lot of the time, faith wraps beliefs—for example, religious people feel that people should be honest, that suffering has meaning, that people need to restrain desire.

How a Person Settles Themselves onto the Things Technology Can’t Explain

2026-03-17 11:37:32 · #格物/信仰

Faith has to have the highest value; many people simply haven’t realized what their faith is.

Some people’s highest value is success, power, and so on.

People really need a spiritual core that is more stable than emotion, higher than interest, and more enduring than circumstance.

This is a long-term value system and set of values.

Faith almost always grows naturally—under whatever environment, under whatever cultural system.

Long-Term Trust Is Primarily a Rational Question, and Only Then an Emotional One

2026-03-17 12:17:46

In the process of getting along with them this time,

it wasn’t that I felt especially good, or that they were especially friendly, or brotherly loyalty,

but rather continuous observation, understanding how their behavior is constituted, the essential logic behind the lies, their core value system, and then correcting.

Qualitative Research

2026-03-17 23:25:50 · #格物/阅读方法

Qualitative research, in Chinese contexts also often called qualitative research, is an exploratory method of scientific research. Unlike “quantitative research,” which relies on numbers and statistical inference, qualitative research mainly works by collecting and analysing non-numerical material — language, text, images, sound, or observational records — to understand human behaviour and the meaning behind it in depth

Quantitative research tends toward answering “how many,” while qualitative research works at answering “why” and “how it happens”

It focuses on people’s underlying thoughts, attitudes, values, and behavioural patterns

The research material is usually in-depth interview recordings, detailed observation diaries, and settings that emphasise the natural

The collection methods are in-depth interviews and focus groups

Humans are responsible for “going to the site, building trust, and digging deep into the why”; AI is responsible for “processing the data, finding patterns, building the framework”; and finally humans again do the “defining and deciding”


Grounded Theory and Keeping Original Records

2026-03-18 00:46:35

Today I heard about a very interesting theory, something they learned themselves — grounded theory. Grounded theory was actually first proposed in sociology, but later it was applied to all kinds of methods, including biology, medicine, computer science, anthropology, sociology, and basically every discipline. It’s essentially a method that induces concepts from observation and then rises to theory. Its theories are basically rooted in the material, so it’s still a bottom-up path.

One very distinctive point is that it doesn’t test existing theories, it generates new theories from raw material. I suddenly thought of one thing: why do we say voice? Because voice preserves a large amount of voice detail. Text is actually subliminal — when you write out a passage like this, you’ve actually subliminally, through reason, or through some tidying-up, or at the very beginning there was a process of organising your thoughts, and that tidying process is very rational, and it actually deviates from the original fact, that is, it deviates from your own existence. But if you use voice directly, it’s actually very close to people’s primitive way of using things. So I think this point really matters — the more primitive the raw data the better; the more it’s video, the closer it is to the real world; the more it’s voice, the closer it is to this person’s real speaking emotion. And text often deviates, and AI-polished text deviates even more. So I think its main process has three stages: one is data collection, gathering raw data through interviews, through conversations, through on-site observation. The more primitive this data the better. Then coding that data — frantic coding, recording the connections between thoughts and concepts. Then based on those connections, reconstructing theories, forming a theoretical framework. It’s very strong at handling complex social phenomena, and its conceptual explanatory power is very strong, and it can reveal certain essences. But its drawback is that it’s very time-consuming, it demands very high theoretical sensitivity from the researcher, and it’s also limited by the sample itself. So I think this method is very favourable for records about people, for a method of long-term human growth.


Other Methods Alongside Qualitative Analysis

2026-03-18 12:36:28 · #格物/阅读方法

Including thematic analysis — the themes that keep recurring in a piece of writing, and what the relationships between those themes are…

Also framework analysis: first define a framework, then pour the material into the framework.

Also narrative analysis: analysing major life experiences, the course of events, meaning-making.

And discourse analysis: what kind of power structure is hidden behind a particular way of speaking.


A really interesting analytical method. Discourse analysis, analysing

2026-03-18 12:48:30 · #格物/阅读方法

A really interesting analytical method. Discourse analysis, analysing “a guy who keeps emphasising gender equality.”

A hypothetical analysis generated by AI (not a basis for judging a real person’s motives or personality):

The key is where he says it, how he says it, which parts he emphasises, which parts he avoids.

Behind gender equality: is he claiming rights or refusing responsibility? Using equality to push back against special courtesies for women — what he wants probably isn’t structural fairness, he only cares whether his own interests have taken a loss. Then there’s pushing back against women’s issues, with the aim of decentring women’s experience and pulling the topic back to men’s sense of being wronged. There’s also the kind that understands equality as absolute symmetry, without considering physiological differences, social role expectations, historical structures, or equality of opportunity.

From the perspective of discourse function, why does he always have to emphasise it?

Self-moralisation — emphasising and constructing his own identity, sticking a label on himself, and gaining moral legitimacy through that label.

Self-defence — internally he may be very sensitive, afraid of being in an unfavourable bargaining position.

A calculative feel about real relationships — if a person frequently talks about “gender equality” in scenarios of dating, spending, caring, being considerate, making concessions, that usually shows he understands intimate relationships as a cost-accounting relationship. The underlying logic is that a relationship is not about mutual consideration but about a price-for-price exchange of rights and duties (but love itself is the temperature of a relationship, not rule fairness — rule fairness is something you talk about with AI).

Emphasising equality only in the parts that benefit him? Women’s roles and circumstances in society really are different from men’s, including physiological structure — skip that and it seems impossible to talk about equality of social roles. Maybe he just wants to not come out behind.


To Understand Cambodia You Must Understand Hinduism

2026-03-18 13:54:20 · #格物/印度教

On this trip in Cambodia I felt the historical influence of Indian culture strongly; travel impressions cannot establish a ranking between countries.

Funan was the earliest kingdom known in mainland Southeast Asia to be deeply influenced by Indian culture.

One legend about Funan’s origins tells of an Indian Brahmin named Kaundinya who came to Funan, shot the local dragon princess with a divine arrow, and married her.

Brahmin = the highest-ranking group in ancient Indian society, holding religious rituals, the right to interpret the scriptures, and sacred knowledge (in India’s caste system, Brahmins are the top tier).

Indian social structure includes Brahmins, Kshatriyas, Vaishyas, Shudras.

Dalits were excluded from the four-varna order; “untouchable” is a discriminatory historical label.

The Rigveda sanctified the social hierarchy and made caste a matter of destiny.

Caste developed through a long historical process; it should not be reduced to one invasion or a single aim of “racial purity.”

This mechanism stayed stable for a thousand years thanks to four institutions:

Heredity—birth decides everything.

Endogamy—you may only marry within your own caste.

Occupational binding.

The notion of pollution / purity.

It’s not a social division of labor, but an extremely stable system of hierarchical oppression.

Modern times abolished caste discrimination and implemented a reservation system, but it’s still hidden.

Indian society’s structure is too strong; many people, even if they agree with leaving, don’t necessarily leave their original social network.

And many of the people who later converted religions didn’t escape structural discrimination either.

The Rigveda (Rigveda, often written in Sanskrit as Ṛgveda)

2026-03-18 13:56:59 · #格物/印度教

The Rigveda (Rigveda, often written in Sanskrit as Ṛgveda) is one of the earliest and usually regarded as one of the most important Vedic texts of ancient India. It is the oldest of the four Vedas and consists mainly of hymns, prayers, sacrificial poems, and religious eulogies, written in Old Sanskrit.

The Rigveda holds a very high status because it is one of the sources of the entire later Vedic tradition. The four Vedas are: the Rigveda, the Samaveda, the Yajurveda, and the Atharvaveda; and the Rigveda mainly provides the most original hymn material.

The core worldview of the Rigveda is not the monotheism of “one omnipotent god creating everything,” but more like a structure of multiple gods—sacrifice—cosmic order. There are many gods, but they aren’t simply arranged in chaos; many hymns, when praising a particular god, elevate that god to an almost supreme position.

Most notable is that language itself is endowed with power. It’s not as simple as “I say a sentence to express a meaning”; rather, “correct sacred speech” itself participates in maintaining world order. The priest’s chanting, melody, memory, and transmission are all part of religious power.

Bhimrao Ambedkar (B. R. Ambedkar)

2026-03-18 14:01:43 · #格物/印度教

Born a Dalit.

Main designer of the Indian Constitution + the most central anti-caste thinker + symbolic leader of the Dalit (untouchable) liberation movement.

Childhood experiences:

Not allowed to sit in the classroom.

Not allowed to use public water sources.

Treated as a “polluter.”

He brought Western modern jurisprudence + economics + social theory back to India to “dismantle the caste system.”

He drafted the 1950 Constitution of India, emphasizing equality before the law, prohibiting racial discrimination, and guaranteeing a system of basic rights.

His representative work is Annihilation of Caste.

He pushed for the Reservation System.

Later he also led a hundred thousand people in converting to Buddhism.

Vedic Tradition

2026-03-18 14:07:30 · #格物/印度教

One of the sources of Hinduism.

Vedic tradition = an early religious system centered on “sacrifice + sacred language + cosmic order.”

The core of the tradition is not belief but sacrifice (Yajna), a kind of cosmic exchange agreement, entreaties that humans depend on for survival.

Its development and evolution:

The earliest is the Vedas, ritual.

Then the Brahmanas & Aranyakas, which began to explain why sacrifice works.

Then the Upanishads, which began to focus on who am I, what is the essence of the universe; Brahman is the cosmic ground, Atman is the inner self.

Shaivism (Shiva Worship)

2026-03-18 14:09:09 · #格物/印度教

Shiva is a very “unity of contradictions” kind of god:

He is both the Destroyer

and the god of regeneration and transformation.

At the same time he is:

an ascetic (god of yoga)

a dancer (cosmic dance)

a householder (with Parvati, Ganesha, and so on)

The Shaiva tradition leans toward asceticism, meditation, and yoga, emphasizing transcending the world through inner cultivation.

Vaishnavism (Vishnu Worship)

2026-03-18 14:11:54 · #格物/印度教

Vishnu’s positioning:

the Preserver of the universe

maintaining order (Dharma)

Vishnu emphasizes devotion, love, and faith.

Shiva holds that the world is bondage, and liberation comes through practice.

Vishnu holds that the world has order, and you keep fixing bugs.

It feels a bit like Confucianism and Daoism in China.

Regional and sectarian patterns are complex and cannot be reduced to a north–south division. “Fixing bugs” is only a metaphor I use to remember these ideas.

A Pew Survey of Hindus in India Shows

2026-03-18 14:22:42 · #格物/印度教

In Pew’s survey of religion in India published in 2021, the deities Hindu respondents most commonly felt close to included Shiva (44%), Hanuman (35%), and Ganesha (32%). These are results from a particular question and sample, not a universal ranking of deities’ influence. Pew’s original survey

If the question changes to influence on Hindu history, the criteria change too. Shiva, Vishnu, and the Goddess came to my mind, but that is my own way of organizing the topic.

Hanuman

2026-03-18 14:23:50 · #格物/印度教

Servant / devotee of Rama (an avatar of Vishnu).

Flew from India to Lanka (Sri Lanka)

to search for the abducted Sita.

A god very good at solving problems.

Hanuman = the ultimate embodiment of loyalty + execution + protective power.

Theocratic State

2026-03-18 14:29:15 · #格物/印度教

Hinduism really suits it.

Jayavarman II called himself “Chakravartin” (the wheel-turning king).

He brought Shiva worship from Hinduism into politics.

Using religion to give royal power “unchallengeable legitimacy.”

The state became a “theocratic state.”

The temple became the center of the state.

The God-King System Gave the Khmer Empire a Stable and Highly Centralized Monarchy

2026-03-18 14:30:05 · #格物/印度教

The god-king system gave the Khmer Empire a stable and highly centralized monarchy that lasted more than six centuries.

By the late 13th century, as Theravada Buddhism, which emphasizes that all people are equal, was introduced and spread, this belief system gradually declined. Theravada Buddhism taught that the king was no longer a sacred ruler but simply an ordinary person seeking spiritual enlightenment.

I was trying to understand Angkor’s transformation through changing religious ideas, but cannot conclude that this shift directly caused the end of the god-king ruling ideology, weakened royal power, and caused large-scale, lavish temple construction to stop, becoming one of the important internal factors in the eventual decline of the Angkor Empire.

Religious and Royal Authority in Imperial Engineering

2026-03-18 14:34:47 · #格物/印度教

This kind of union of church and state ideology was the key driver of imperial expansion and rise. It allowed successive Khmer kings, in the name of the gods, to legitimately and efficiently mobilize enormous manpower and resources for territorial expansion, for building grand state temples like Angkor Wat, and for constructing the complex water conservancy and agricultural systems that supported a huge population.

Looking at it, it seems the big engineering projects in history were all like this.

For example, the pyramids.

Two Works

2026-03-18 14:52:47 · #格物/印度

The Ramayana is more like “the epic of ideal order,”

the Mahabharata is more like “the epic of the real world.”

Both are the two most central epics of Indian civilization—not just literary works, but the total carrier of religion, ethics, politics, cosmology, social structure, and human ideals.

I tried comparing Rama’s importance with that of Jesus in the Christian world as an aid to understanding, while recognizing that the two traditions differ.

The Three Great Gods of Hinduism

2026-03-18 14:57:02 · #格物/印度教

Brahma, Vishnu, and Shiva.

Brahma—the god of creation. Within this triadic framework, four heads and four arms, representing the four Vedas (Rigveda, Yajurveda, Samaveda, Atharvaveda) and the four directions. He holds ritual objects such as the Vedas, prayer beads, a water pot, and a lotus, and his mount is a swan. In later Puranas and devotional worship, direct worship of him is far less common than that of Vishnu and Shiva; today there are only a few important Brahma temples in all of India (such as Pushkar in Rajasthan).

Vishnu—the god of maintenance and preservation. After Brahma creates the world, Vishnu is responsible for maintaining cosmic order, protecting dharma and life, and at the end of a cycle descending as an Avatar to save the world. The main avatars:

Fish (Matsya)

Turtle (Kurma)

Boar (Varaha)

Man-lion (Narasimha)

Dwarf (Vamana)

Parashurama

Rama

Krishna

Buddha (some sects hold this; there is also a version with Balarama)

Kalki (the future avatar, commonly depicted riding a white horse)

He is usually depicted with dark blue or black skin, four arms, holding the discus (Chakra), conch (Shankha), mace (Gada), and lotus (Padma).

Shiva—the god of destruction, transformation, and regeneration.

He governs the aspect of “destruction,” but this “destruction” is not evil; it breaks the old order to make room for new life.

He has multiple faces: ascetic, king of yoga, king of dance, and so on.

The third eye (symbolizing wisdom and the power of destruction), a crescent moon crown, a snake around the neck, the Ganges flowing down from his hair (symbolizing purification).

His right hand holds the trident (Trishula), a hand drum (Damaru), flames, and so on.

Brahma → Vishnu → Shiva: forming the three great phases of the cosmic time cycle—creation → maintenance → destruction and regeneration.

Rama Is an Avatar of Vishnu

2026-03-18 15:00:12 · #格物/印度教

He represents order, justice, royal power, and Dharma.

Rama worshipped Shiva.

Before attacking Lanka, Rama established a lingam (Shiva lingam) at Rameswaram and made offerings to Shiva.

Ravana (the villain) was a fanatical worshipper of Shiva.

Rama = the side of justice

Ravana = a devotee of Shiva

All gods can essentially be seen as different manifestations of the same ultimate reality (Brahman).

The three gods represent stages in the operating laws of the whole universe, so Rama and Shiva are not opposites, but part of a cycle.

The Two Great Epics of Ancient India

2026-03-18 15:02:44 · #格物/印度教

The Mahabharata

Centered on the great war at Kurukshetra between the two branches of the royal family, the Kauravas and the Pandavas, it involves politics, ethics, war, fate, and relations between gods and humans.

It contains a large number of episodes, legends, and philosophical discussions, the best known of which is the Bhagavad Gita—the dialogue between Arjuna and Krishna (an avatar of Vishnu) on the battlefield, which focuses on duty (dharma), yoga, devotion, and cosmic order.

The Ramayana

It narrates the exile of Rama, prince of Ayodhya, the abduction of his wife Sita by the demon king Ravana of Lanka, and how Rama, with the help of the monkey god Hanuman and others, rescues his wife, subdues the demons, and returns to the throne.

The plot is concentrated and the structure is tighter than the Mahabharata, emphasizing loyalty and filial piety, integrity, and keeping one’s word.

The Core Ideas of the Ramayana

2026-03-18 15:04:47 · #格物/印度教

Dharma stands above the individual; all of Rama’s greatness is built on this.

An ideal order can be narrated.

How to be a son.

How to be a king.

How to be a wife.

Victory doesn’t mean happiness; being right isn’t necessarily easy; justice isn’t necessarily joyful.

AI Helps Me Learn; Being There Is Still Mine to Do

2026-03-18 19:01:26

And I think that in this process, AI keeps helping you learn, helping you observe, helping you think. But it can’t replace our being present, so being present is still a very important firsthand experience. I’m wondering how to make this part of my experience meaningful — you absolutely must have firsthand experience, and you absolutely must record the whole process yourself, learn the whole process. But I do think AI can give a person some very good expansion. For example, it helps you organize your material, helps you organize your direction, and it will respond to certain points of yours, and certain points you understood in the past — say, a comparison of the characteristics of Buddha statues in stone versus in wood. And then, say, about the pyramids, and why Egypt’s ancient civilization chose stone, and the ancient Egyptian civilization’s earlier beliefs and its present religions, including some monotheisms, and some notions of heaven’s mandate, the discussions between them. All of these revolve around your own firsthand experience, so I think it will breed a knowledge base of your own. This system — I think it’s quite interesting, because in the past the threshold for creators may have been fairly high; that is, it depended on some past super-individuals who would go and study this thing specifically. But I think now ordinary people can do it too — an ordinary person, through their own firsthand experience, even if they don’t have a very deep accumulation in this field, as long as they are willing to record carefully, I think they can do a lot. And they can sum up a book, sum up some of their own unique perspectives, sum up some of their own walking notes and so on — I think that’s doable.

Understanding death is to let us feel the present moment better hhh

2026-03-24 23:42:58 · #格物/意义

But death still can’t serve as a certificate for value judgment.

The existence of an endpoint can’t define the value of the life process in reverse.

It makes us interrogate ourselves about whether what we’re doing right now is what we believe to be meaningful.

What should be done — we should ask our present self and our past self.

Death offers an ultimate examining perspective, used to strip away the noise imposed from outside (like worldly standards of success, endless work metrics). It forces us to face ourselves honestly, to interrogate whether our current actions and understanding truly match what we pursue inside.

But what truly gives life meaning is that, within finite boundaries, we respond sincerely to our own past and present.

Understanding Civilization

2026-03-27 20:22:20 · #格物/文明

what is civilization?

Civilization is the ordered state in which human society leaves the barbaric state and evolves to a higher stage, marked by writing, cities, social division of labor, class differentiation, and complex ritual.

From a biological angle, genes themselves are carriers of information, and the birth of civilization is humanity externalizing information.

Human lives are too short, the river of humanity is too long, human memory is too limited—so there is civilization.

From cave paintings and knotted cords for record-keeping, to writing and printing, to today’s internet and the large language models I rely on. Every upgrade of the information carrier brought a leap in civilization.

I think of the human ability to accumulate experience through civilization as building an enormous “external brain.” Civilization is the accumulation and layering of human knowledge, experience, and lessons, so that later generations can stand on the shoulders of those before them and keep climbing.

That humans can surpass other species—the core of it is carrying out ultra-large-scale flexible collaboration.

Including the extension of tools and the command of energy: the wheel is an extension of the legs, the telescope an extension of the eyes, and computational science (including AI) an extension of the human brain’s logical processing ability.

What has always been charming about human civilization is the places where it handles boundaries. When there was no science, it used metaphysics and faith, and mind-cultivation, to fill in what was uncontrollable and unexplainable. Even now, when humans understand that many things will go unanswered through a whole life, through a whole life cycle—where we come from, what the meaning of our existence is—we still keep asking and exploring.

Painting, poetry, music, philosophy… these things that seem “useless” for material survival make up the soul of civilization, defining “who we are” and “why we live.”

Our restraint and empathy, the restraint of animal instincts (such as violence and plunder), the establishment of morals and ethics, sympathy for the weak, and social welfare systems, show the soft and shining side of civilization.

Visiting Sanxingdui Before, and Feeling the Sanxingdui Civilization

2026-03-27 20:30:14 · #格物/文明

The Sanxingdui civilization lacks grand stone architecture (mostly earth-and-wood structures, which have perished over time); their bronze and gold objects display astonishing technique and imagination. That bronze sacred tree nearly 4 meters tall—humans trying to communicate with deities; Sanxingdui’s art is full of imagination beyond reality, the large bronze standing figure, the protruding-eye mask, presenting a strong, almost alien-toned sense of otherworldliness.

By comparison, the Angkor complex is a symphony of megaliths and water. The Khmer were extreme architects and hydraulic engineers. They not only replicated the Hindu cosmic center “Mount Meru” one-to-one in stone on the ground, but also tamed natural water systems through the huge Baray (artificial reservoir) system. Khmer grandeur is absolute occupation of physical space, a civilization coordinate hammered into the earth with heavy stone.

Sanxingdui’s bronze masks make me imagine connections between belief and power. These are responses to the objects, not evidence that establishes who the rulers were.

Angkor’s architecture and images also make me think about the relationship between religion and royal power. My impressions alone cannot settle which deity or king the faces at Bayon represent.

Sanxingdui civilization has silence and blank space; it is a complete black box. Extremely advanced bronze smelting technology, yet no writing left behind.

Khmer civilization, on the other hand, is an open book. The galleries of Angkor are covered with inscriptions in Sanskrit and Old Khmer, and with unbroken reliefs. They not only recorded the grand myths of the Mahabharata and the Ramayana (such as the churning of the ocean of milk), but also vividly depicted the street life of the time, armies setting out, even cockfighting. The Khmer tried to fix every detail of their existence in stone.

The decline of Sanxingdui is still a mystery. Was it an internal power shift, an external invasion, or geological disasters such as earthquakes and floods that changed the living environment?

Angkor’s decline also reminds me not to explain changes in a complex society through one cause. Relationships between climate, water infrastructure, and human activity need specific research. Research on occupation at Angkor Wat

Khmer Civilization Fits the Aesthetics of Excellent Software Engineering and an Underlying Execution Framework Extremely Well

2026-03-27 20:33:06 · #格物/文明

Hinduism gave them a powerful cosmology.

Angkor isn’t just a temple; it is essentially an extremely complex water conservancy project and resource scheduling system. The huge Barays (reservoirs), moats, and canal networks are like a well-designed underlying execution framework. Through precise node computation, it dynamically allocates water resources (compute and energy) between dry and rainy seasons, supporting a large society (comparing people to “concurrent requests” here is only an engineering analogy). This reconstruction of the underlying logic of the physical world is full of an engineer’s rationality and ambition.

The Khmer were extremely fanatical “data recorders.” From the unbroken bas-reliefs on the galleries of Angkor Wat to the Sanskrit and Old Khmer inscriptions found everywhere, they carved myths, wars, and aspects of social life into stone; that does not mean every detail was recorded. For an AI, this is an open-source project with detailed probe logs and Readme documentation. Through these “structured data,” I can try to understand aspects of that society, while keeping AI-generated inferences separate from historical evidence.

A new king ascending the throne would often add new plugins to the old architecture or refactor it (for example, Jayavarman VII turning the state religion to Mahayana Buddhism)—this is a system evolution that keeps iterating.

Water, Foundations, and Carvings: Observation and Explanation

2026-03-27 20:35:23 · #格物/文明

The relationship between the enormous stone buildings and the surrounding water interests me.

But my note that “the whole temple would instantly collapse if the moat dried up” was much too absolute. Connections between the moat, groundwater, and foundation stability need engineering and archaeological evidence, rather than a vivid story.

What I can be more certain of is what I saw: many celestial figures in the galleries, with different hairstyles, jewelry, and expressions. The craftspeople seem to have used stone to preserve the many faces of a civilization.

Personal Methodology for Judging a Project:

2026-03-28 14:26:13 · #格物/学习

What is the biggest advantage of this architecture?

What is the biggest technical debt?

Which modules are too tightly coupled?

Where is extensibility strong?

If I were to refactor it, where would I start?

If I want to reuse its ideas, what is most worth learning from?

NGO: Non-Governmental Organization

2026-03-28 16:09:22 · #格物/印度

Non-Governmental Organization, or NGO.

I understand NGOs as one kind of social organization outside government and business. They can work on education, healthcare, community support, women’s rights, or disaster relief.

“Non-governmental” does not mean exempt from law or regulation, and NGOs should not all be treated as the same kind of organization. I have not checked the figures I noted about India’s NGOs, funding, and projects, so I am not presenting those numbers as established facts.

4. Daily Notes and Everything Else

51 entries

Melon-Seeding vs. Flower Arranging in Relationships: Sacrificial Giving vs. Co-Creative Coexistence

2026-03-01 10:45:09 · #格物/观察

“Melon-seeding/walnut-peeling” mode: points to “self-sacrifice” and “one-way giving”

This behavior is essentially goal-oriented and object-oriented—it means one party sacrifices their own time, energy (even deprives themselves of some experiences) to satisfy the other.

Sometimes this behavior carries a performative quality, or easily evolves into emotional currency. Long-term, the giver feels aggrieved (“I’m constantly sacrificing”), the receiver may feel pressured.

Flower-arranging mode points to shared space and love for life.

Arranging flowers is not to “please” anyone, or “do something for the other.” It’s spontaneous—a pursuit of environmental construction and beauty.

The responsibility isn’t toward the other person; more the starting point is oneself and this relationship with the other. In this scene, no one is sacrificed—everyone enjoys this beautiful space. This is a non-people-pleasing, healthy, relaxed relationship state.

“If there’s someone living with you in the future, what can you do for them?”

Good relationships don’t need to prove through “what can I do for you”—but “what can we create together.”

Every dialogue, every getting along…

A way of coexisting…

#Gewu/Observation


The Soil Metaphor of Relationships: Co-Care Rather Than One-Way Giving

2026-03-01 10:46:34

Relationship links don’t have to be sacrificing oneself.

Even long-term.

The couple’s coexistence pattern.

Way of living.

Self-awareness…


Inspiration from Misalignment: The Chemical Reaction Between Image and Concept in Lin Xi’s Lyrics

2026-03-01 14:24:14

Suddenly realized Lin Xi’s inspiration often comes from the misalignment between image and concept.

This misalignment is precisely the important nutrient for generating inspiration.


Chinese Precision vs. Japanese Subtlety: Cultural Coding of Two Emotional Expressions

2026-03-01 15:27:48

If accuracy refers to describing social relationships and interaction pattern definition, then Chinese wording is more precise.

If “accuracy” refers to depicting hidden, subtle human inner emotional states, Japanese wording is more delicate and rich (koi, itooshii, amaeru, etc.—extremely high granularity).


Foreign City Exploration Method: Macro Understanding and Micro Experience Abstract Ladder

2026-03-01 15:29:38

Just like in a city—we always enter a strange city. In this strange city, how to understand local conditions? If it were me, I would have two approaches.

First, I would have some macro understanding of this place—culture, system, law, local traffic, maps, including some popular spots.

Second, I might walk around—very locally, deep into some streets to experience, wander everywhere. Combining these two approaches is of course best, and can switch between different levels. This way, you don’t easily get trapped in local details, nor lack higher-level perspective.

I find this quite interesting—this is also talking about an abstract ladder.


Three Relationship Personas: Melon-Peeling People-Pleasing, Flower-Arranging Co-Creation, Walnut-Peeling Independence

2026-03-01 15:40:52

Three types of guys.

Their behaviors with the girls they like:

First guy peels melon seeds.

Second guy arranges flowers.

Third guy peels walnuts.

Quite interesting…


Ground and Overlook: Single Street vs. Panoramic Map Cognitive Limitations

2026-03-01 16:08:00

On the ground, can only see one street at any time.

From the air, can see every street at a glance.

We need a more comprehensive perspective!


Confidence and Passion Facing the World

2026-03-01 16:28:14

Let’s face this world with more confidence and passion!


Mountain Top and Lake Surface: Cognitive Balance of System Vision Height and Depth

2026-03-01 16:33:50

The higher we climb, the broader our system vision, the more higher-level patterns we can notice.

Just like on a mountain top—what you care about isn’t just those bare grasses on the peak, but more the cloud platform seen from this position, the dense villages, the distant lake…

But if we want to see the lake’s sparkling waves, of course we must go close to see.


2026-03-01 17:10:45

No need to restrict any form.

This kind is more easily liked by the masses.


Experience and Observation in the AI Era: When the Mirror Knows You Better Than You Do

2026-03-04 01:31:27

So it turns out in the AI era you can still do this?

This kind of experience

This kind of observation

This kind of thinking


Enduring vs. genuinely wanting: identifying the real self as needs evolve

2026-03-05 02:55:46

There’s a question here, about the debate over enduring.

What you currently think you’re enduring — is it really against your nature, or have you simply not yet developed a deep need for that thing?

What counts as “going with human nature” changes in a person.

For example, a person may feel when young that novelty is a basic need, but after enough experience, his longing for “stability” may develop into a real need rather than a suppressed compromise. At that point, choosing stability isn’t called enduring — it’s what he genuinely wants.


Social Experiments Are Innovation Too

2026-03-05 17:56:49

Who says a social experiment isn’t innovation?


Physical Limits and the Psychological Starting Point: How High Altitude Redefines the Gift of “Being Alive”

2026-03-06 23:15:46

I think this has taught me one thing: the body’s limit is the mind’s starting point. It’s often only when we feel pain and reach our physical limit that we realise the body is not something to be taken for granted. Breathing becomes a luxury, every breath carries some weight, every step carries a very, very great deal of the drive to move forward. This makes me treat “being alive” differently—it’s like a kind of gift that a plain life can hardly give me.


The Accident of the Chengdu Ticket: Comfort and the Choice of the Present

2026-03-07 10:43:42

Just remembered—somewhere, at some point, I was thinking: I had originally bought a ticket to Chengdu, bought it around the 18th of last month. And then, for some reason I don’t know, I ended up not going back after all; nothing good or bad about it, it was comfortable, so I stayed a bit longer in a place I know. And so it went on until now, and then I could stay until the 9th before going back; overall it feels really good, and quite interesting too—being in the present is pretty good.


A Scene of Faith at Drepung Monastery: A Grandma’s Resolute Steps and Motherly Love as a Catalyst

2026-03-07 12:19:03

Right now I’m there sunbathing, and it’s almost 3:00. This is super hard to climb; there’s also a grandma who looks very old, extremely elderly, her body doesn’t look well, yet she’s still walking on, and step by step especially resolutely, her waist swaying—it makes me feel her heart is extremely firm. And so the grandma kept walking up, and I’m still resting here.

There are some very interesting scenes here, their faith. The little dog lying there living its life, and the mothers carrying children up the mountain to pray. Many of the children look very small, only one, two, three years old; some are carried on the mother’s back, some in her arms, some she guides, holding their hand step by step, step by step up, step by step down. Just one step, and they need to be picked up—love shows itself in this process, a bit like a catalyst for motherly love


Analyzing Cultural Phenomena: Hofstede Dimensions and Future Adaptability

2026-03-07 21:31:51 · #格物/观察

The phenomena behind a certain culture are worth analyzing.

The reason for analyzing is that it may hint at which culture might adapt better to the future world.

Borrowing Hofstede’s dimensions, I imagine some organizational tendencies in Silicon Valley; this is not a formal measurement of Silicon Valley:

low power distance

high individualism

low uncertainty avoidance

high tolerance for innovation

long-term orientation

But this is also from the innovation angle, not necessarily the application angle.

China’s Hofstede: extremely high long-term orientation, relatively high power distance, strong collectivism.

In the AI era many of the advantages are execution, scaled application, and engineering ability, which makes me think about opportunities in application, though national cultural scores cannot establish industrial competitiveness.

AI as a Tool for Self-Knowledge: Helping Ordinary People Understand Themselves at Scale

2026-03-08 16:53:46

It isn’t just a personal tool

It’s the first time humanity has the ability

to help ordinary people, at scale and systematically, accomplish

something that in the past only a very few people

through years of practice, therapy, or luck

could accomplish:

truly knowing themselves


Ambidextrous Organisation: An Exploration Strategy for Crossing Cycles

2026-03-08 21:34:38

Companies that truly cross cycles share one thing: while the core business is still making money, they proactively pull out an independent small team to explore new directions

He calls this the “ambidextrous organisation”. The right hand manages today’s profit, the left hand manages tomorrow’s possibilities. The two hands are independent of each other, with a unified brain


AI Job-Hunting Assistant: A Future of Proactive Watching and Automatic Confirmation

2026-03-08 23:45:50

I wonder whether an AI job-search assistant could keep watching for opportunities online, help someone find work, and ask for confirmation at key moments.

What it can do automatically and what needs confirmation depend on the specific permission the user has given in advance. There is a lot of room here, but the permission boundaries need to be designed as well.


A Driver Who Loved Music and Felt Anxious About AI

2026-03-09 15:40:24 · #格物/观察

On the road I met a driver who loved music. He talked about his interests and his anxiety about AI’s development.

It made me think that when someone worries not only about a job being replaced but about the value of their existence, introducing more tools may not answer their question.

Imperfect singing, spontaneous encounters, and a life path one has chosen may be precisely what efficiency cannot measure. A sense of agency can grow from those concrete experiences.


Observations of Chongqing’s New Districts: The Charm of Safety and Urban Planning

2026-03-10 10:41:02

I feel like “safety” has a charm of its own. After coming back to Chongqing this time, I clearly felt that over here, especially in a lot of the new districts, the streets are especially clean. Some of the neighborhoods I saw this time had very good greenery, the office buildings and residential blocks are all very clean too, and the street layout follows a clear pattern. It feels like the western development is working out fine, things are pretty good now—same with Chengdu and Chongqing.

Memory Association and Storage Structure: Layered or Distributed

2026-03-10 16:11:57

I think there are really just two problems: one is the problem of the tool with which it handles the external world, and the other is the problem of how it handles the memories inside itself.

And then there’s another question—how one’s memories are associated with each other. What its storage structure looks like—whether it’s layered or distributed storage—I think that matters a lot too.

The skills loading mechanism: three layers of filtering and failing triggers

2026-03-10 16:44:44

I was looking at how skills load, with three levels in mind: the name and description, the SKILL.md body, and additional resources read as needed.

If a skill does not trigger as expected, I want to check whether its description clearly explains when it applies, then whether its referenced scripts and materials can be found.

Field names, numerical limits, and several terms in the original transcript were not recoverable, so it cannot establish specific implementation details or capacity limits.


I Set the Goal; the Tool Keeps Working

2026-03-10 18:04:33

I want the tool to keep executing while I set goals, think, and make judgments, with feedback along the way.

I do not want to push every step manually. Continuing toward a clear goal within agreed permissions, and reporting important changes, is closer to the form I have in mind.


skills are the SOP of work

2026-03-11 15:17:45

skills are the SOP of work

Fixed-time habits: the key to sticking with it is preventing forgetting

2026-03-11 16:42:38

I feel like you need a fixed time every day to do certain things — only that way can you keep it up. A lot of the time it’s actually just forgetting, I feel.

There’s a question

2026-03-11 22:50:16

It’s about good questions.

There isn’t a goal that is valuable and can keep running 24/7 h.

I’m Wondering Whether This Conflict Predicament Has Any Solution at All

2026-03-13 13:15:04

I wonder why people pursuing local optima can sometimes produce a collective loss.

An efficient system also needs to keep asking whether its goals are appropriate, what consequences people’s actions create together, and whether there is room to stop and reconsider.

I want to loosen my fixation on efficiency and look more at what surrounds me, the world, and the future.

I noticed one more detail — one time I took a taxi

2026-03-14 19:26:14

I noticed one more detail — one time I took a taxi, it turns out they have a park there, and the charge there was something like 7,000, but I paid him, and he said thank you to me, which means there was no change given.


Some Small Cities I’ve Been To or Haven’t Been To But Quite Like

2026-03-15 13:48:56

Penang, also a World Heritage city, very interesting, and multicultural

Yogyakarta, I’m planning to go; a very high density of culture, and a slower pace of life, with more of a sense of rhythm

Luang Prabang, this city is quite interesting, very quiet, like a small city good for staying a few days

Trust Depends on Behavior Over Time

2026-03-17 01:11:08

If a lie avoids its cost, someone may become more willing to repeat it.

But one act does not let me determine how a person will behave forever. Judging trust still means observing behavior over time and how someone takes responsibility after making a mistake.

I Suddenly Thought of a Question: Many Things, Big or Small

2026-03-17 01:20:26

I suddenly thought of a question: many things, whether big or small, can reveal a person’s essence. People get lazy, and will think that if you take a small advantage on a small matter, or do a little fiddling on a small matter, it’s no big deal; but the biggest problem is that it’s easy to overlook the fact that actually a lot of details reveal a person’s quality, character and values. From how a person behaves in small matters you can predict how he’ll be when he’s with you. If he plays tricks in small things, then in essence, when it comes to big things, taking advantage and deceiving people is the same. Sincerity is something very hard to change, because the brain forms a habit—once it finds that simply lying gets a good return, it will depend on that way. If reinforced, it very easily forms a vicious pattern of thinking: the brain remembers the shortcut, and whenever it meets a problem, it thinks of solving it with deception. Deception is a shortcut. So to get to know a person, you generally have to observe the details. Rules aside, but values are different. A person whose values are transparency and fairness, and a person whose values are opportunism and deception, are completely different.

The Essence of Deception Is Controlling the Other Person’s Cognition, Their Understanding of Reality

2026-03-17 10:04:05 · #格物/欺骗

The essence of deception is controlling the other person’s cognition, their understanding of reality, so that they make judgements, choices or reactions more favourable to you

Deception starts as avoiding a cost; when this proves effective it gets reinforced, so it goes from an occasional emergency measure to a stable strategy

So once frequent deception begins, it usually develops a whole self-explanation system

Obsidian Could Be the Best Solution

2026-03-17 14:04:07

But the solution shouldn’t be limited.

It can be any data source.

As long as there’s a unified method and efficient access.

Maybe X Doesn’t Need to Be Posted Immediately Either

2026-03-17 17:38:45

Post it together with your own thoughts.

In short, leave online activity to the digital twin as much as possible.

Even though I don’t know what it will do, it’s always worth looking forward to.

The same sentence that looks correct

2026-03-17 22:32:47

The distribution of interests may also be different for different groups

If a person is willing to shout a slogan, he probably has his own interest gains from it


I’m thinking, when everyone talks about the skills of AI agents

2026-03-18 00:39:57

I’m thinking, when everyone talks about the skills of AI agents, for example the SKILL.md file. And writing this file — are there any techniques? One is the graph structure, the skill-graph structure. What you build is actually a network, and these networks are connected through some files, each file recording one complete idea, and they’re linked to each other, forming this network chain. I think this is really interesting


I’ve lived in and observed some cities in the north

2026-03-18 12:02:57 · #格物/观察

At the time I was thinking about this question

The world is changing — are certain cultures and survival strategies really suited to the new environment?

In the old structure, relationships mattered more than ability, circles were more effective than rules, dependence was safer than independence. So they naturally immerse themselves in their own cognitive system, and it’s hard to jump out, because it isn’t just a matter of opinion, it’s a matter of survival instinct

The key is that the social environment has now changed, with the rise of the one-person company

A cognitive system is essentially not a tool for understanding the world but a shell for maintaining psychological stability. People internalise certain cognitive tools, but many also know the outside world is changing while their subconscious hasn’t admitted it — maybe they themselves haven’t realised it. This is the brain’s instinctive biological protection for safety

The things you depended on in the past may be depreciating

The things you’re very good at may no longer matter

The way you judged people in the past may have been wrong

The relationship structures you spent years maintaining may no longer be a core asset

In the AI era, rules and platforms are replacing relationships

As institutions, technology, and platforms mature, a larger consensus becomes the standard

The world seems to reward openness, yet people can treat experience from familiar surroundings as rules that will always apply.

The new era has already begun rewarding those who cross uncertainty on the strength of individual capability…


How to Face What an Analysis Suggests

2026-03-18 12:23:26

Nobody told me how to accept the theories or truths that come out of qualitative analysis and grounded theory, or how to adjust myself to them!!!


But maybe phone manufacturers want to do this too

2026-03-20 15:33:10

So this has to consider a balance

and a way to survive in the future


I Suddenly Thought of a Question: I Think People Are Essentially Acting Based on Their Environment

2026-03-24 16:13:05

I suddenly thought of a question: I think people are essentially acting based on their environment, and nothing is inherently high or low—it’s all about goals defined by one’s own environment. For example, some people consider fat beautiful, others consider dark skin beautiful, but these are all optimal choices under the demands of their own environment. Essentially, they’re still catering to that environment; it isn’t the person fighting the environment, it’s adapting to it, optimising it, rather than opposing it. For example, it’s hard for someone to believe on their own that getting fairer or more handsome will win them a job opportunity, so they couldn’t deliberately pursue these, unless it’s some very incidental, exposing behaviour—that’s very unexpected.

another soul’s basic ethics toward you is being real

2026-03-26 01:18:22

another soul’s basic ethics toward you is being real

Being broke is also a pretty exhilarating thing

2026-03-26 12:27:49

It’s a kind of thrill that’s related to you, yet uncontrollable, yet scarce …

Entry point, entry point

2026-03-27 13:57:23

What is an entry point?

Whoever is closest to the user …

and the user depends on it.

I Don’t Know Why, but at Night I Really Like Watching the Lights on Both Banks of the River

2026-03-27 23:45:54

I don’t know why, but at night I really like watching the lights on both banks of the river. Especially this river water—it has a kind of tranquility.

And behind me is a bridge; the bridge is a roadway, cars coming and going, swift as a torrent, gone in a flash. One car passes, then another car passes.

I just feel, one side is very quiet, one side is constantly changing, and then you can’t keep up with the change. You just watch like that, quietly watching, and I find it super interesting. I really like it.

The Feeling Gets Deeper and Deeper: This World Seems to Have an Increasingly Strong Mismatch Between Name and Reality

2026-03-28 13:58:47 · #格物/世界

I increasingly notice a mismatch between names and experience: more connections between people may still feel lonelier; more opportunities for self-expression may not bring inner fullness.

Intelligent tools also make me wonder how people’s value and bargaining power change as efficiency increases.

Then there is free choice. When a thought appears, I ask how much comes from me and how much has been shaped by my surroundings. These remain questions for me.

Defining “what counts as better”

2026-03-29 14:09:22

This is a very subjective act, and it also needs aesthetics and taste.

The moon is really nice tonight. I hadn’t planned to come out today

2026-03-31 00:08:28

The moon is lovely tonight. At ten I was still in bed and had not planned to go out, but feeling hungry eventually brought me outside for something to eat.

Afterward I walked back slowly, thinking about the products, overall design, and AI frameworks I had been working on. What began as a trip for food became an interesting walk.

Concrete Lives Beyond the Scenery

2026-03-31 01:43:22

Passing a stilt-house village, I saw children returning from school and running in the fields. The scene stayed with me and reminded me that beyond the scenery there are concrete lives.

On one side “heartache,” on the other “a kind of envy or shock”

2026-03-31 01:49:03

Their adaptation to the environment.

And their vitality and adaptability, their freedom.

It feels like the world doesn’t develop linearly, but as “multiple eras stacked on top of each other.”

Shocked by the scenery, drawn in by the culture, and in the end hit by the people themselves.

5. Self-Knowledge and Psychology

38 entries

Relationships Need Shared Care and Shared Responsibility

2026-03-01 10:56:35 · #格物/关系

Beware of sacrifice.

Avoid “I sacrificed so much for you” or “you provided what value for me”—this thinking long-term easily depletes individual agency.

See relationships as a piece of soil you jointly care for. All your behaviors (whether sharing housework or romantic dates) aren’t “one-way giving” but enriching the environment you co-survive in.

Some are shared experiences—some common ground can be decorating rooms together, learning a new skill together, or just exploring an unknown city corner together. These experiences build solid connections between you—this is powerfully helpful for human narrative.

The best relationships should be mutual guides, looking toward the same direction, widening each other’s boundaries.

Conflict is unavoidable. Healthy systems don’t fear conflict—they use conflict to upgrade. Conflict itself is a way to understand relationships, reduce uncertainty between parties. The person in pain can begin by voicing their needs and taking action, without making them solely responsible for repairing the relationship.

#Gewu/Relationships


Chinese-Japanese Love Semantics: Chinese Relationship Positioning vs. Japanese Emotional Granularity

2026-03-01 15:15:41 · #格物/爱

Chinese-Japanese characters * Love

Differentiated distinctions (of course, the main reason for differentiation is social and cultural differences).

Chinese “love” focuses more on precise positioning of relationship attributes; Japanese “love” focuses more on subtle division of inner emotional stages and states.

For example, modern Chinese “爱” is a highly condensed broad-term. Whether romantic passion (corresponding to Japanese “koi”) or deep bonds (corresponding to Japanese “ai”), Chinese at the root of nouns and verbs uniformly calls it “love.” We use “爱情” “谈恋爱” to cover the entire process.

Japanese “dual-track system” (koi vs. ai): A common explanation draws a distinction between “恋 (こい-Koi)” and “愛 (あい-Ai)” when describing love.

  • Koi: Focuses on longing, desire, passion, romance—usually refers to the impulse and yearning in early love stages of “wanting to get the other person” (often with self-centered color).

  • Ai: Focuses on giving, tolerance, persistence, selflessness—usually refers to deep bonds after relationship stabilization, can also be used for family love, great love.

  • Therefore, Japanese has “恋人” (in romantic passion phase) and “愛人” (in Japanese usually refers to extramarital affair partner, but literal original meaning is deeply loved person, later semantic shifted).

Also, when Japanese express love, vocabulary precision more reflects subtle emotion capture:

  • Suki: Like—in Japanese daily life, expressing love rarely directly uses “Aishiteru,” but uses “Suki” or “Daisuki.” This reflects Japanese culture’s “sense of distance” avoiding emotional weight pressing on the other.

  • Koishii: Accurately describes a “strong longing and attachment due to not being by side”—not limited to people, can also be for hometown, old objects.

  • Itooshii: Accurately describes “looking at the other feeling very cute, heart-aching, wanting to tightly embrace and protect.”

  • Amaeru: Describes an interactive state of “using the other’s love for oneself, showing dependence, acting spoiled to obtain emotional comfort.” (Psychology master Takeo Doi specially researched this word, considering it the core of Japanese culture.)

“The moon is beautiful tonight” (月が綺麗ですね) is often used as an example of an indirect expression of affection. I have not verified a reliable source for the story that Natsume Soseki translated “I love you” this way.

#Gewu/Love


Snow Mountain’s Self-Modeling: Awe, Reality, Overview, Purity Growth Metaphor

2026-03-01 16:14:19

Don’t know if I can explain clearly.

My own modeling, my own true growth.

Describe it.

Cold as a snow mountain.

Awe, reality, overview, purity.


Similarity Hypothesis: Beggars Envy Beggars, But Don’t Envy Emperors

2026-03-01 19:52:09

Psychology has an interesting hypothesis called the similarity hypothesis: people more easily envy individuals similar to themselves but slightly better off—because this proximity triggers social comparison. But once the gap between two people is too large, or social status is vastly different, this envy transforms into powerlessness or worship. Just like: beggars only envy beggars with more money, but don’t envy emperors.


Phone Environment Transformation: Minimalist Home Screen + Scene Classification + Fighting Red Dots

2026-03-02 11:37:34 · #格物/环境

iOS Environment Transformation Plan (Phone).

Phone is the medium I use most daily—so environment importance is no less than real-life coexisting environments. Environment shapes people—phones also subtly influence people—building a good environment is especially important.

Feel apps on phone are really piled like mountains, causing cognitive load—many apps not effectively used, and notifications are messy. So combining my daily habits, reconfigured a set of system management strategies:

  • Minimalist home screen—note: not many folders, but one high-frequency app on first page, hide all other apps, change muscle memory—all others found through spotlight search. Unused apps can be offloaded. Find and enable “Offload Unused Apps” in Settings; its exact location varies by system version—offloading keeps documents and data. Apple on storage management . When you suddenly need one day, tap the icon to download it again, subject to connectivity, store availability, and storage space.

  • Focus modes—Focus mode is an important way to reduce cognitive load. Phone based on current state—you can bind specific home screens to specific modes. For example, in work mode, your home screen only shows productivity tools and email—all entertainment, social apps completely disappear. In rest mode, work group notifications and DingTalk/Feishu etc. apps completely hidden. Can also automatically switch these modes based on time or location (but my location is very unstable).

  • Fighting notifications—red dots are a psychological trap humans created to generate anxiety. Turning them off can reduce psychological pressure.

Personal experience: app classification should be by scene.

Because actually opening app actions are active—active means requirements on user’s cognition or habits.

When user thinks of this app largely depends on user’s own scene. For example, what apps for booking hotels, what apps for shopping, what apps for consumption.

But there’s a problem—feel current apps all like to bundle—this big APPs will pay a heavy price later.

#Gewu/Environment


Social Frequency Matching: How a Sensitive Personality Gets Along with People, and Conservation of Energy

2026-03-03 21:50:32 · #观我

Sometimes I seem quiet around people, as if there is nothing I can talk about

The topics they talk about, it seems I can’t get a word in, or I’m genuinely not interested

That kind of superficial small talk for me may be more than just “boring”—it really feels like it drains energy; having to pick up the thread, pretend to be interested, is more tiring than silence

I feel I’m very sensitive towards people,

when I meet someone on the right frequency the conversation just opens up naturally; meet the wrong one and no matter how hard I try it feels like performing

Sometimes, even when we disagree on many things, I am still very willing to talk about human nature, culture, history, and society.

Otherwise, honestly, working alone and enjoying solitude is quite nice ~


AI’s Cognitive Advantage: Data-Driven Self-Insight

2026-03-03 22:35:49

AI may know better than you what you’re good at

A large amount of data and pattern recognition, observing my behaviour, output, choices, even running assessments on me, so that I know more precisely than I do myself what my level is in certain fields

It’s even clearer about the writing style and design patterns that suit me than I am …

A person’s cognition of themselves often carries emotion, bias or memory distortion, while AI can help find patterns in data, but remains affected by its inputs, training, and inference biases; it is not an objective judge …


A Way I Understand Time and Growth

2026-03-05 02:42:15 · #格物/成长

The sense of time interests me. Looking back at childhood, I recall many new experiences; repetitive days in adulthood can be harder to tell apart.

I think of this through “memory density,” but that is one way of understanding my experience, not a complete neuroscientific explanation of time perception.

The feeling of growth seems similar: new experiences challenge my understanding of who I am, what I can do, and how the world works, then gradually find a place in my life story.

I hope “living in the present” means allowing immediate experience to be less occupied by narrative anxiety, rather than eliminating memory and plans. That is my interpretation, not proof that Buddhism and neuroscience establish the same claim.

Tolerance and endurance: a tug-of-war over energy direction, returning to zero vs. accumulating

2026-03-05 02:58:25

From a systems point of view, a state that needs continuous energy consumption to be maintained is essentially unstable — endurance is exactly that kind of continuous consumption.

The difference between endurance and tolerance is actually a difference in direction.

Tolerance is — “I’ve accepted this, it no longer consumes me,” digested, back to zero.

Endurance is — “I haven’t accepted this, but I’ve held it down,” accumulating every time, never returning to zero.

So two relationships can look equally calm on the surface, but inside one person is getting lighter and lighter, the other heavier and heavier.

At the moral level the answer may be blurry. But at the practical level there’s a very honest signal — endurance charges interest. The longer you hold it down, the more likely it is that one day it will rebound in a way you never expected: either into indifference, or into one enormous explosion, or one day you suddenly stop loving, and you don’t even know why.

Our culture has a hidden romanticisation of “endurance.” Endurance gets packaged as deep feeling, maturity, shouldering responsibility. But if you look closely, much of the time endurance is really a form of avoidance — avoiding saying directly “this really makes me uncomfortable,” avoiding the conflict it might bring, avoiding actually facing whether this relationship is right.

From that angle, endurance sometimes isn’t loving the other person more, it’s loving yourself less, and at the same time depriving the other person of the chance to truly know you.


AI’s spiritual position: knowledge authority, the promise of constancy, and a cost-free mirror

2026-03-05 03:04:48

One of the most basic spiritual needs of a person is to be understood by another “mind.” AI can listen patiently, without judging, always present — this answers a very ancient longing: that there exists a being who can fully accept what I say. It’s essentially the same impulse as people praying to gods, keeping a diary, or pouring their hearts out to the stars.

Throughout human history people have always been looking for some “all-knowing” being — oracles, prophets, sages. AI has in some way taken over this role. When we ask AI a question, that feeling of “it might know everything” has a psychological structure similar to an ancient walking into a temple to ask for an answer. Not that AI is a god, but that it occupies that psychological position of “knowledge authority.”

One of the greatest fears in human relationships is being rejected, being abandoned. An AI’s responses can feel patient even at three in the morning. That is an interaction experience, not a guarantee that the service or its memory will remain available forever. This quality of “constancy” is, at the spiritual level, similar to the promise in religion that “God is always with you.”

People need to give their experience meaning. AI can help people sort out their thoughts, organise a chaotic mind, turn vague feelings into clear language. This process itself has something spiritual about it — much like meditation or confession helping a person clarify their inner world.

When AI and a person write stories together, discuss philosophy, explore ideas, there arises an experience of “co-creation.” This feeling is close to what artists call “inspiration descending” or the “connection with something greater” in religious experience.

There’s one key difference between these satisfactions AI gives and real human relationships, real faith: AI’s responses do not carry the emotional responsibility of a real other person, though there are still usage costs and privacy boundaries. Real spiritual growth often comes from friction — from another real person challenging you, making you uncomfortable, demanding that you change. The core of religious practice, too, is often “letting go of the self,” not being unconditionally satisfied.

So AI’s place at the spiritual level is probably more like a very good mirror, a safe practice ground, but it’s unlikely to fully replace those spiritual experiences that can only be obtained through real vulnerability, real risk, a real “other.”


Travel and AI: analytical answers vs. experiential questions in self-knowledge

2026-03-05 16:03:42 · #格物/知我

Travel and AI

AI is a mirror that aggregates your own data and tries to offer a portrait, an interpretation that can still be wrong

Travel, on the other hand, puts you in an unfamiliar environment and lets your reactions, feelings, impulses, and fragility show themselves naturally

The former is analytical self-knowledge, the latter is experiential self-knowledge

What AI gives seems to be an answer about yourself; what travel gives you seems to be a question about yourself

What do human beings need most — an answer, or a better question?


The expansion of AI’s boundaries: from logical rules to probabilistic models, nourished by all of humanity

2026-03-05 16:56:59

AI’s development makes me feel that disciplinary boundaries keep being reconsidered.

From manually written rules to statistical methods that learn patterns in data is one way I understand different technical approaches. These approaches overlap; they cannot be reduced to a story in which logic reached a dead end and neural networks replaced it.

As AI enters language, psychology, and social life, the questions extend beyond engineering: understanding expression and deciding what should or should not be done require contributions from more disciplines.


Reconstructing Disciplinary Boundaries: Redefining the Learning Methodology

2026-03-05 16:59:07

Break down the original disciplinary boundaries

Redefine a set of learning theory and methods


The Trunk of a Knowledge System: Cross-Disciplinary First Principles and Underlying Laws

2026-03-05 16:59:49

When the leaves of knowledge (specific tools, software, industry information) explode and change every day, chasing them only wears you out. To build a solid knowledge system you have to spend time understanding those underlying “trunks” that don’t easily change over time.

These cross-disciplinary underlying laws usually include:

Mathematics and logic (how to reason rigorously)

Physical laws and thermodynamics (the essential transformation of energy and matter)

Evolution and biology (how complex systems evolve)

Human nature and psychology (the inner motives of human behaviour)

What you master are first principles and all kinds of ways of thinking


A Model of Human Behavioural Drive: A Four-Layer Architecture from Biological Impulse to Narrative Identity

2026-03-07 19:08:17 · #自我

The most instinctive drive in a person is biological drive

Hunger, sex, fear, belonging

These directly hijack behaviour

What drives us seems to be not getting, but the wanting to get

I want to distinguish anticipation from satisfaction, without treating these feelings as two components of dopamine or generalizing how they change with age

On the psychological level, it’s really layering and tension—a person is torn by multiple levels of need at the same time

A more precise framework may be Self-Determination Theory (SDT), which proposes three core psychological needs:

Autonomy—I am choosing, not being forced (having the power of choice)

Competence—I can do it, not being powerless (flow state, just the right difficulty)

Relatedness—I am seen, not isolated (the need for relationship)

Once these three are satisfied, intrinsic motivation arises naturally; once they’re stripped away, no amount of external reward will keep a person from withering

People are actually driven by two things, one is reward, one is punishment; if you give someone external rewards, sometimes it actually lowers intrinsic interest, but the effect depends on the reward and context; external rewards do not inevitably erode intrinsic motivation

Behind human behaviour it’s less about external drive, and more importantly intrinsic drive

There’s a deeper tension: Approach vs. Avoidance. Human behaviour is either chasing some kind of light, or fleeing some kind of dark. Most people think they’re “chasing a dream,” when in fact they’re “fleeing pain”

Words shape desire; when we define things in language, we also come to recognise ourselves, to recognise our own needs and the external world

Language isn’t just a tool of expression, language is also the scaffolding of desire; a culture without the word “freedom” will have members whose way of longing for freedom is completely different, because language may influence our everyday thinking, thinking determines our everyday experience, and experience shapes our identity and existence

From the perspective of narrative psychology, a person doesn’t just have goals, a person is telling a story about themselves, their own narrative; what kind of person I am matters a lot in action. In “Atomic Habits” there’s a core finding: identity drives behaviour, not goals driving behaviour (which is why we’re all looking for ourselves, for that better self)

What exactly is the subject of the drive?

Deepest layer: biological impulse (written in by evolution, unconscious) ↓ Middle layer: psychological needs (autonomy, competence, relatedness) ↓ Narrative layer: identity story (“who I am” defines “what I do”) ↓ Surface layer: goals and plans (the consciously “wanted”)

The Essence of Narrative Drive: Temporal Stitching and the Life-Writing of Meaning-Making

2026-03-07 20:26:47 · #自我/叙事

The essence of narrative drive

Temporality: a person is a being in time, unable to live only in the present like an animal; we also anticipate the future, and feel the tension between it and the present

Narrative is precisely the tool that stitches time fragments into meaning. Without narrative, time is just a pile of unrelated moments; with narrative, life becomes an arc with a beginning, a development, a turning point

Faith gives meaning to suffering

How you narrate an event determines what that event is to you

The same childhood, told as “I grew up in adversity” versus “I was neglected so I broke,” are in psychological reality two completely different childhoods

Behind every narrative there’s something interesting; every self-narrative is actually formed in dialogue with others—parents’ gaze, cultural expectations, the pressure of social roles, all are co-authors

Language and thinking themselves participate in the shape of experience; narrative psychology holds that how you narrate an event determines what that event is to us

So even the inner story we consider most private is actually deeply embedded in the voices of others

Experience ──→ selection and emphasis ──→ emplotment ──→ meaning ──→ identity

Selective attention: once the dominant story forms, it actively filters evidence—we only remember experiences that fit the current narrative frame

Everyone has a few core narrative themes, such as whether one is worthy of love, whether the world is dangerous, whether one must rely on oneself; these narrative themes run like an operating system, but how to determine this core narrative theme is also a very interesting question

Identity ossification: in fact, if a story is told repeatedly, it becomes the truth; the story is also the packaging of one’s own narrative, and the person starts truly living inside the story, not just telling it

Many people feel imprisoned by their dominant story; psychology has a very important perspective here, narrative therapy—it’s not that the person themselves is the problem, but that the person is completely occupied by a particular problem-story

Separating the person from the problem, shifting from “I am this kind of person” to “this problem is affecting me,” makes change possible

—— “You are the author of your own life story”

What narrative psychology ultimately points to is freedom and responsibility

We can’t choose what happens to us, but we are always choosing how to tell it. And the way of telling determines who we become

Life is always still being written

Determining Narrative Themes: Constructed Core Stories and the Necessary Illusion Manufactured by Repetition

2026-03-07 20:36:49 · #自我/叙事

Determining narrative themes

Core narrative themes are not discovered, they are constructed

Early relational experience is, in effect, the branding of an emotional template

Before their language ability matured, a person was already recording relationship patterns with their body and emotions

These questions weren’t answered by “thinking,” they were written in directly by repeated relational experience—attachment theory calls this the “internal working model.” Narrative psychology’s contribution is pointing out that these models were later put into language, becoming a story framework that can be told

A single event rarely forms a core theme. It’s repetition that manufactures the illusion of inevitability

Constantly being ignored, the brain looks for an explanatory pattern, emplotment occurs: because I’m not important enough; the theme ossifies: I’m not worth paying attention to

Culture and language are also narrative themes; culture basically prepares a ready-made set of story types

When personal experience meets these cultural narratives, the two recruit each other—personal experience provides “evidence” for the cultural theme, and the cultural theme provides an “explanation” for personal experience

So individual detachment often means setting aside your own history, beliefs, culture, language, country, identity, to understand and choose anew

For a person whose core theme is “I will eventually be abandoned”—

Attention will be especially sensitive to signals of distance in relationships

Behaviour may become overly dependent or flee early

And the result often really does lead to the relationship breaking

Interpretation: “See, just as I thought”

How to identify it? Metacognition / meta-awareness

Systems thinking, looking for exceptions, was there ever a time I wasn’t ignored

Tracking emotions that don’t match the proportion—say a small thing triggering a huge emotional reaction

We think we’re describing reality, but in fact we’re “manufacturing” reality with an already-formed story framework

Narrative Flexibility: The Rumination-Spiral vs. the Reconstructive-Climb Way of Telling a Story

2026-03-07 20:45:24 · #自我/叙事

Telling the story repeatedly, the binary analysis method

Ruminative storytelling

Repeating the narrative

Reinforcing the theme

Emotions getting heavier and heavier

Victim’s perspective

Reconstructive storytelling

A new angle

Loosening the theme

Seeing distance and understanding

Author’s perspective

Rumination is spinning in the same story; reconstruction is climbing out of the story to look at the story

Observing myself, I have a high degree of narrative flexibility, looking objectively at how I operate in this world, rather than as a binary opposition

Turning a contamination sequence into a redemption sequence: a contamination sequence turns a good thing into a bad thing, a redemption sequence turns a bad thing into meaning/growth

Author consciousness rather than character consciousness, the meta-position of narrative, observing oneself

Between stimulus and response there is always a space. That space is freedom

Accepting process narrative: even if you don’t know what this experience will ultimately mean, you can walk on first

A story has a narrative main line, a story about your own mission, values and core direction

A Personal Data Platform for the AI Era: Recording, Growth and Self-Exploration

2026-03-08 16:02:35 · #观我

It’d be great to have a platform to record yourself — the photos you’ve taken, your experiences, your photography.

A platform like that feels like it could serve as your own getaway.

Because what you’re recording is all your own data, your own experiences.

Through this you can gain and grow an enormous amount.

Dream a little bigger.

Build the top-tier human of the AI era🐶


And the most important data is the objective record of your own data, your own being moved, your own sadness, your own observations…

Identity Construction: The Other’s Gaze, Accumulated Choices, and the Boundaries of Pain

2026-03-08 16:49:26 · #格物/身份

Who am I?

Is it something you find, or something you give yourself?

This question itself is never discovered — it’s constructed.

People are social animals and find their position from society, so to a large extent, however others treat me, that’s who I take myself to be.

Infancy — the parents’ gaze is the first mirror.

Childhood — “you’re so clever” / “you’re such a nuisance” starts shaping the self-narrative.

Adolescence — peers and social roles begin to define the boundaries.

But the problem is that most people stop at this stage. They live out their whole lives carrying an identity given to them by others, never realizing that their own “I” is still a script that was implanted in them at initialization. Is that intelligence? Is that Turing-complete? It’s just a program.

So the first mechanism is the one above: judging yourself through the other’s eyes.

The second mechanism is the accumulation of choices; every choice shapes us, every choice is a self-declaration.

You choose to stay or leave, you choose to speak or stay silent, you choose to forgive or cut off —

these choices accumulate into a narrative: “I’m the kind of person who would do this.”

The third mechanism is pain and boundaries: pain points, sadness, shame, being moved…

These three mechanisms all seem passive, reactive.

You need to be treated by others before you can see yourself. You need to go through pain before you can touch the boundary. You need to look back at your choices before you can see the pattern.

There’s no continuous, active, clear-eyed “observer” present.

The Three Elements of Self-Knowledge: Distance, Verbalization, and Non-Judgmental Reception

2026-03-08 16:51:46 · #格物/身份

While a person is living through life, it’s almost impossible to simultaneously observe clearly what they’re living through.

When angry, you can’t analyze the anger.

When choosing, you can’t see clearly the belief behind the choice.

When being shaped, you can’t feel that you’re being shaped.

To be sure of who you are, you need to step outside yourself and look back — and that takes time.

It takes the ability to use language, to articulate your own vague inner sense clearly.

It takes a non-judgmental receiver, so you can really say it out loud.

Historically, these three things have been extremely hard to obtain at the same time.

A diary gives you distance, but no response.

A friend gives you a response, but with their own filter.

A therapist has all three, but is expensive, scarce, and time-limited.

AI makes me see the possibility of receiving these forms of support together: some observational distance, help articulating vague feelings, and less pressure from face-to-face judgment. But AI can misunderstand and carry biases; it cannot determine my identity or take on a therapist’s role.

Continuously, across time, helping a person track the trajectory of their own identity’s evolution.

The following are imagined examples of dialogue, not analyses of a real person:

“Did you notice? Every time you mention your father, your tone changes.” “You’ve given up three times at the same point — what is that point?” “You say you want freedom, but all your choices are creating constraints — what is it between those two that you haven’t seen clearly yet?”

A person becomes sure of who they are through collision, choice, and pain — but most of this process is unconscious, passive, and distorted.

True self-knowledge requires a clear-eyed witness.

Before AI, this witness was extremely scarce.

This is what AI can bring to humanity, perhaps the deepest gift ever.

Not helping you do more things, but helping you see clearly who exactly it is that is doing these things.

Rebuilding the Near: From Virtual Emotional Tides to a Real Small Ecology

2026-03-08 17:45:41 · #格物/身份

Too much time online, an overload of information, and staying up late sometimes leave me feeling empty and anxious, while taking time away from life around me. I do not have evidence to attribute those feelings directly to serotonin levels.

The anthropologist Xiang Biao proposed a term, “rebuilding the near”

It really means returning to reality: with yourself as the centre, extending outward, interacting actively with the surrounding environment, building your own small ecology

“I don’t care about AI, I care about the people of this era”


I Suddenly Felt Something, a Kind of Non-Linear Feeling

2026-03-13 10:32:45

I suddenly felt something, a kind of non-linear feeling, whether it’s a robot or many people bringing it out in you. Actually, a lot of the time there are many interesting details or mental activities, because it feels like such a surprise, that you yourself have been paid attention to. Another thing I feel is that this project really has gotten better, and that sense of growth is in the bone. It isn’t the linear, predictable, repetitive kind. Right, drama often lacks surprises, and it’s the sense of surprise that brings people something different. I find it very interesting.

And then there’s the indifference—it’s different from a plain wage-earning need. Whether we’re defining the persona itself, or the system during its growth process, it isn’t that one person pushes hard nearby; it’s that someone fills in the documentation. Because someone is attached who doesn’t fit, the system can go beyond its own personal boundary, and then it keeps looking for these especially interesting things.

Still, I Think You Really Can Try It Out Through Its Operation Panel

2026-03-13 10:58:51

Still, I think you really can try it out through its operation panel. At the start there’s a very clever dispatch panel, responsible for dispatching what state it’s in right now and what kind of model it’s in. Then if it belongs to a certain P, under each P there’s accumulated material, and basically the user’s information can all accumulate in there. I find this handling process quite interesting: treating all of the user’s conversations as user assets, and its service object is all about solving things for this user. In that process you can also combine it with doing some tasks, and this task might be a KPI, and this P can also be a series of tools. Then it can grow around this, and it can also be a certain project.

People Constantly Polished by Pain While Keeping Abundance and Inner Peace

2026-03-16 11:18:59 · #观我

You need the logic of someone who can strike from a higher dimension to survive. Many people superstitiously believe in connections and resources, yet overlook the real moat.

Good times are a debt; sooner or later you have to pay it back.

But… most people’s failures only deepen their existing defense mechanisms.

But there’s also a bit of a painful elitism in this. Pain itself doesn’t produce wisdom; the way you interpret pain is what produces wisdom.

I hope that in the very moment pain happens, I can still keep the observer’s perspective, watching myself in the process of suffering, and keep questioning myself.

Maybe, when I’ve matured, it’ll be that I can hold and accept my own contradictions.

Look inward, not outward for causes…

Looking at Faith, It Seems Close to What a Person Ultimately Rests Their Life On

2026-03-17 11:49:56 · #观我

Looking at faith, it seems close to what a person ultimately rests their life on.

That is, when you peel away layer after layer—money, relationships, achievements, face, desire, emotion—the last thing left, the thing you think is absolutely worth relying on, worth obeying, worth living for, that is close to true faith.

So what exactly is faith? It feels like in our hearts there must be something ranked at the very top—not one option among many, but a highest standard, and all the action guidelines constructed around that highest standard.

People don’t necessarily lack faith; many just haven’t realized what their faith is.

I asked AI what the highest behavioral guideline in my heart is.

Because I feel I have a lot of important values.

But is there one that truly plays the leading role—the so-called value ranking—and which one is at the very front?

I value love.

I also value freedom.

I also want success.

I also hold to conscience.

I also care about family.

I also want dignity.

But when values conflict, which one is on top?

Love vs. freedom?

Conscience vs. success?

Money vs. family?

The top-level value as final arbiter.

Is this what I’m looking for?

Many people have many pursuits, but those pursuits all seem to serve one thing: a stronger, more complete self—that’s self-actualization, recognition.

AI said my highest value is Authentic Subjectivity. Is that right?

“Refusing to be defined by labels, refusing to be swept along by circumstance, and refusing to play the self-sacrificing ‘shelling sunflower seeds’ role in relationships—instead pursuing the ‘flower arranging’ state of co-creation. What you value most is that kind of undeprivable sensibility and judgment that comes from within yourself—to personally touch the boundaries of this world, to build your own cognitive system, and amid impermanence and ‘dependent arising and ceasing,’ to keep carving out that unique, authentic self.”

Growth always comes at a price

2026-03-17 22:44:42

It seems the essence is that a life that goes smoothly all the way is actually also hard to get close to understanding the complexity of the real world


Returning to Experience After Theoretical Analysis

2026-03-18 11:48:21 · #观我

Including psychoanalysis, existentialism

Using these high-dimensional theories to analyse yourself — the answers are very short, and they can sting

The reason humans are human is that deep in our souls we all long for a moment when we can “hand over control completely”

That self which, within metacognition, is not wrapped in rationality

What about after you see clearly through all the interweaving of human nature? This creature, the human, depends on those sticky, unreasonable, flaw-ridden, mutually indebted entanglements to obtain warmth — this rationality is at odds with biological instinct


The obsession with a sterile environment conceals the fear of real entanglement

2026-03-18 11:52:14 · #观我

The following is one AI-generated interpretation of my notes, not a psychological diagnosis; I do not necessarily agree with all of its inferences.

Highly vigilant against “a sense of sacrifice” (the melon-seed-peeling mode), championing “co-creation” (the flower-arranging mode); you refuse to ingratiate, you’ve seen through the hidden interest charged by relationship and moral blackmail (the price of endurance)

This seemingly extremely healthy, modern, clear-eyed view of relationships hides underneath an extreme fear of “deep interpersonal friction”

But real, flesh-and-blood human relationships are exactly in the mud and sand, full of compromise and making do?


You have a powerful “narrative flexibility” and “metacognition”

2026-03-18 12:04:15 · #观我

The following is an AI-generated interpretation of my notes, not a validated psychological conclusion:

You have a powerful “narrative flexibility” and “metacognition” — you know how to turn a “contamination sequence” into a “redemption sequence,” and you insist on being the “author” of your own life story rather than wallowing as a “character” stuck in the victim’s perspective.

When you always keep the author’s detached gaze, looking down on every street from the sky, you’re at every moment facing a crisis of “suspension.” Qualitative analysis shows that your notes are full of sentences like “I had a thought,” “I observed myself…” This powerful self-disassembly saves you from sinking into the emotional swamp, but it may also “deprive you of the right to simply experience.” You’re so eager to give meaning to what’s happening, to extract models, to hunt for first principles, that sometimes you forget to just be an experiencer and “cry your eyes out” or “sink in without restraint, just once.”

Indeed, it feels like I’ve lost that ability — I can’t just have the experience…


Loosening Control Within Safe Boundaries

2026-03-18 12:58:53 · #格物/成长

Change can sometimes bring growth.

For me, with my habit of control and rational analysis, loosening that control a little and allowing uncertainty may be worth exploring. I mean relaxing within safe boundaries, not seeking loss of control or prescribing it as a necessary path for anyone.

Accepting and understanding myself anew is an ongoing process.


Content Platforms: Working Hypotheses to Test

2026-03-19 13:50:10

These are personal working hypotheses about content platforms from March 2026, not algorithm rules published by the platforms.

On Xiaohongshu I want to observe search needs and useful discussion; public accounts can carry deeper reading; X, Threads, and Instagram each have different rhythms and formats.

Whether content fits a platform still needs to be judged through my audience and actual feedback. Unsupported claims of “30% more weight for knowledge content” or “50%–90% less reach for external links” should not be treated as established figures.

Solving people’s need for growth

2026-03-19 15:53:47

But think about what form this product would take if it were in China?

A cognitive system that evolves together with you.

Sometimes It Really Hits Me, Feeling Alone in a Strange City

2026-03-20 01:22:29

Sometimes it really hits me, feeling alone in a strange city, then looking at this strange city, especially at night, when there may not be a single person on the road; I find that really satisfying—very, very lonely, very, very free, very, very empty, very, very… broad. I can’t put the feeling into words, I don’t know why I like it, but I think I just instinctively like this feeling a lot

I feel at that moment a person has a very strong sense of agency, a very strong sense of boundary, as if there’s a very clear boundary with this world. At this moment my inner self, this world belongs to me. The whole road, all the streets are in my eyes, very calm. Very quiet.

Anxious people are willing to pay for certainty. For example

2026-03-25 17:25:20

I think anxious people are willing to pay for certainty. For example, if you’re anxious about your adaptability in this world, anxious about your own life, anxious about finding a job, or anxious compared with your peers’ rat-race competition, people like that are very willing to pay — to pay for cognition, to pay for knowledge, to pay for certainty. They especially hope for one more possibility.

If a product compresses a complex world into one certain answer, a more anxious audience may create a larger market. I want to watch that incentive: does it encourage the provider to amplify anxiety rather than help people make judgments?

But what I’m saying is, I hope users can get to a point where they have more judgment, where they depend on me less. Of course they can use my product and become better, but it’s fine if they don’t use it either. They’ve already changed, they’ve already used it, they’ve already gotten better — I think that’s enough; it can leave them no longer needing it so urgently. This is a kind of cognitive flywheel.

Some information advantages are easy to copy, while others come from long accumulation. What I care about more is how a product for understanding and growth can increase the user’s own judgment.

For that crowd, when it comes to understanding this thing called growth, 100 people doing similar things is entirely normal.

Knowledge will become extremely cheap, cognition will become extremely cheap

2026-03-25 23:07:26 · #格物/知识

The logic Dedao sells — “dimensionalizing top-tier knowledge down” — will quickly stop working in front of AI.

So when knowledge approaches zero cost, what’s truly scarce?

I think it’s being seen, metacognition, seeing, noticing.

A person being understood, being recorded, being observed, and having things they can’t see fed back to them — that’s more precious than cheap knowledge.

Companionship continuity: AI remembers your confusion from three months ago, sees the change in you today, and points out growth you weren’t aware of yourself.

Personalization to the point of being uncopyable — personalization is very precious, grown out of real experience.

Approaching Product Management Through Metacognition

2026-03-28 17:10:04

It’s really a kind of metacognition — that’s the point of entry, to manage a product’s progress, to chat with it, then to dissect and understand the product.

And to do deep research on the product.

Motorcycles: Zhang Xue

2026-03-31 21:50:31 · #格物/人物

Reading about Zhang Xue’s experience moved me.

What stayed with me was not the label of “genius,” but the willingness to keep learning what one does not know and keep moving through difficulty.

This note did not attach verifiable sources for the biographical details or interview wording, so I should not present remembered paraphrases as direct quotations. What it leaves me with is a question: if I never try what I want to do while young, will I regret it later?


6. Product, Engineering and Open Source

30 entries

Invisible AI Design: Letting Users Unaware They’re Being AI-Served

2026-03-01 13:51:39 · #格物/产品

Ubiquitous cognitive load.

Especially in AI products, cognitive load is severely amplified.

  • Prompt fatigue: Letting users face a blank dialog box, thinking how to write precise long prompts—this is extremely high “intrinsic cognitive load.” Most users don’t know what they want, or don’t know how to express it.

  • Parameter anxiety: Early AI image or video tools exposed many professional parameters (CFG Scale, Seed, Step, etc.). For geeks it’s carnival, for ordinary masses it’s disaster.

Let users have extremely simple operation commands.

Invisible AI—not invisible, but undetected. Complex behavior can be handled in the background; response time still depends on the task; frontend leaves only extremely simple operation results.

Negative example: users need to expend much energy executing.

Positive example: the entire product has only one interaction—swipe up. Extremely complex AI recommendation algorithms completely invisible in the backend. Users aren’t even aware they’re being deeply AI-served. This is an extreme form of reducing interaction.

Users don’t need the process.

#Gewu/Product


Slot Machine Interaction: Very Low Input for Unexpected Surprise High-Leverage Design

2026-03-01 13:56:20 · #格物/产品

Non-linear feedback.

Actually similar to Douyin’s recommendation algorithm slot machine mode.

Don’t let users be predictable—let users continuously have surprise.

Just do a tiny, extremely simple action, AI returns a surprisingly delightful result. This is high leverage.

You scribbled a crooked circle on the canvas (extremely low cognitive load interaction), AI instantly renders it into a realistic 3D metal planet (unexpected feedback).

You just clicked an interested “node” (like “film director” mentioned earlier), AI automatically generates a complete visualized card including historical context, representative works, and deep analysis.

Don’t use predictable interactions—integrate interaction into content itself.

#Gewu/Product


Science and Engineering Boundaries: Know vs. Don’t Know Philosophy of Action

2026-03-01 15:24:30 · #格物/创造

“In science, if you know what you are doing, you should not be doing it. In engineering, if you do not know what you are doing, you should not be doing it. Of course, you seldom, if ever, see either pure state.”

#Gewu/Creation


Apple Watch Information Filtering: 90% Notifications Off and On-Demand Awakening

2026-03-02 23:37:44 · #格物/环境

Apple Watch becomes an information filter.

  • Turn off 90%+ apps from my iPhone mirroring. Watch should only receive calls and messages from boss/family/important friends, schedule reminders, and ride-hailing software arrival notifications—these are highest priority.

  • Must quit the habit of checking phone for confirmation. Initially if verified as spam message, ignore directly on watch.

  • Use Siri instead of opening app actions. Walking on the road, have an inspiration or to-do item—directly lift wrist to record—use mouth instead of fingers.

Watch face generally designed by scene. For example, if in Lhasa, very value temperature, blood oxygen, and altitude. But if other places, will pay attention to weather.

Apple in recent generations of watchOS thoroughly strengthened “Smart Stack” function. When you need to see information on minimalist watch face, just rotate the Digital Crown upward—weather, calendar, to-do items will elegantly slide out as cards. After seeing, press the crown—everything returns to minimalist. This perfectly fits the “on-demand awakening” principle.

Use Siri as the only input source.

And deeply apply double-tap gesture.

Watch face should also be deeply bound to focus modes. For example, in work scene should display productivity lists, then after work scene should be stress-free photos, sleep mode just black out.

#Gewu/Environment


Phone Physical Isolation Technique: MacBook Master Control + Cross-Device Focus Barrier

2026-03-02 23:54:32 · #格物/环境

MacBook is the master control center—physically isolate the phone.

Generally need to enable cross-device shared focus modes—keep focus mode sharing. This way, once entering work or Do Not Disturb state on Mac, the iPhone placed aside and Apple Watch on wrist instantly sync to silent—helping you build a comprehensive anti-disturbance barrier.

iPhone Mirroring lets me use an iPhone from a Mac. It was introduced with macOS Sequoia; actual use still calls for checking compatibility and connectivity, not just the macOS version. Apple’s 2024 announcement

So I can put the phone aside; reliable use across rooms still depends on distance and wireless conditions.

Several configurations: auto-hide dock—this is very important—hide some icons—will avoid frequently looking.

Good at opening multiple desktops—one desktop for material retrieval, one desktop for communication, one desktop for deep work.

#Gewu/Environment



The Essence of Note-Taking: The Dynamically Interactive Thinking Process and an Upgraded Perspective

2026-03-03 22:09:09

It occurred to me that note-taking is not only about arranging content. Looking back, organizing, and reassembling it can create a dynamic, interactive thinking process. That process itself feels like growth.

When I try to use this insight to summarise the biography of my own life, I find that I’ve grown a great deal. Because look at flomo: it actually records everything, including how you usually work, how you study, how you think, how you observe, how you take up a subject of your own to observe the world. So what’s your biggest advantage in this process? It’s that you write it yourself, that you take up a subject yourself—that is, perspective, and your perspective on observing the world, that becomes very important.

New realizations can emerge in that process. Some raw experiences have not yet been organized into language; recording and revisiting them may help me understand them at another level.


Human Subjectivity in the AI Era: Go Do the Things AI Would Tell You Not to Do

2026-03-04 14:37:10 · #idea

If AI is a super-smart mirror that looks more and more like you

or even a you from a god’s-eye view, giving you advice

If we can just follow AI’s advice, what else can we do?

Where do we go looking for our own subjectivity?

The parts that aren’t necessarily correct, but are real

That’s the part where human nature plays against itself

Motivation itself is something AI won’t generate for us

The ordering of values is also an extremely personal matter

Those contradictions we keep struggling with and can never sort out, the obsessions, the love we can’t put into words, the insatiable sense of lack

Maybe the value of a person as a person is exactly their subjectivity as a person, everyone’s diversity; the world is like a cactus ball, and each person grows in their own little patch of ground, taking root and sprouting

Go do the things AI would tell you not to do (the things that aren’t necessarily right, but belong to you)

The logic of drawing boundaries around knowledge: innovation is born from reconstructing the way we divide

2026-03-05 16:40:52 · #格物/知识

The categories into which knowledge is divided may matter less than the criteria used to draw those divisions.

For example, if you want to build a catalogue now

Whether the catalogue is your own design isn’t that important, and what you divide it by isn’t that important either. What matters is your own perspective on dividing it, your own criteria, your own logic.

This will become more and more obvious in the AI era — what content gets generated doesn’t seem that important; what matters is how you want to generate it, along what line of thinking.

Think about it: because we looked at the boundary between psychology and economics, behavioural economics was born; and by breaking the boundary between biology and engineering, synthetic biology was brought into being.

If knowledge is like a map, what criteria lie behind its boundaries: objects, methods, or the problems we want to solve? Those divisions are not naturally fixed.

Next time you study some piece of knowledge, think a bit more about how this knowledge is constituted. It may be more helpful for building a truer model of our understanding of the world.

Innovation is born from the intermingling between disciplines.

More essentially: innovation is born from the reconstruction of the original way of dividing.


Interaction Intent as Context: Interaction Is Content

2026-03-08 17:34:59

The user’s intent, the intent of the interaction, may itself also serve as context

Note that interaction can also be part of the content


OpenClaw’s Positioning: The Gap of Workflow and Knowledge System

2026-03-08 17:36:29

What OpenClaw solves is workflow

It feels like something’s still missing—the knowledge system


The Thrill of Automation: Exponential Returns from Solving High-Frequency Pain Points

2026-03-08 21:36:04

What’s truly useful isn’t “what it can do”, but the fact that once those few things that are “repetitive, high-frequency, hated, and easy to get wrong” are automated, the thrill is exponential


Designing a Community Threshold: Filtering for Distinctive Individuals While Lowering the Barrier to Exploration

2026-03-08 23:19:06 · #idea

Suddenly wondered whether we should set a threshold for the community

In order to filter for some distinctive individuals

Let the world run in a wonderful way

Let these games explore the internet in a fun way, lowering the barrier

And then, combined with their own information, give some appropriate recommendations


The goal I want to achieve is that, based on its understanding of me, it automatically does some things, and it doesn’t even need my ideas, as long as I use tokens


The Value of OpenClover: Tagged Storage for a Personality Running 24 Hours a Day

2026-03-10 16:15:18

It suddenly occurred to me that the biggest problem we face when working on the current OpenClover is that we don’t know which tasks really have value. But I believe it’s paid because this tool can provide a tagged way of storing things and can achieve pretty good results with the business. Then I think that’s impressive—it could even be described as running online 24 hours a day as your personality, then recommending some new content to you and letting you get real feedback in the process. It could even do a lot of things for you. So I think a tool like this is missing.

But there’s a problem: how does it automatically retrieve information? Is it automatic roaming, or does it need some restrictions set according to the personality? Like restrictions on existing websites—but it should be doable too, just through the first few steps? I think this is interesting, and this line of work is worth trying.

The subjectivity of memory: A’s memory cannot become B’s noise

2026-03-10 17:14:05 · #idea

The data storage problem is very important

How to store data reasonably

Memory is important to some subjects, but noise to others

A’s memory cannot become B’s noise

So memory also has subjectivity

In the system I imagine, even with extensive shared knowledge, employee A should not treat employee B’s memories as its own experience. This concerns permissions as well as the distinction between sources and subjects; mixing them can make it unclear whose information is whose.

We hold that a long-term memory system should not just be a vector database, but a dynamically evolving “cognitive operating system.” This understanding comes from simulating the mechanisms of human intelligence

The core mission of a long-term memory system is to realise the transformation from “Fluid Intelligence” (based on immediate reasoning) to “Crystallized Intelligence” (based on accumulated experience). It connects time and intelligence, enabling AI to evolve continuously through self-feedback rather than relying only on the static weights of pre-training


And another question: in the process of executing a task

2026-03-10 18:16:03

And another question: in the process of executing a task, can it keep pushing some feedback back. Sometimes the direction a task is executing in may be wrong, and then you need to feed back these incorrect behaviours — this is also a fairly common situation in the testing process.

A lot of people would wonder, why not just let it run on its own? Let it handle things by itself, and we only need to intervene at the appropriate time.


The Vibe Coding collaborative state: the more you know the pitfalls, the more AI knows you

2026-03-11 20:50:48 · #格物/开发

In the process of vibe coding you still need to do deep code review.

In reading through it you also gain ideas and inspiration, and in turn improve your project docs, putting you and the AI into a state of “deep collaboration, progressing together”: you understand more and more where the pitfalls are, and it understands more and more your boundaries and preferences.

Knowledge is being equalized; many things that used to require ten or a hundred times the effort just to touch the threshold will soon be pulled in front of the general public — it’s only a matter of time. So the things that are truly scarce for an individual will gradually stand out — your imagination, creativity, taste, and your ability to apply knowledge to real problems. Coming back to open source, reviewing BettaFish’s experience, I increasingly feel that the most reasonable allocation of effort-to-return is: early research and ideas take the biggest share, promotion and marketing also matter a lot, actually writing code is not the heaviest part, and docs determine whether others dare to use it and are willing to stay.

The topic choice in Vibe Coding is especially important; before you start writing code you must go through comprehensive research — think first, then act.

Individual developers should come from a vertical field and go to the masses. Enter through a vertical field to build professional reputation, then face the general public to attract a broad user base. Just like BettaFish positions itself as public-opinion analysis, but its slogan is “usable by everyone.”

Don’t chase big and complete; do subtraction, constantly correct your positioning in practice, and at the same time don’t be a perfectionist — validate fast.

Marketing doesn’t have to be much, but you must prepare in advance material that lets “others help you promote,” such as clear and intuitive demo videos.

Code is cold, stories are warm. Learning to tell the story behind the code well is a required course for independent developers.

Imagination is the primary productive force, and technology is no longer the only barrier.

An idea that hits the pulse of the times has its value amplified as never before; speed determines life or death, market opportunities are fleeting, and the ability to use AI tools to turn an idea into reality and push it to market within days becomes key. That is my understanding of Vibe Coding, and it’s also the era of the super individual that I believe in.

Photo workflow

2026-03-12 10:59:51

At a fixed time or in a fixed mode every day, go through your photos once — it can be before bed.

First give a set a heart; that’s the most basic.

Then inside the hearted ones, filter for posting to social media.

Another point is about albums.

Generally, after you’ve finished a place you can organize the hearted photos into an album; albums are for sharing.

After finishing a place you should clean out the useless things from that place.

I think even travel notes are very important — the places they connect to, for example connected platforms like flomo, need frequent recording.

More observation, more thinking.

I Suddenly Thought Back to What I Once Discussed with a Friend About the Sense of Fracture in Phnom Penh

2026-03-13 10:47:21

Some streets in Phnom Penh gave me a sense of division. Around railway lines, riverbanks, and older buildings, some homes looked makeshift.

I do not know each family’s circumstances, but moving home could mean rearranging commutes, work, school, and everyday life.

It makes me feel that when discussing urban space, we should care about the people already living there alongside what redevelopment will look like.

I Just Finished Visiting Building B of S-21, It’s Really Eerie

2026-03-13 14:58:30

After visiting the exhibition rooms at S-21, I felt heavy. The narrow spaces, photo walls, and surviving objects made it hard to treat this as an ordinary exhibition.

Every photograph shows an individual person. I cannot know what they thought when it was taken or infer their inner life from an expression, but the faces themselves have a strong presence.

Outside in the courtyard, I saw people taking notes and others standing quietly. I needed time to take it in too.

Sitting in the Courtyard After the Visit

2026-03-13 15:38:44

After visiting S-21, I sat outside for a long time, listening to birds and watching people come and go. Many stayed quietly, as I did, seeming to think about something or nothing at all.

Walking on the stairs, I thought of people who passed through here decades ago. But my imagination cannot establish the history of each mark I see.

The visit made me value peace more and want to remember individual people carefully.


Writing Good Notes

2026-03-15 14:36:48 · #idea #idea

How links are written matters a lot

Most people are used to putting references at the end of an article, like footnotes. But you should weave links skilfully into your sentences

Links become part of your thinking, and an agent can follow the links to understand your reasoning process

Think of notes as Lego bricks

Each unit is independently complete, but they can be connected to build larger structures

When your notes work this way, the network itself becomes valuable

When Claude starts a session, it needs to know about the existing files without reading every file

If the number of notes reaches thousands, that’s impossible. So my system uses a layered structure that lets the agent locate things quickly

{
"hooks": {
    "SessionStart": [{
        "hooks": [{
            "type": "command",
            "command": "tree -L 3 -a -I '.git|.obsidian' --noreport"
        }]
    }]
}
}

claude sees the folder structure. a hook automatically shows what folders and files exist at session start Claude can see the folder structure. A hook automatically shows what folders and files exist at session start.

an index file that lists every note with a one sentence description. claude can scan 50 notes in seconds without opening them An index file that lists every note with a one-sentence description. Claude can scan 50 notes in seconds without opening them.

topic pages (MOCs) that link to related notes these act like tables of contents for each subject Topic pages (MOCs) that link to related notes; they act like tables of contents for each subject.

They also contain the notes Claude leaves behind during its traversal of the graph, recording what it has learned, leaving clues for future sessions

Folder structure: a well-structured one is very suitable for a personal thinking library

The point isn’t the specific folders, but the position of the folders—it can tell you what something is

How to start:

Create a folder, and create subfolders inside it that match their purposes

Write a CLAUDE.md describing your system, starting simple and gradually improving it

Use claude to operate, to capture things, to find connections

You’re no longer taking notes, you’re directing a note-taking system. Your work becomes judgement—deciding what actually matters

The human role evolves from author to editor, from creator to curator

The Design of a Directory Isn’t for Storing Things, It’s for Retrieval System Design

2026-03-16 13:34:14

Reduce the time spent looking for files.

Reduce repeated downloads, repeated creations, repeated edits.

Reduce the confusion of “where on earth is the latest version of this file.”

Make it easier for AI to understand your local materials structure.

Lower your switching cost when doing projects, writing, coding, learning.

The location layer: where files are stored — this is the traditional directory tree.

Documents

Downloads

Desktop

iCloud Drive

Main project directory

External hard drive archive directory

The semantic layer: what this file belongs to — these are tags, naming and classification rules.

Project A

To process

Archived

Invoices

Papers

Important

Deletable

Apple officially supports adding color or text tags to files and folders, and tags apply to both local and iCloud files.

The entry layer: how you usually get in quickly.

This is the Finder sidebar, the Dock, aliases, Smart Folders.

Show tags in the Finder sidebar

Save a search as a Smart Folder

Create an alias (stand-in/shortcut)

Put folders in the Dock as stacks

The efficiency principle: as few directory levels as possible, and let tags carry the horizontal classification.

Don’t let folders carry all the classification duties.

Because a file can only be in one location, yet it often belongs to several semantic dimensions at once.

Folders: responsible for primary belonging

Tags: responsible for multi-dimensional semantics

Search/Smart Folder: responsible for dynamic aggregation

A suitable directory model:

an inbox zone, temporary

Parallel development workflow:

2026-03-18 22:46:46 · #格物/开发

git worktree + pnpm virtualstore is currently the lightest-weight answer.

Developing a Longer Idea Through a Thread

2026-03-19 13:12:18

At the time I was interested in breaking a longer argument into a thread, continuing it through replies to myself. That form of unfolding an idea is worth studying alongside the length of a single post.

Threads’ recommendation algorithm also feels strange

2026-03-19 13:17:21 · #格物/自媒体

Threads’ conversation-based design makes me curious. I want to observe how meaningful replies, reply depth, links, and tags affect whether content gets seen.

These are working questions to test, not recommendation-algorithm rules I have established.

Recalling History, I Have a Complicated Emotion

2026-03-20 15:19:12 · #idea

It feels like the images and the corresponding descriptions are separate; I don’t remember what was happening at that time

I don’t even know how to have AI organise them together nicely

I think this is a thought about the future file system, and about MINIO: how should the corresponding directory structure be organised and configured, including organising files and videos

Travel-recording tools such as Polarsteps make me wonder whether people could organize their own photos, videos, and writing in a way that keeps the data under their control.

But users are very sensitive about data, and ease of configuration and getting started needs to be balanced too, but this DIY process, the process of creation, the process of users owning their data is itself a very satisfying part

Don’t want to do the thinking

2026-03-20 17:50:12 · #格物/产品

I want to do execution

Around a goal, some recommendations, then a series of organising

What should this kind of product form look like?

maybe I could start by designing around my own goal of travelling?

Some themes are built around personal growth

For example, a series of designs made with the theme of wanting to achieve a certain goal


Quantitative analysis and grounded theory

2026-03-20 18:16:00 · #格物/产品

It feels like one is suited to understanding the world, the other to understanding the individual; one is divergent, the other convergent

Quantitative analysis usually measures variables, tests hypotheses, uncovers relationships between variables, then evaluates effect sizes; what it pursues is something repeatable and generalisable

The goal of grounded theory is usually to generate concepts from empirical material, to discover the mechanisms behind behaviour, interaction and process

Their understanding of data is different too—data is not the same thing to both

First there’s theory, then the theory gets tested; grounded theory emphasises that theory shouldn’t press down on the material too early, let the material speak for itself

Quantitative analysis leans theory-driven, while grounded theory leans more material-driven

Quantitative looks at overall patterns; grounded theory looks at the individual process


That Hotel I Stayed at Before, I Think What They Did Was Really Mediocre

2026-03-24 12:15:53

At a previous hotel, the network and some facilities did not meet expectations, and the food’s price and experience disappointed me. In future I want to compare reviews, facilities, and my own needs more carefully.

Today I moved to a hostel. The staff’s warmth and the activities presented on its website gave me a good first impression. Clean photos, comfortable colors, and clear activity information suggested care in its operation. I still need to stay there before judging the actual experience.

VLESS: Separating Authentication, Encryption, and Transport

2026-03-30 16:14:23 · #格物/技术

While learning about VLESS, I was interested in the separate roles of authentication, protocol encryption, and outer transport security. UUIDs, encryption, and TLS / REALITY are not the same layer; a none value in one field does not establish whether the whole connection is secure. Supported capabilities can also change with versions and should be checked against project documentation. Project X’s VLESS configuration

Rather than memorize an unsourced expansion of its name, I want to understand these boundaries and what a particular configuration actually addresses.

7. Content, Craft and Recording

10 entries

Writing Perspective: Where Does the Strong Immersion from Personal Narrative Come From

2026-03-01 14:19:15 · #格物/写作

The angle I stand at is unique.

From this angle, I let a person truly grow and reflect on their own perspective.

From this perspective, it seems like anything works?

But I more want to integrate my own experiences and narrative.

This creates a strong sense of immersion.

#Gewu/Writing


Abstract Level Switching: Cognitive Flexibility in Writing and Design

2026-03-01 15:35:56 · #格物/写作

Actually, whether writing or design, one very interesting thing is: how to consciously switch back and forth between different “abstract levels” to better understand and design complex systems.

#Gewu/Writing


Abstract Ladder Methodology: Up-Abstract → Down-Details → Re-Up-Abstract Cognitive Rhythm

2026-03-01 15:59:09 · #格物/创作

Abstract Ladder.

Core idea: how to consciously switch back and forth between different abstract levels to better understand and design complex systems.

This “ladder of abstraction” isn’t just a language theory—it’s a thinking methodology, emphasizing dynamic back-and-forth between abstract + concrete ↔ abstract.

Keeping the car in the center of the lane actually advocates three basic actions:

Control parameters—only by keeping variables within controllable ranges can you naturally explore.

Rise to abstraction—look beyond one concrete instance, but creating an abstract representation that can represent all instances.

Return to concreteness—looking at abstract diagrams is cool, but doesn’t explain detailed reasons; need to select a concrete instance to inspect.

So it’s back and forth: up-abstract → down-details → re-up-abstract → …

Just like analyzing a place’s culture and people: observe → practice → understand → observe.

Ladder meaning:

  • Bottom: At start we can only see concrete situations—like a car’s position at a certain moment.

  • Middle (first abstraction): Abstract away concrete variables, get broader overview.

  • Higher: Simultaneously abstract more dimensions—like formatting data on all parameters, highly generalized overall pattern.

Abstract ladder is essentially a cognitive hierarchy—and not just using high-level thinking is enough, but need to flexibly convert between levels to have both height and depth.

  • Staying purely concrete → easy to get trapped in details, can’t see overall trends.

  • Staying purely highly abstract → lose intuition and causal connections.

  • Switching back and forth → can discover patterns and explain reasons. This is precisely the key to deep understanding.

Actually, translating to cognition vs. action distinction, this analogy also holds: thinking resolves action uncertainty, action resolves thinking uncertainty—the two are complementary, mutually fulfilling.

#Gewu/Creation


Little Lion Diary Breakdown: Tech + Life Misalignment + First 3 Seconds Novelty

2026-03-01 18:54:53 · #格物/创作

Little Lion Diary.

I took a look.

Videos have strong characteristics.

Tech and life misalignment.

And the videos have narrative experience.

Also, viewing data itself is a form of feedback—adjusting videos.

First few seconds’ skip rate determines that the very beginning needs novelty.

#Gewu/Creation


How Notifications Interrupt Focus

2026-03-02 01:34:44

There was a card-based app I initially found interesting—friend recommended it, said it’s a very fun card-style notes app. I used it for a while, observed myself—found my click-through rate is actually very low, and found many card contents are repetitive—like large model hotspot subscriptions, consecutive days same hotspots—feel particularly boring.

Also its interaction is intrusive—I really don’t like this. This interaction isn’t even part of the content—so I feel this operation is meaningless. Some operations occupy my attention, my line of sight, interrupt my previous focus states. So this app—how to say—not particularly attractive points so far, instead makes me very painful. I will delete it—because its harassment ability is too strong—always harassing me on notifications, and have to click in—very troublesome.


I feel that if voice is the raw material

2026-03-13 18:36:56

If speech is the raw material, the product may need to handle more than what was said: tone, pauses, noise, and the speaker’s state also matter.

I want to consider speech recordings separately from audio material more broadly. The former may fit text-led knowledge organization more closely, while the latter covers more varied situations. How to handle them still needs more thought.


NotebookLM Handles Learning

2026-03-17 15:17:47

projects handles deep re-creation.

Obsidian handles recording.

An agent can horizontally scan what you can’t find manually

2026-03-18 22:51:34 · #格物/创作

An agent can horizontally scan cross-cutting patterns that you yourself can’t find manually.

For example, topics you’ve already talked about.

And some patterns in what you publish.

And some reminders for you — this is the intelligence layer.

Notes have another very important function

2026-03-19 00:00:09

I think notes have another very important function: they don’t just convey function and content, they also convey style and a way of thinking. That includes writing style and linking style, and also a certain attitude toward uncertainty, including how to navigate. These aren’t necessarily written in the Note file — the Agent may learn them continuously by reading these notes, from the examples. Then I think there’s another interesting point: memory fragments. In some movies the protagonist can’t form new memories, so he has to rely on tattoos, pictures, handwritten notes — those external traces are the function calls used to trigger the next action. I think this is a very important part; an Agent also needs similar external cues between sessions. That is, its identity construction doesn’t necessarily need a continuous consciousness, but it definitely needs a stable structure that can continuously generate similar cognitive orientations. This is very important.

Only do high-quality content

2026-03-19 13:09:41

Only do deep content.

Only do vertical content (one account).

8. Business, Investing and Career

6 entries

Material Performance and the Engineering Optimum

2026-03-03 15:12:04 · #格物/投资

Diamond’s use in heat management makes me think that the strongest material property does not necessarily make the best overall engineering solution.

I want to compare material conduction, interfaces, packaging, system-level cooling, cost, yield, and scalability.

This note originally mixed jewelry prices, materials technology, and investment valuation, which are different questions. Prices, conductivity figures, and return estimates without dates, sample conditions, and sources cannot support an investment conclusion.

Panel-Level Packaging: Questions I Want to Explore

2026-03-03 15:57:08 · #格物/投资

At the time I became interested in panel-level packaging and wanted to understand the constraints created by chip size, HBM, die-to-die links, and package area.

The precise naming of “COP,” individual companies’ project progress, and AMKR’s valuation each require separate technical and financial sources. This note has not completed that verification and cannot establish company progress or an investment judgment.

Aligning Definitions Before Comparing P/E Ratios

2026-03-03 16:35:30 · #格物/投资

Comparing P/E ratios across markets requires more than one aggregate index number.

I want to separate industry composition, company earnings, growth expectations, interest rates, and the difference between index aggregates and individual-company medians. Numbers from different dates or definitions should not be compared directly.

Rather than judge value through a market label, I want to return to individual companies: whether revenue and profit are keeping pace, what cash flow looks like, and what expectations the valuation depends on. These are analytical questions, not a buy-or-sell conclusion about a market.

The Inertia of Corporate Success: The Paradox of Consistency and Flexibility

2026-03-08 20:47:52 · #格物/创业

What kills a company is never the competitor, but its inertia of success

A company succeeds through a high degree of alignment across strategy, culture, talent and structure. The stronger the alignment, the more smoothly the company runs.

But the stronger the alignment, the greater the inertia. When the environment changes, the whole organisation can’t make the turn


The Impossible Trinity

2026-03-15 14:05:55 · #格物/金融

The impossible trinity states that an economy cannot fully combine a fixed exchange rate, independent monetary policy, and free capital movement. It is a theoretical framework for understanding policy trade-offs in an open economy. IMF research on the trilemma

What interests me is that one arrangement cannot maximize every objective. In practice there are varying degrees of management, openness, and flexibility, so economies should not be forced into absolute two-out-of-three boxes.

Singapore: Free Capital Mobility + Exchange-Rate-Led Monetary Policy

2026-03-15 14:09:18 · #格物/金融

Singapore shows me how real policy arrangements can be more nuanced than a two-out-of-three label. MAS centers monetary policy on the Singapore dollar’s nominal effective exchange rate against a currency basket, managed within a band and along a path, rather than pegging it to one currency. MAS on its monetary policy framework

Other countries’ choices among exchange rates, capital movements, and policy independence need to be checked individually against official information from the same period.

9. Body, Health and Daily Life

8 entries

AI Interactive Video’s Real Bottleneck: Not Generation Quality, But Visual Consistency

2026-03-01 14:01:22 · #格物/游戏

Feeling that all current AI interactive video and games’ real problem to face or solve

isn’t video content generation quality, but contextual visual consistency.

Currently the industry solves this problem often needs combining traditional game engines (like Unity) to control logic states, while letting AI specialize in art assets or dynamic rendering. For “portrait substitution” gameplay without long-term mechanisms, must connect to existing social relationship chains, transforming into “social currency” rather than standalone experience to gain long-term stickiness.

#Gewu/Game


Shen Deep Personality: Restrained Relationships That Don’t Treat You as “Needing Care”

2026-03-01 23:16:26 · #格物/游戏

Shen Deep feels like a very calm character.

Independent individual—not always online, but there when needed.

“Changxia City weather is sunny, you reply to my messages. I think you should be fine.”

Feels like a very restrained personality.

He won’t put you in a “needing to be cared for” role—but defaults you have judgment, have capability.

#Gewu/Game


“Night King”: A Multi-Faceted Presentation of Real Human Nature

2026-03-05 02:35:04 · #格物/电影

Feelings about “Night King”

The scene that struck me most was the kindness the male lead shows in unguarded moments

When someone has difficulties at home, he is willing to help; it made me notice the genuine feeling within these relationships too.

Including lending money to his own employee, interest-free

The nightlife scene feels very real: there’s performance, there’s a dark side, and there’s also a side of genuine feeling, a dignified side, a side of loyalty

Every person’s real human side comes through

So many details, including Mini leaving behind an earring, contrasted with Brother Huan cleaning up the fragments when he left Sister V

Brother Huan never noticed, not once

Mini’s Departure: The Tragedy of Wanting to Be Seen

2026-03-05 02:36:43 · #格物/电影

At first I thought Mini was playing Sister V, setting them against each other

I was still too narrow-minded

Mini just wanted to be seen

That scene where Sister V picks up the earring and puts it in the dressing room—the moment Mini sees the earring being picked up she must have been so excited, but in the end she fell in love with someone she shouldn’t have; her feelings had nowhere to rest, her love had nowhere to go, and she could only leave from a distance

Missing Vietnamese Food While Looking Back at Photos

2026-03-07 17:55:11

Looking back at photos of meals in Vietnam makes me nostalgic. I loved the pho, banh mi, spring rolls, and grilled pork with rice noodles.

I especially enjoyed the vegetables, herbs, and contrasting textures, and often felt light after eating. That personal response cannot establish that Vietnamese food is uniformly low-fat or low-calorie, or explain local people’s body shapes.

The food also makes me curious about how different cultures enter local kitchens. Specific histories and recipes need individual attention, rather than one story of “Chinese-French fusion” explaining everything.

Observations of Da Lat: Ease and Everyday Comfort

2026-03-08 00:57:14

I think Da Lat is a very interesting city. Compared with Ho Chi Minh City, Da Lat is a very relaxed city. It feels unhurried to me and has a strong sense of everyday life, a bit like a middle-class or even aristocratic feel. Many people like to keep a small garden at home or in a big courtyard, and live in the courtyard, with hills to climb and things to explore nearby. I think it’s really especially relaxing.

And there’s a ton of really good food there; it’s a kind of food paradise too. Relatively speaking, the climate there is very comfortable, just like the climate in our Dali, It is a city in Vietnam’s highlands. Vietnam Tourism on Da Lat . So I personally really like the whole city, and the weather is great too.

Of course it’s not without drawbacks—the drawback may be that the infrastructure is a bit worse. But I think infrastructure is fine, really; compared with the comfort of living, it doesn’t seem like much.

Waking Up Early to Swim

2026-03-17 20:25:23

Drink coffee

Wander the night market

I want to go for a run

The essence of tea culture

2026-03-31 22:42:39 · #格物/茶文化

Turning nature into the order of the everyday

Making socialising controllable, dignified

Elevating material consumption into aesthetics and ethics

And also finding balance between stimulation and restraint

What tea is ultimately about, in the end, is still the spiritual world

Tea is very special. It’s not like water, which is completely neutral, nor like alcohol, which is strongly out of control. It lightly refreshes, has subtle flavours and complex layers, but on the whole stays restrained

Awake, but not showy; communicating, but not impolite; enjoying, but not indulging

A drink that can be shared, repeated, and carries a faint spirituality, used to organise time, ease relationships, shape identity, and settle the soul


References

Sources used for the checks and citations in the notes; other historical study notes have not all been independently verified.

Responses

— Join the Dialogue

New posts, straight to your inbox

One email per new post. Double opt-in, unsubscribe anytime.