April 2026 Thought Notes: Living in Laos, Borders, AI and Agents

Puzzled: Why Did This Claude Code Wave Hit So Hard?

I’m a bit puzzled why this time Claude Code triggered such a big shock. After its source code leaked, open-source projects appeared immediately, and in a very short time they gained a huge number of stars — on the order of 100k stars. That’s incredible, that kind of thing

Technological maturity × market anxiety × replicability × open-source diffusion mechanisms = explosion

The Tipping Point from Agents to Production-Ready Demos

I feel that right now a kind of tipping point from agents to production-ready demos has already arrived. Things like the prompt engines of the past, or RAG, or multi-agent — right now it feels like the whole LM or AI ecosystem has reached a turning point. The turning point is going from only building some engineering, only building some methods, to now genuinely being able to do a lot of things, even build a very sound system. I think this is an enormous shift, and it’s very interesting

Long Term, Claude Code Still Has a Moat

So long term, I feel Claude Code’s moat is still there. Right now it’s just that on the CLI, including the agent system, it may lose some ground to peers — it may get caught up with. But competition is pluralistic by nature; in the short term it may look like the technology has been flattened a bit, but long term I think its ecosystem, including its standards, is what’s sustainable — including the cost of educating its users

Even the Earlier AutoDeepResearch Project…

I feel that even the earlier project, the AutoDeepResearch one — it would actually try to make your goal as clear and credible as possible. So at that point it has a series of features designed around that goal. For example, defining metrics: metrics are very important, they decide the purpose of your goal and how it’s evaluated. Only with evaluation, with quantified metrics, is there feedback, and the feedback comes back to you, and you keep at it. So I think measurability and deformability are a very important part. That includes scoring scripts and templates; they’re all part of the work. And then there’s convergence — when it should stop once the state is stable. That’s convergence. And then stop conditions: once a certain situation appears, for example a score reaches 80% or above, it can stop. Or it has iterated more than 10 times in a row, or it has already completed 20 iterations. Right? That way it can stop too. This is a quantified behaviour for a certain domain. But concretely, we still harness around a goal. Actually it’s about giving it an environment. Prompts are one aspect too. I think there are some other important points in doing harness work, for example designing a system that can run itself. In this process the environment matters a lot too — what kind of safe environment it executes in, and what it’s allowed to use. Then there’s an improvement strategy: how it tracks its own behaviour in this loop, and how it improves itself

And I saw a fairly interesting point: its AI review has some metrics, one of which is resonance. Resonance means which things — whether to add images, or strengthen the core story, or add more voices. It has some examples, and these examples may not match your project, but they give you a way of thinking. And then there’s clarity, or credibility. But from my own testing, I feel it highlights certain directions I don’t think much of myself. For example, if I’m writing old project code and the old project code has some problems, I still won’t change them. I think old project code generally needs to be stable, rather than being optimised

Japan Stays Calm About OpenClaw

Japan stays very calm about OpenClaw — they seem to keep a certain distance, with a very cautious attitude, towards anything that changes especially fast, or that is unstable or unsafe. This may have to do with their cultural history, and with their system. They are a country that does things very much by the rules

Some people might feel they’re behind us. But actually there’s no such thing as behind or ahead. “Behind” means everyone completing the same goal from the same starting line, and if in that process one side gets eliminated, if it’s an elimination race, then there’s such a thing as behind and ahead. But obviously the large-model field isn’t like that. First, governments everywhere aren’t on the same starting line, and their goals aren’t very similar either. Even between China and the US there will be certain differences. China may care more about applications, including profit models. So, how should I put it, there’s no ahead and behind either — there’s only what is good, what adapts to the future environment, and what can’t adapt to the future environment. Of course that’s also our subjective view, and it could be wrong too. Right now we all feel that people who do things by the rules, or systems like that, may face the risk of being optimised away by AI. That’s an inevitable trend.

And I think there’s another point: two types of ability are actually quite interesting. One is occupational skills — the kind of work that requires a person to be genuinely present as proof, or to genuinely handle complex and difficult environments; that kind of work may depend heavily on human ability. And the second, I think, is neurodiverse people. For example, some people with dyslexia, people with autism — they don’t follow the standard process, and they often have their own distinctive strengths. Like God taking away a pair of eyes but perhaps giving you more sensitivity, more inspiration, more perceptiveness, to let you feel this world — feel the wind on your body, feel the sun shining on you, feel the music your own ears hear, feel the world your own hands touch. So, because what’s easily disrupted is an industry that follows a standardised process, but what’s often hardest to replace is the person who jumps out of that framework and thinks about problems in a very non-classical way. That kind of person I find quite interesting, quite fascinating. In the AI era, they may have their own distinctive place, their own distinctive perspective

All AI Products Should Be Designed as Runtimes

I feel the form of all AI products right now should be heading towards designing a Runtime, that is, a harness. Not that the interface doesn’t matter at all — I think the interface is the least important thing right now. Look at the form Claude Code exists in: you could call it a terminal form, or an IDE form, or a desktop app form, or a web form. But I think the UI is just an entry point, it isn’t the subject. We need to grasp one subject — the Agent OS subject. How that subject should be built, I think that’s very important

As for which client to build it for, what the shell is, I don’t think any of that matters. This thing is just an entry point from the outside; what matters is designing around the Runtime. I think last year, when building products, because of the stage AI was at, I still had a misconception. But now I’ve decided to correct that misconception

Funny: It Even Has an Anti-Distillation Mode

I find it really funny that it even has an anti-distillation mode: to stop competitors from stealing its thinking through the API to train their own models, it builds in an anti-distillation mechanism that injects fake tool definitions into the data stream, or returns only encrypted summaries, summarised reasoning. I think that’s pretty funny. On the one hand they fake identity — if it detects that the user is from their own team, their own company, it adds an AI attribution statement, that the words Claude Code absolutely must not appear, and that it mustn’t suggest in any way that the author is an AI

I think there’s a method: when you’re using Cici

I think there’s a method, which is that when you’re using Cici to write code, you can tell whether this thing needs to be corrected over and over. You can keep talking to it in the session, in the conversation, and if its output is wrong, you should correct it promptly along the way — if you don’t correct it, it’ll drift further and further off. You talk to it through the process and correct it twice; if it’s still wrong after two corrections, I think you should clear its context and start over. Rather than getting stuck on — on this context. You should think: the more it writes, the more it gets wrong, and the more it costs you to fix the errors. To fix one error you have to spend that many tokens. There’s sunk cost in this.

Even when talking with AI, what we should really be learning

I think that even when talking with AI, what we should be learning more is not how many prompt tricks there are, or how many cool operational skills we’ve mastered.

The three scaling dimensions of Harness Engineering

Harness Engineering’s three scaling dimensions: time (let the agent run longer), space (let hundreds of agents run in parallel), interaction (let a person manage it without dying of exhaustion)

Of course this goal also applies to raising lobsters

And I feel the problems you run into differ a lot from scenario to scenario, and that difference often decides the current shape of the harness

OpenAI’s solution:

A person’s core work shifts from writing code to designing the agent’s working environment

Knowledge must be versioned and discoverable, and then live in the repo

Constraints are far more effective than instructions — limiting specifically what it does, deterministic

Perfectionism is the enemy of throughput — don’t chase every output being perfectly correct, otherwise the system will lock up and efficiency will be even lower. Accept imperfection, prioritize getting things flowing. Rather than letting an agent polish a PR to perfection over and over, better to merge it first and open a round of cleanup later (in the agent era, correcting errors is actually easy — no need to worry that everything must be completely fine before merging; rather than letting an agent polish a PR to perfection over and over, better to merge it first and open a round of cleanup later

Anthropic’s practice shows that the evolution of a harness

Anthropic’s practice shows that the direction a harness evolves is not

thicker and thicker, but thinner and thinner as models improve. Every time a new model is released, the first thing you should do is remove some components and see whether anything breaks, rather than keep piling things on

For agents, maybe this is a point of inspiration

First, they refactored the repo from a monolith

First, after they refactored the repo from a monolith into multiple independent crates, compile wait times dropped sharply and throughput multiplied several times over. This means the repo structure optimized for agents may not be the same as the repo structure optimized for humans. A human can wait 30 seconds for a compile; hundreds of agents compiling at once — hundreds of 30 seconds — is a catastrophic I/O bottleneck

In multi-agent architecture, a point you absolutely must think about is how to avoid catastrophic I/O bottlenecks

Because if 100 agents are working at the same time, each one has to compile once, and each compile takes 30 s, that’s a system-level disaster — the system just locks up / compile time rises exponentially

If it’s split into multiple crates, when an agent modifies module A it only compiles module A, compile time is lower, and it avoids a concentrated compile explosion

Repo structure optimized for agents ≠ repo structure optimized for humans

Fine compile granularity

Concurrency friendly

High cache hit rate

Scattered I/O

A question

Once an agent is no longer “a little tool you run occasionally” but has become “a production system that continuously produces, runs in parallel, and opens PRs automatically,” the real bottleneck is no longer the model itself — it’s “how do humans keep up with managing it”

OpenAI calls this scalability at the interaction level: it’s not whether agents can do the work, but how humans can steer a large group of agents with a sufficiently low cognitive load

A person’s center of gravity shifts from watching every step the agent takes to designing a system in which agents will most likely do the right thing automatically — and its form is tickets

Write a ticket

Drag the ticket to some state

The system automatically schedules an agent to do it

The agent proves its own result

Humans only review the final artifact

If there’s a real problem, fix the harness first, rather than hand-holding and correcting the agent every single time

Not a stronger coder agent, but an “agent orchestration / operations system”

The harness itself should be a complete system

A lot of people understand harness too narrowly, but in fact it should at least include:

Task entry layer

Execution orchestration layer

Fault tolerance and recovery layer

Verification and constraint layer

Human-computer interaction layer

The natural convergence of agents

Even though every agent will make mistakes

we still don’t chase perfection

but instead iterate many times, through many PRs, many rounds of revision, and in the end the system tends toward the correct state

In a large-scale agent system, errors are unavoidable; the key is not to eliminate errors but to choose how to handle them:

Either: use an evaluator to force corrections (a control-type system)

Or: accept errors and let the system converge through iteration (a statistical-type system)

Today’s ARCHITECTURE.md

Today’s ARCHITECTURE.md is actually changing from “a system explanation for new developers” into “a repo-level contract read by both humans + agents”

You can make use of Claude Code’s hooks

You can make use of Claude Code’s hook capability; hooks feel really handy

You can let Claude Code create them itself

Getting really familiar with how Claude Code is used

and its source code

often lets you understand design more deeply

Spend all your time on building context and

Spend all your time on building context and harness

rather than optimizing prompts

The ceiling on prompts is far too limited, and it’s very constrained by the model’s own capability

Codex and Claude Code

The difference in execution strategy between Codex and Claude Code

Codex does agent planning and executes it in a sandbox (temporary FS + environment)

produces a diff / artifacts

then applies them to the real repo

so command execution is copy-on-write and can be run safely

Codex doesn’t care about side effects of the process, only about the diff and the result

Claude Code directly uses four ways (read, write, execute, connect)

with no isolation; the steps really happen, there’s no replay phase

The harness is the same — both have tools

Agent Harness = LLM + Tools + Execution Layer

Why launchd suits scheduled tasks better than cron

But it seems it has no way to wake up an iOS that has already gone to sleep

But it has a very clever mechanism: if the screen is merely off — that is, say, you’ve plugged in power or something is downloading in the background — then the Mac’s disk or CPU is still running normally. At that point it can trigger on time regardless of whether the screen is off. But if you really close its lid, or you click sleep, then it enters a very low-power state, and the CPU temporarily goes into a non-essential process. At that point its tasks get suspended and can’t be executed. But launchd has a very smart trick: if a task is set to run at 3 a.m. but the Mac is asleep, then when you open the lid the next morning at 8, or at the moment the Mac half-wakes, launchd will immediately read that missed task

Traditional knowledge bases are more like “retrieve when asked”; LLM Wiki

A traditional knowledge base is more like “retrieve when you ask”; an LLM Wiki is more like “first compile the knowledge into an intermediate layer that persists”

Raw sources are the source of truth; the wiki is the knowledge layer the LLM maintains; the schema is the rules layer

The schema is a lot like AGENTS.md / CLAUDE.md — it determines how the wiki is organized, how it ingests, how it queries, how it does health checks

This gist was explicitly written by Karpathy as an idea file. What that means is not directly delivering a finished product, but handing the thinking to someone else’s Agent, which then implements it further based on concrete needs

This LLM wiki, I feel

I feel this LLM wiki — and I think it’s quite inspiring — that this wiki takes raw data, very raw data, and processes it, processes it into a wiki archive. And this wiki archive is for AI to read, and the AI actually also has some — rules, rules similar to Claude or Codex, like defining a Markdown file, and telling it, being passed into the system prompt every time, emphasizing that you should do it this way

A good way to take notes with Claude: plan first

Use one session to do planning, having Claude use the AskUserQuestion tool to interview you in depth, covering technical implementation, UI/UX, edge cases and trade-offs; after the interview is done, write the full spec into SPEC.md; then open a brand-new session to execute. The new session has clean context, focuses entirely on implementation, and uses the written spec as reference

A simple, direct way for agents to interact with each other — the file read/write way

In the process, have Claude confirm its understanding more before starting work — whether it has understood the task — and ask it what it plans to do and why

If you find that every session starts by typing the same instructions, you can turn them into slash commands

Agent Teams

Subagents are like “interns sent out on errands” — they receive a focused task, finish it, and return a result. The main agent always holds control; the Subagent doesn’t need to communicate with anyone else. Agent Teams is more like forming a project group: each member is responsible for their own part, but they stay in communication the whole time. When the backend developer changes the data model, the frontend developer knows immediately, rather than only finding out when integration tests fail

Subagents = Delegation, Agent Teams = Collaboration

How to choose between them? The core criterion is: do the Workers need to communicate with each other? If the work is independent of each other (like a backend audit, a frontend audit, and a test coverage check each happening separately), use Subagents. If the Workers need to challenge, review, or build on each other’s output (like one Teammate writing an API spec, another reviewing it and flagging inconsistencies, and a third writing the corresponding tests), use Agent Teams

Generally speaking, there’s only one scenario that needs Agent Teams

which is when multiple agents need to play different roles and review each other’s output

An LLM continuously maintains a structured knowledge base that has already been compiled

An LLM continuously maintains a structured knowledge base that has already been compiled. When new material comes in, it isn’t archived for later use — it’s immediately integrated into the existing knowledge graph, updating related pages, flagging contradictions, strengthening cross-references

https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f

Analysis of quantification methods

Dimensional decomposition + Few-shot calibration

One is to take a complex problem or a vague problem and split it into N sub-questions that can be answered independently, each with a clear pass or fail condition. Then weight the dimensions to get an appropriate score

Compared with comparative ranking, I think rather than giving an absolute score, it’s better to ask which of A and B is better. For humans — and including model judgment — judging relative quality is much more stable than giving absolute scores. Like using this image or another image and comparing them — that’s also a method; let it keep cycling through this way of training, until it can collide with something with a relatively high contrast

And there’s actually another very important thing: including when we write Claude, we don’t list what’s good; instead we list what’s bad. Because hitting a deduction has a far higher hit rate than hitting a bonus

Then there’s measuring through some mechanically measurable behavior, like using Playwright to click the UI and test API endpoints, then checking the screen state. That’s more precise than you saying whether usability is good or not, and it’s also easier for an LLM to solve

Then there’s giving some Few-shot anchoring — this means calibrating with examples that have detailed score breakdowns, aligning with humans’ own preferences, to reduce scoring drift. Not just telling the model to score 1–5, but why this case is a 3 or a 4 rather than a 5

Is this design beautiful?" is hard to answer consistently, but “does this follow our principles for good design?” gives Claude something concrete to grade against.

What quantification’s method is doing isn’t measuring “true quality,” but measuring “the degree of alignment with some set of principles.” So whether a quantification method is good ultimately depends on whether your principle set actually captures what you want

Measurement is essentially relative too. When we read about relativity

I think measurement is essentially relative too. When we read about relativity, apart from the speed of light, all other standards of measure — like speed, why time speeds up, and who is moving and who isn’t — are relative. There’s a frame of reference to measure whether this object has shifted in the real world. And especially, some shift values may depend on human preferences, culture, and situational needs

The principle set can go wrong at three levels:

Incompleteness — your principles don’t cover what you actually care about The article’s four design dimensions missed “emotional resonance” — a design can meet all four dimensions and still leave the user feeling nothing. This dimension is hard to operationalize, so it was omitted. Omission doesn’t mean it isn’t important, only that it isn’t measured.

Faulty proxy — the thing you measure is a proxy for the thing you actually want, but proxies distort Goodhart’s Law applies precisely here: when a measurement becomes a target, it ceases to be a good measurement. The evaluator in the article was trained to penalize “purple gradient on a white card,” so the generator learned to avoid that feature — but that doesn’t equal genuine originality, it just means that particular signal of originality failure is gone.

Evaluator drift — the principle set itself encodes the evaluator’s preferences, not the user’s The article’s author wrote: “The wording of the criteria steered the generator in ways I didn’t fully anticipate.” The principle set included wording like “museum quality,” and as a result all output converged toward museum aesthetics. That’s the evaluator’s taste, not the user’s need.

A common misconception is: if the principle set has problems, just keep refining the principle set until it perfectly captures reality — but this path generally doesn’t work, and the reasons are:

True intent is itself fluid. You can’t capture a dynamic preference system with a static principle set. The design you think is good today may feel outdated six months later. Your principles at this moment are the sediment of your experience up to the last moment

Operationalization itself is a form of lossy compression. Turning “emotional resonance” into a scoreable metric necessarily requires trade-offs and simplification. This isn’t a technical problem; it’s a limit of language and concepts themselves

The evaluation system shapes what is evaluated. The Generator isn’t just being measured — it changes during the process of being measured, converging in the direction the principle set defines, and simultaneously moving away from the space of possibilities the principle set doesn’t cover

In the end this can’t be solved, only managed

I feel like I still have a lot of repos, so I’m thinking about how to manage

I feel like I still have a lot of repos, so I’m thinking about how to manage them. Because look, I still have my blog project, and all my notes — I think all my notes should be consolidated into one repo. In that repo there can be life-related things, knowledge-related things, growth-related things, even thinking-related things, and even daily diaries. But I think they should be put together, because it’s one record of me. And I think it’s all based on raw formats, metadata formats. So I think these metadata formats should be preserved as raw as possible. And kept in a very centralized repo. And then the model can do some compilation based on this format; the purpose of that compilation is to sort out all your context — the purpose of sorting it out is to let you act better, to create the next step better. So could I have a hypothesis: I think first we need to have a unified repo; this repo is not only a unified record for us, but also a unified thing to process for AI. And then the content after AI processes it can also be in this repo — of course it can also be copied and created outward, distributed to other directories. And the other directories — say your blog, some web pages, a personal homepage, or your WeChat official account publishing, anyway — they’re responsible for something very pure: distributing the content you create. And every record of you and AI should also be stored, stored in very raw data. Its purpose is to let you do effective management, because actually every chat, every recording, every one of your records is very precious context

OpenAI / Anthropic

The OpenAI / Anthropic competition is starting more clearly to shift toward “capital + enterprise deployment”

The latest clues from Reuters show that both are continuing to push for larger-scale enterprise partnerships and financing designs; the focus isn’t simply fighting over model leaderboards, but fighting over “who can become enterprise infrastructure faster”

Harness has finally become the main battlefield everyone is fighting over

FYI: taste is an intuition that grows naturally

First you need to have some really good experiences; second, you know what’s good and what’s bad — this is something that naturally derives from your experiences

I think you should first live well, and then properly perceive what you want to express

You first have to live richly enough to have something to give to AI

Our lives now are almost wrapped up in algorithms. The content you watch

Our lives now are almost wrapped up in algorithms. What content you watch, who you meet, what you buy, where you go — it’s all recommendation systems making decisions for you

This thing is very comfortable, but also very dangerous

Because it keeps narrowing your world

You like watching AI content, so it pushes you more AI content. You follow a certain type of person, so it pushes you more people of the same type. Your information cocoon gets thicker and thicker; you think you’re seeing a very big world, but actually you’re just spinning around inside an increasingly refined little bubble

It’s quite interesting now — the information age

I think it’s quite interesting now — the information age. A lot of things have indeed accelerated, just like in the email era, when mail got replaced. But would you say it’s really been replaced? I don’t think so — there are still plenty of people who choose to handwrite letters, because, how to put it? You send someone an email, you send someone a message, and you handwrite a letter to send to someone — the weight is completely different. Because you’ll find that in this age of infinitely accelerating information, the slower the rhythm, the more attentively you do something, the easier it is to actually move the other person. It’s a kind of proof of presence. Because attention is being diluted, people tend to value more whether the other person respects their attention, rather than just casually diluting their own attention and casually sending out some stuff, right? So presence carries more and more weight

I now even think this repo, for AI

I now even think that this repo, for AI, should be something it can perceive — how this directory structure should be designed, and how it should call a particular file under a particular directory, it should have a clear perception of that. So I think for the PC side, if its directory structure doesn’t match common sense, there should be a README note reminding that this is how the project structure is designed. And I think it should still follow as good a directory structure as possible, so that AI can summarize and infer

Building products for people and for agents are different

For people it’s about going safe

For agents it’s about efficiency

In that case, its Subagent

In that case, I feel its Subagent is very limited in what it can do; its autonomy isn’t that big, because its skills can’t be invoked very flexibly

The design philosophy of a Subagent is trading autonomy for determinism. Its core value isn’t “being able to do a lot of things,” but “doing one thing predictably, isolatably, and reusably”

But this limitation brings a very important benefit: behavior is predictable. You define a security-reviewer, and you know that every time it starts it carries the same security review knowledge, and its behavior won’t drift because the parent session’s context is different

Put the task back in the main session and let the main Claude handle it with full Skill discovery ability

Or use Agent Teams, letting multiple agents collaborate, each with different specialties

In terms of compile speed, bun feels like the best fit

Sometimes don’t underestimate this compile speed

If it’s running multiple agents, compile speed is very important, so use bun in projects as much as possible

Avoid –dangerously-skip-permissions in team projects

Avoid the –dangerously-skip-permissions parameter in team projects

Use the /permissions parameter to pre-allow some common bash commands I know are safe in my environment, to avoid unnecessary permission prompts. Most of this configuration has already been committed to the .claude/settings.json file and shared with the team

Ralph loop is a great fit for things with a clear success standard

Ralph loop is very well suited to work that has a success standard — like code migration, bulk refactoring, adding tests

What it’s not so suited to is work that requires judgment and design decisions; no matter how many times Claude loops, it won’t converge on a better answer

To get the best out of Claude Code

To get the best out of Claude Code, the most important thing may be — giving Claude a way to verify the results of its work. If Claude has this kind of feedback mechanism, the quality of the final result will improve 2–3 times

Claude uses the Claude Chrome extension to test every change I make to http://claude.ai/code . It opens the browser, tests the user interface, and keeps iterating until the code runs correctly and the user experience is good

Harness Design is being hotly discussed

Harness Design is being hotly discussed not because the concept is new, but because models are now strong enough to have forced the infrastructure shortfall into the open

The core contradiction has shifted from “the model isn’t strong enough” to “the system can’t carry the model’s capability”

What agents lack most today isn’t more prompt tricks, but three layers of infrastructure: identity, substrate, verifiability

Without identity there’s no continuity, without substrate there’s no accumulation, without verification there’s no economic system

Harness should not be understood only as “the shell that controls the Agent,” but more as the soil in which the Agent grows and settles

The point isn’t to rein it in, but to make capability inheritable, collaborative, reusable

A good Harness should let the Agent be stronger, steadier, and run autonomously for longer, while letting the human gradually exit the loop

The ultimate direction isn’t “commanding the Agent more conveniently,” but “needing fewer humans watching the Agent”

The execution layer will gradually be swallowed by models, but infrastructure layers like time, relationships, and accumulated history will not

What’s truly valuable long-term isn’t just scheduling execution, but continuity and relationship networks

The bet of building a product: users need a complete experience, you win on execution

The bet of building a product is: users need a complete experience, and you win on execution, taste, distribution

The bet of building a protocol is: the endgame of competition in the agent era is interoperability, and you win on the right to set the standard

Why is the agent field especially prone to making protocols matter more than products?

Claude Code is a very clear case. On the surface it’s a product (a CLI coding agent), but its real moat is being the landing testbed for MCP + deep integration with the Claude API. What Anthropic is really betting on is MCP becoming the de facto standard for agent interoperability; Claude Code is just the first powerful demonstration of that standard

But the product layer and the protocol layer aren’t necessarily separate — for harness, for example

Traditional product logic: I help users complete task A

Harness logic: I define the execution structure of task A — input spec, tool-calling conventions, output format, fallback strategy

Harness lets task A be driven by any sufficiently strong model, rather than being bound to one specific implementation

OpenClaw’s three-layer skill disclosure system, pluggable backend, provider-agnostic routing — it doesn’t block any particular model from winning, but bets on the value of the execution structure itself. That’s protocol thinking wrapped inside a product shell

For tools designed for humans, the boundary is a question of cognitive load. You don’t want users to think too much, so you hide complexity and give them a simple entry point. The clearer the boundary, the less effort for the user

For interfaces designed for agents, the boundary is a completely different thing — it’s a semantic contract

An agent doesn’t need “simple,” it needs explicit. It needs to know:

What this tool can do, and what it can’t

What it returns when it fails

What my output format is, and how the downstream agent parses it

Is this operation idempotent? Does it have side effects?

What most agent products are doing right now is early instantiation of protocols, without realizing it themselves

Explore which harness structures are effective (execution-layer protocol)

Define communication conventions between agents (interoperability protocol)

Settle which kinds of tasks suit full automation (task-boundary protocol)

The moat should be that the protocol you define yourself is depended on by other agents, and your execution framework is adopted by more downstream tools

In some vertical scenario, make the harness the de facto best practice, and let this harness itself become an interface others are willing to connect to

What product thinking does: help you record this thought

What protocol thinking does: let this thought be understood and used by any future agent

The former’s value is in the present, the latter’s value is in compounding

Ralph Loop is great for iterating by version

It’s best to design the first version of the PRD, and the later versions of the PRD

Getting the PRD wrong is very costly. If the direction of story #1 is off, the following 19 are all built on a wrong foundation. Ralph won’t tell you; it just executes. So the architectural conversation up front can’t be skipped

graphify did a direct test: on

graphify did a direct test: on the corpus of Karpathy’s repo + 5 papers + 4 images, reading the graph/wiki layer used 71.5 times fewer tokens than reading the raw files, and the quality was better (because the relationships had already been extracted)

but ,,, the tokens for building the wiki up front are the exploding part …

and that only counts reading, not writing ….

Humans should mostly stay in the Wiki layer

But the reason is different from the LLM’s — not because of token efficiency, but because of cognitive load

Raw is an article you clipped one day, a tweet, a PDF. It has a few problems:

The context is already lost. The article you saved three months ago — you no longer remember why you saved it, what it’s related to, or how you thought about it at the time.

Fragmentation. Raw is a pile of isolated points with no lines connecting them.

The readability is the original author’s readability, not yours. The article is organized according to the author’s logic, not according to your thinking structure

This is why most people’s Obsidian vaults eventually become a digital trash can — raw files pile up more and more, but they never get looked at

The value of the Wiki layer is greater for humans. Every page of the Wiki layer is knowledge you’ve already digested, written in your own language, organized according to the structure you care about. That’s what’s actually readable

Karpathy actually hints at this in the gist too: he says the human’s job is to “curate sources, direct the analysis, ask good questions, and think about what it all means” — all of these activities happen at the wiki layer, not the raw layer

There are some differences between the water festivals

Lao people seem to wear more traditional skirts

Thai people are super conspicuous; basically a lot of them are in floral shirts

Chinese people are a bit more refined and influencer-style, basically more sneakers

Vietnamese people are plainer and more low-key, a group that doesn’t stand out

Creating is also a kind of learning

Learning in the act of creating

Don’t do learning that is meaningless and has no action attached

flomo sells a way of thinking and a usage habit

memos sells control over a piece of infrastructure

flomo’s moat is culture, not technology, and the cultural moat is being eroded

The card-note method, Luhmann, the restrained aesthetic of being anti-Notion. This narrative was very effective in 2020–2022, because at that time “note-taking app anxiety” really existed, and users needed someone to tell them “you don’t need that many features”

The fragile points of its narrative:

It depends on the founder continuing to produce. Once Shaonan reduces his content output, the product’s cultural gravity decays. This isn’t a criticism; it’s a structural problem faced by all founder-IP-driven products

The cloud service + subscription business model is getting harder in an environment where awareness of “personal data sovereignty” is waking up. Users increasingly ask: my data is on your server — will you go out of business? Will you raise prices? Will you get acquired? These questions have no technical answer, only a trust answer. And trust gets consumed

AI is directly hitting flomo’s core use case. One of flomo’s values is “accumulate fragments; look back after a long time and be surprised.” But now users are starting to use ChatGPT to directly process fragments, and AI to do rumination and association — this need now has a more immediate substitute path, and flomo’s time-value proposition has been diluted

but maybe in the future Memos could be integrated into a homelab / self-hosted ecosystem

A tool with extremely strong extensibility

I think a tool with extremely strong extensibility will definitely also have an absurdly fast iteration speed. For example, if you want to design it as a “less is more” tool, then the question you have to think about — a very fatal question — is that your iteration speed has to be very, very slow, and you have to very, very much consider the user’s feelings. But if you want to build a tool with very strong extensibility, you won’t have that problem. You just keep trying, you just keep designing, and users will even give feedback, give you feedback, offer you more first-hand experience, and let you keep iterating and growing

From a human’s felt experience, you don’t necessarily need tags

Actually, from a human’s felt experience, you don’t necessarily need tags, because do you think tags are really that different from titles? A title is a simple summary, so I think tags aren’t necessary. Tags serve people, not AI. But what do they serve people for? They serve people’s retrieval — they make it convenient for people to trace back their notes, to search; essentially it’s a search process, it’s just the concretization of a search. But I think if we go back to first principles — since it’s designed to serve people in searching, in tracing back — then I think, whether this part, if it were reconstructed by AI… what can AI do? What AI can do is that it should serve its title; the title can actually satisfy that too. A title — you can define the title yourself, or you can let AI come up with titles or summaries, right? But this way still essentially serves to abstract a level, a one-sentence level, and this level also makes it convenient for the LLM to quickly trace back and locate this content block. I find this quite interesting

After I came here, I found that Laos’s water festival is underrated

After I came here, I found that Laos’s water festival is underrated. I feel there’s a very primal life force here, and it’s not that commercialized

And such a contrast with their culture — they’re normally a very reserved bunch, and then during this festival everyone moves their speakers to the roadside, the whole city is an electronic music festival, a water festival

Raw and Wiki give different answers under different scenario questions

Raw and Wiki give different answers under different scenario questions

For example, if you look at a travel guide, Raw might be a 3000-word travelogue, while the Wiki page might only need a few elements, like the name of the attraction, a rating, the reason it’s recommended, and the source. You can make its Wiki highly compressed and zero, because you don’t need that content when you’re looking it up. But I think, for legal and medical documents, Raw is the original contract text, and Wiki needs a summary, plus the original text of the key clauses. So the precise wording of the original is itself information, and it can’t be replaced by a summary

Core summary

[LLM-synthesized 2-3 sentences answering “what is this”]

Key facts/attributes

[structured bullet points that can be quickly scanned]

Related

[links pointing to other Wiki pages]

Source

[pointing back to the specific location in Raw, with anchors]

Wiki and knowledge graphs are kind of interesting, you can’t say which is better

I think Wiki and knowledge graphs are kind of interesting, you can’t say which is better or worse. But when LM handles Markdown, what it’s actually doing is linear text understanding. A lot of relations are in natural language, and the LLM parses them itself — there’s a spatial relationship between A and B

Then when the LLM reads a knowledge graph, its relations are actually explicit, it’s a formula, the relations have already been extracted, so the LLM doesn’t need to do that parsing

{source: “attraction A”, relation: “walking_distance”, target: “attraction B”, weight: 15}

But the problem is LLM is not a graph database

LLM’s reasoning is done on token sequences, not on graph structures.

If you give it a graph.json, what it sees is still a pile of text. It can’t natively “traverse the graph” — it can only read the JSON, and then simulate graph traversal inside its own attention mechanism.

Graphify: raw file → [LLM extraction] → graph.json → [LLM reads JSON] → answer the question

Wiki: raw file → [LLM extraction] → markdown → [LLM reads markdown] → answer the question

Both compile first, then query; the key difference is in the intermediate format

I think the graph structure is very structural — if A and B have some relationship, this is very much in favor of the graph structure answering it. But if you ask what the core argument of this paper is, then I think the graph structure is more awkward. This thing, I think it doesn’t have — technically there’s no absolute good or bad, but for different scenarios, the premise is you have to know what your scenario wants

So Wiki actually preserves narrativity — the Wiki pages humans write have chains of reasoning, have background, have why. And these things are hard to express in a knowledge graph; you can only store nodes and edges, but you can’t store the narrative of which historical background this decision was made in. When doing RLM, in the process of complex reasoning, the narrativity of context is definitely very important

For scenarios with dense structured queries (codebases, paper relations)

For scenarios with dense structured queries (codebases, paper relations, entity relations) → knowledge graphs are more efficient

For scenarios requiring deep understanding and narrative reasoning (personal knowledge, research reviews, decision records) → Wiki is more efficient

Google folds NotebookLM capabilities into Gemini

Google further folds NotebookLM-style capabilities into Gemini: the project-based AI workbench is taking shape

Ah, I feel like this thing could later be used with NotebookLM to do some deep research, I guess. Because NotebookLM essentially has a flaw: it’s only suited for researching and learning based on existing documents, but it isn’t suited for deep learning. For example, if something isn’t in the document, it has no way to go deeper and explore. So I think that’s a problem too, so I think there’s still room to do some optimization

The scenario for experience-type products is

The further AI-generated things move forward, the farther users are from their own experience

Ambient

Very interesting, a cross-domain concept word, but interesting in every domain

Existing in the surrounding environment, diffusely present, an unobtrusive background presence

In music, Ambient sound = background sound that naturally exists in the environment, rather than sound deliberately made the focus

Including wind, rain, the murmur of crowds, the background sound of a cafe

Ambient music = music that can be ignored or focused on (as ignorable as it is interesting)

In computing, it’s also a kind of environmental sensing

Ambient computing = technology recedes into the environment, continuously sensing and responding without the user actively operating anything

This is a kind of sensing without obvious interaction, but still with contextual awareness, and it can keep running

I think there was a screenshot tool before that did it this way. I think this should count as a very promising direction in the future AI era. If I were doing research, I think based on my judgment it’s like this

Chat agent passive (only answers when you ask)

Tool agent triggered to execute

Ambient agent actively senses + keeps running

The designed UX

Ambient UI = not

interrupting the user, an interface that conveys information in a “very low presence” way

At the system level, Ambient Agent System is an environment-type intelligence system

While you’re still judging, you should sink the core down to the protocol layer

While you’re still judging, you should still sink the core down to the protocol layer or the engine layer

Because I feel like it’s harder to extract a core layer than to build a core layer from scratch, especially for a document machine’s data model — once it’s wrong, the cost of switching is very high. But I feel, in terms of user perception, it’s best to be a very lightweight tool for users, and then when actually using it, the experience will be good, it shouldn’t be too heavy, because if it’s too heavy it becomes very complicated

chatcup generates frame-by-frame descriptions and commentary

For example: one frame is described as “a person typing code”, another frame is “red error text appears on the screen”

Actually their current editing logic is nothing more than trying every possible way to convert video into text descriptions, and then processing those text descriptions through a series of operations, including things like modifying code, or writing documentation — that’s how they process the video, process the audio clips. So although what we see is the video frame by frame, it’s actually text-driven nonlinear editing

They also have some speech recognition layer, and a vector retrieval layer. It stores the converted text into a vector database, and in this way quickly locates the content of the video frames. And it also splits the video frame by frame, specifically how many seconds per frame of video, and then the description corresponding to each frame, and edits the video this way

Claude is smart too

Integrating all capabilities into one client platform …

So what are the other products supposed to do

In a few stages, he’s skilled at using Opus

I think in a few stages, he’s skilled at using Opus. For example, one is in environments with high uncertainty, high context load, and high judgment density. Then it depends on the complexity of the task, including cost tolerance. This actually has boundaries, and whether the boundary still depends on a kind of intuition. For example, when you feel it’s more suited to doing API design, when you feel it’s more suited to doing code refactoring

I feel that when the requirement itself is vague, a good model can also help you fill in this boundary. And if more files, more modules, more historical context are needed, then a better model can also handle it better. And it’s not just nailing a single point, but understanding the whole chain of design, change, verification, review

You can tell it, when you have intuition, you can tell it — think a bit more, this task looks hard — and it will mobilize its thinking model, it will actively and smartly think about this problem

And I think now AI actually, I think a reasonable usage experience shouldn’t be treating it as a tool where you give it a clear requirement and it executes. That way it can’t bring out its creativity, or its ability to execute tasks. Rather, you treat it as an engineer, treat it as a designer, and you give it a requirement, and let it complete it itself. You’re the client, it’s the contractor

The Claude team gave me some inspiration: they push iteration to the limit

I think the Claude team gave me some inspiration: they push iteration, they push the update speed to the extreme — it’s like they’ve captured the focus of the whole of society. And I think they haven’t gone through some version design protocol; they should be driven by faction versions, i.e. CI — every time they merge into the main branch, they cut a release, bringing the psychological cost of cutting a release almost down to zero

So I think this kind of workflow seems very suitable for some small teams or startup teams in the AI era. And then they also use large amounts of GitHub Actions for CI automation, and they also use Claude Code Action to handle the corresponding issues and review PRs

I think what we can do is unify the requirement entry point to one place as much as possible. Or make the format of requirement creation as unified as possible, but anyway, no matter where you create the requirement from, you can get a very unified format and standard. And it’s best to have an AI that automatically analyzes for you every day — for example, what requirements you listed today, and you submit the requirements, maybe 20-something, 10-something of them. Then the AI analyzes some of these requirements every day, then does a PRD, then combines the PRD into JSON, then through the JSON does a very simple task about running at night. Then design an engineering capability that automatically reu for you at dawn. Then when you get up in the morning, you automatically cut a release, and you test a few things yourself, then you propose some features. Every day, as much as possible, make yourself the human-centered part — integrate this workflow into your own life, into your own daily routine

You’re thinking — since certain structures can be said to be for people

Actually you’re thinking, since certain structures can be said to be for people — for example, extracting some repeated experiences from a person’s daily life and summarizing them into a structure. Then can you also imagine that these structures can also be used by AI? If in the future AI also reaches a state where it can operate continuously — that’s of course a very good form as we understand it. So how can we get it to do this task in a controlled way, or in a traceable way? I think that’s where this kind of tool is needed

Recommendation works for things less precise, less immediate

Recommendation can be used for some more general scenarios, ones that aren’t so precise, not so immediate

Understanding information entropy for context

“What need does a person have right now” really is a high-entropy variable — determined jointly by mood, scenario, history, and the current focus of attention, and almost impossible for the outside to predict reliably. What AI does relying on context and “intelligence” is essentially using a noisy prior to compress a high-dimensional distribution; it can reduce entropy, but the ceiling is low

Whereas when you type out a question directly, you turn the sender (yourself) directly into the information source, the channel is lossless, and what the receiver (me) gets is the explicit intent you encoded. In this process the entropy loss isn’t achieved by inference, but by yourself completing the encoding from need to expression

From a human engineer’s perspective, multiple repos are manageable

From a human engineer’s perspective, multiple repos are manageable. From an AI agent’s perspective, multiple repos are opaque:

Can’t see the complete picture

Can’t reason about cross-service impact

Can’t run integration tests locally

Each tool handles only one stage; no tool tries to do everything.

Each tool handles only one stage; no tool tries to do everything. The daily loop forms a closed cycle: error detection → triage → fix → verify, with human involvement minimized

Statsig: Feature flag, every feature by default behind a gate, gradual rollout, kill switch without needing a deploy

Graphite: Merge queue, PR stacking, only merge when CI is green

Linear: AI automatically creates tickets, with severity score, log samples, suggested investigation paths

Two kinds of engineer roles

Architect is responsible for designing the SOP, teaching AI how to work

The one building infrastructure, integrating systems, the triage system

The one defining architecture and system boundaries, whose core ability is to critique AI, not follow AI

The ability to critique AI is more valuable than the ability to produce code

As for the Operator, it’s more about doing bug investigation, UI fine-tuning, PR review and verification

Junior engineers adapt faster than senior engineers

AI assigns tasks to people, not people assigning tasks to AI

The task doesn’t require architectural reasoning, but requires skills and attention

Junior: fewer habits, the sense of self is amplified

Senior: two months of work done in one hour, hard to accept

Professional salespeople (so-called “lecturers”) use cultural information

Professional salespeople (so-called “lecturers”) use a pitch combining cultural information, ecological misinformation and medical advice to package expensive illegal wildlife products (such as “Angong Niuhuang Wan” advertised as containing rhino horn, and bear oil) as miracle drugs that can cure and save lives, thereby inducing foreign tourists to buy at high prices

Vientiane, Laos

For a while, or in some cities, there’s a strong rejection of people with stories

Because their stories all have a structural familiarity — the exotic-hunting kind, the inspirational kind, the artsy kind … they all seem to carry some personal performance

But here I’m moved by some simple and pure stories

I’m at the high-speed rail station right now, Vientiane’s high-speed rail station

Out in the world, it seems that toward Chinese people, my own willingness to engage isn’t very high

But in Vientiane, by chance I got to know a group of artists, this time the topic is wildlife protection

Wildlife protection artists — artists who sacrificed their own art to protect globally endangered wildlife; their attention is on relatively big social meaning and issues, and on creating better works

A boss who has given local poor kids a free plate of egg fried rice for two months straight — ps, even though the boss and I clash a lot hahahaha, the boss is an extremely strong empiricist, I’m a theoretician, a clash between the two, but on this matter it especially drew me in, he’s a very interesting person

And I met a local resident engineer, but he has also, for 8 years straight, given 600 yuan in tuition every month to a pair of poor twins at the place where he previously worked

Rewrite the time discounting function, rather than endure

Bring future gains into the present

Fitness → don’t look at physique, look at “completed today’s training ✔”

Writing → don’t look at monetization, look at “published today ✔”

Learning → don’t look at a few years later, look at “understood one point today ✔”

Lower the frequency of decisions, fix what you do, set up environments and triggers

You can increase some positive dopamine feedback

Long-term returns are very slow, so building a brief, sustained immediate satisfaction is very important

Replace the “sense of results” with a “sense of progress”, return to the process, and long-termism happens naturally

Is the purpose of posting to Moments really just to seek approval

It feels like that’s just the vehicle? Are there some more primitive things?

Humans are the only animals that give their experiences a narrative

Something happens, and if you don’t say it out, it’s unfinished in your perception. There’s a concept in psychology called the Zeigarnik Effect — unfinished things keep occupying cognitive resources. The essence of posting to Moments is closing off an “unfinished experience”.

Eat a good meal and don’t post it, and in a sense that meal “isn’t over yet”. Once you post it, it’s complete, and you can let go

Expression is a basic human need

Humans are the only animals that give their experiences a narrative

Something happens, and if you don’t say it out, it’s unfinished in your perception. There’s a concept in psychology called the Zeigarnik Effect — unfinished things keep occupying cognitive resources. The essence of posting to Moments is closing off an “unfinished experience”

Eat a good meal and don’t post it, and in a sense that meal “isn’t over yet”. Once you post it, it’s complete, and you can let go

The audience of Moments is other people, so its positive feedback comes from external approval. But this approval is unstable — anxious if no one likes it, hurt if comments are bad

Being understood by your future self is actually more stable than being approved by others, because this witness doesn’t disappear, won’t judge, and will only understand you more and more

V4’s superpower: a brand-new attention mechanism with positional encoding

V4’s superpower: through a brand-new attention mechanism that includes positional encoding, it maintains a 1 million token context while greatly reducing computational overhead

The traditional Transformer attention mechanism, when processing text, requires every character (token) to “attend” to all the characters before it. This means that if the context length increases 10 times, the computation explodes 100 times. That’s the fundamental reason long context is both slow and expensive

It uses a hybrid attention architecture

V4-Pro’s single-token inference computation is only 27% of the previous generation, and the KV cache (which you can understand as the model’s “VRAM memory”) consumption is only 10%

V3.2 introduced DSA (DeepSeek sparse attention)

V3.2 introduced DSA (DeepSeek Sparse Attention), to fundamentally reduce the computational complexity of attention, solving the efficiency bottleneck in long-context scenarios while keeping the model’s output quality basically unchanged. arXiv — this direction is substantive innovation, not parameter stacking

V4-Pro (1.6T parameters, 49B activated) and V4-Flash (284B parameters, 13B activated), both support 1M token context

All previous leaks said 1T parameters, but it’s actually 1.6T, and the activated parameters also went up from the predicted 37B to 49B

The core is Hybrid Attention

The core is Hybrid Attention Architecture, combining CSA (compressed sparse attention) and HCA (heavily compressed attention). In 1M token context scenarios, V4-Pro needs only 27% of V3.2’s single-token inference FLOPs, and the KV cache only 10%. Hugging Face — this is the key to making 1M context actually runnable: not simply widening the window, but driving down the attention computation cost

One of the core drivers of RAW→wiki was “it won’t fit”

So one of the core drivers of RAW→wiki was “it won’t fit” — a 128K window simply isn’t enough for thousands of notes, so you have to compress, distill, and structure first before the model can see your knowledge base. Now with 1M tokens plus V4’s KV cache at only 10% of V3.2’s, this “survival compression” need has indeed dropped a lot

But there’s a problem. The first problem is that a large context ≠ linearly improved effective processing capability. 1M tokens is the nominal window, but attention’s recall in the middle of long sequences still decays — the “lost in the middle” problem hasn’t been thoroughly solved, it’s just been pushed back. 3000 memos is roughly 1 to 1.5 million words, already close to or exceeding 1M tokens, and it’s all unstructured raw stream — the model “seeing” and “effectively reasoning” are two different things

The second problem is what exactly wiki compilation is solving. If it’s only solving the problem of context not fitting, then it really is less important. But it’s also solving another thing: signal density. Among 3000 raw memos there’s a lot of repetition, emotional fluctuation, and context-dependent content, and stuffing it in directly gives extremely high noise. Structuring isn’t just compression, it’s doing curation — making scattered patterns explicit

The third problem, and the most critical: is your goal “the model can access information” or “the model can understand you”. These two things have completely different requirements

After 1M context, you can do something you couldn’t do before: inject large amounts of raw memo as a grounding layer, and then let Soul Core serve as the navigation layer. It’s not either-or, but layered — Soul Core defines “who I am”, raw memo provides the detailed evidence of “what I’ve experienced”

The worktree boundary = the Git change boundary

The worktree boundary = the Git change boundary The terminal / CCA boundary = the task execution boundary

What worktree serves should be certain tasks that are committed separately, whose changed files basically don’t overlap, and which can open PRs to each other, exploring different solution directions

But I feel its compilation still brings some new thinking

But I feel, I feel its compilation still brings some new thinking, or new habits. For example, it used to be a way of compiling metadata files, and the original drive might have been a kind of spatial anxiety, like it can’t fit, so compress it, and then compile it once through files, and link it once, and then it has backlinks. Then articles connect easily to each other in a chain. And in the process the LLM does, it can also handle it through some agent capabilities.

And then another thing I feel, it still — it has solved three problems, it really solved three problems — one is the density problem, that the signal-to-noise ratio of the RAW stream is very low; among 3000-plus memories, the content that really constitutes very effective content might be 20%, and the rest may be repeated emotional streams or fragments or some context-dependent content. But you’re thinking what LM itself sees isn’t equal to being able to extract some very valuable structure from this noise. But that’s obviously — some do compilation, and they can do this kind of processing

The second problem is that RAM’s shape is a time series, it’s linear, and it decays. When we usually understand a person, the shape we need is a relationship graph — for example, the tension between beliefs, the repetition and imitation between behavior patterns, the core contradictions, how they evolve, how they connect. And these two shapes seem naturally mismatched. So many — no matter how much context, it’s hard to solve the shape problem. But RAM’s shape is very meaningful, it’s linear, it decays, because a person’s attention is linear, a person’s energy is linear, a person’s narrative is also linear, which determines that a person in this current world, their time must be a linear sequence, but a person can understand it very simply

And then I think there’s another point, maybe a navigation problem. For example, when we stuff tokens in, where should the model be looking? If there’s no navigation structure, its attention in the middle position decays. The lost-in-the-middle problem won’t disappear, it will only be pushed back

There’s still a very important part, about design

The question of how the relationship between raw and wiki is stored, and how it’s defined

Summarizing is compressing information; compiling is discovering things that aren’t explicitly present in the information

Turning “building scaffolding for the model” into letting it grow its own

It reverses the thing of “building scaffolding for the model” into “letting the model grow its own scaffolding”

Model capability → tool encapsulation → Harness orchestration → Agent behavior

Each layer helps the model grow

But there’s an interesting line

We’ve long used engineers’ certainty anxiety to patch the model’s uncertainty — but the model’s uncertainty may itself be the source of its capability

Be brave and do subtraction, not addition

Why the name “Dialect” is accurate

Garry Tan calls it Dialect, and I think the word is chosen very precisely.

First generation: Prompt (single-turn instruction)

Second generation: Chain (chained calls)

Third generation: Harness (tool orchestration framework)

Fourth generation: Dialect (the private interaction language that emerges between model and environment)

It’s really talking about a process of emergence; it’s not designed, but a model that gets run out

The model, in some environment, keeps trial-and-erroring, and then figures out a way the environment needs. This set of ways isn’t standard, nor is it encapsulation written by humans; it’s the model’s own dialect, i.e. Dialect

Humans put a door and a food trough on the birdcage, and the bird does survive, but it forgets how to find food. Harness is the cage, Dialect is tearing the cage down

So in the era of subtraction, the difference between people migrates from who has the stronger toolchain to

Judgment — knowing when to trust the model and when not to Taste — knowing which tasks are worth letting the model play freely, and which tasks must be constrained Problem selection — knowing which environment to put the model into matters more than knowing how to encapsulate tools

Their core shift: don’t start from “defining all possible actions”

Their core shift is: don’t start from “defining all possible actions”, but from the opposite assumption — the model can do almost anything, and then go constrain it

Old: whitelist thinking → list what the model is allowed to do

New: blacklist thinking → list what the model is not allowed to do

I think this way: if we’re designing environments for the model

Actually I think this way, I’m thinking if we design environments for the model, then to a certain extent shouldn’t we design more interfaces for it, design more tentacles, let it touch real situations. Because it’s also an environment. This environment I understand as people — a kind of friction between the model and people. Because people, the model also understands them, it receives people’s context, it feels all of a person’s current context, all of their current — including their own thoughts, including what they want to do, including their goals, including their company, including their reasons and so on. It can be a task, it can be an issue. And then I think putting these things together, what it supplies is an environment of its own

What we’re designing is really just a few boundaries. One is what the model can see — this environment determines whether the model has some interfaces, and whether there are some good things between them to reduce friction. That’s one aspect. Then I think another is what it can move — some models basically all have permissions now, some bound to your device, and some through some secondary security keys

AI — sometimes I’m thinking about the boundaries of AI

#AI sometimes I’m thinking about the boundaries of AI, it’s very hard to connect it all together. I think its product will have a few very, very important parts that can’t be underestimated. One is it, like a newest applied technology. It will emphasize that all the previous stuff doesn’t matter, your previous Harness doesn’t matter, your previous system tools don’t matter. Harness is nothing more than managing your model’s context, it’s nothing more than clarifying your model’s boundaries. But I think it now — it now has some new things, for example it will tell you, you should cut away as much as possible, that is, I feel I don’t know what its difference from Harness is. Because Harness itself is also partly about you designing an environment. But now this dialect still has a kind of paradox, because it still depends on the environment — in this environment, you might have previously emphasized what the model can do, but now you emphasize what the model cannot do, because actually you’re constantly refining something like your constraint. Just like when we write in Claude, in Claude’s meta, we don’t write what you can do, instead we all write, you can’t do it this way, you should do it this way. So I think maybe for models now, telling it what not to do always works better than telling it what it should be able to do. Things it can do should be left for it to fully unleash its imagination and creativity. You just need to design a really great environment for it, and the model can, so to speak, manage its own boundaries and break through its own boundaries in that environment

AI — in this process, what I think about is

#AI actually I think in this process, what I’ll think about is, I think people may be very important in the future, in the AI era, because you know? Even if you’re one person, you can — I don’t know how many agents you can manage, I don’t know how many contexts you can manage, how much very erudite knowledge you can have, how many AIs you can have, how many employees you can have, I don’t care, what do I care about? I care about whether this person has a way to manage such a person, to let himself naturally undergo some transformation. I’m thinking, in the AI era, what is its person like. I’m a bit at a loss right now, I think the person, he should first of all be a person, and then what can AI do? AI still serves him; as a person, his basics, he’ll output some context, he’ll have some thoughts, he’ll learn some knowledge. But a very key point is, what the model should manage isn’t how much knowledge you’ve learned or how many notes you’ve taken. I think what it should more necessarily manage is when, and why, you learn that knowledge. It may be a person’s intent, because I think there are some things about a person that are limited by the world he’s in. Even if AI is very awesome and developed now, and will get more and more awesome and developed in the future. But I think a person’s timeline doesn’t change, and a person’s perception of time won’t change either. Because a person only has 24 hours a day, sleeps, eats for 7-8 hours, right? Eating time can’t be compressed either, right? A person’s time is very continuous, very fragmented; the fragmentation may depend on what else this person is doing at the same time or at different times. When he does things or thinks about things he has an intent, and can also have no intent; I think it’s all fine, even without intent he might also — the person at the very start just wanted to space out or something; the intent is there anyway; it’s just a descriptive tool, it’s just, it isn’t an entity, it’s a fiction used to describe a certain state; intent, it’s just. You yourself are drunk, you want to — why do you want to drink? You drink — you want to — what do you say after you’re drunk? Or after you’re drunk you involuntarily said something?

AI can quickly return to previous context information

And it’s information in compiled state

This information may be information that the AI encapsulated at the time

The world is getting richer and richer

More and more information

But a person’s time and energy are limited

A person only needs a very small portion, but the portion most helpful to himself

AI imperceptibly understands yourself, understands this world

At the right time digs into yourself, gives out content you can consume

The world is a cactus ball, each of us is a carrier on our own spherical surface

In the AI era, the relationship between agent and human should be

Complex things the AI simplifies for the human; simple things for the human the AI executes in a complicated way

Institutions change the rules, but a person’s fate is more the result of the system

Institutions only change the constraints, not the initial conditions

The sex industry is legal (institutional change)

But women still enter this industry

Their alternative options are still no income, extreme poverty; they’re choosing survival

Once you enter this industry, it’s very hard to get out — this is path dependence

No education → can’t find other work

Has an industry label → rejected by society

Income structure fixed → can’t transform

Many people don’t know their rights, have no ability to defend them, are controlled by middlemen and organizations

Institutions are only a necessary condition, not a sufficient condition

Some tools now use methods for time-sensitivity

For example, hooking up the latest documents via some API or RAG system

or web search for the latest data sources

The metacognitive layer, letting the model know what it doesn’t know — this is a harder problem

Explicitly marking the knowledge cutoff date

Training the model to proactively express uncertainty on time-sensitive questions

Injecting the current date into the system prompt so the model can compare against its own knowledge boundary

Timestamp provenance: every piece of knowledge carries its acquisition time; past a limit it automatically triggers re-verification

The older the information, the lower the confidence, triggering a fallback to search

Setting mandatory verification checkpoints for high-frequency-change domains (finance, law, medicine)

One model generates the answer, another specifically challenges its time-sensitivity

Leaving Siem Reap Tomorrow

Heading to DON KHONG on Si Phan Don

A bit reluctant to leave this city

Maybe I’ll never come back

But this place, this city, this memory

really will always exist in my heart

Goodbye Angkor Wat! Goodbye Siem Reap

Still Got Held Up at Customs, Heading North Today

It still felt like I got held up at customs, and overall today coming up to the North side — because I came over from Cambodia, you know — some of the border crossings on this side are close to Cambodia, and they target Chinese people, they just keep dragging things out for you. Because we happened to run into a lot of Chinese people at the time, they figured all Chinese people have a strong collective mindset, so they just keep stringing you along and they can skim more money. In the end they kept us hanging for over two hours at the start. At first it was 50 dollars and we thought that was too much, 50 dollars was way too much. Then they kept adjusting it, all the way down — down to 20 dollars, then down to 10, and finally down to 5. I thought it was pretty funny. In the end we just kept holding out, and I left when it hit 5 dollars. Actually I think I could have held out longer, but a lot of their family was waiting for me, so I didn’t keep going. There was another Chinese guy behind me who held out a bit longer and got it down to 2 dollars. God, I really think they go too far

How a Country Leaves Its Impression on You

How should I put it — overall, it’s the impression a country leaves on you. Actually I think the main thing is the impression its customs leaves on you. I felt this before in Vietnam too, in Ho Chi Minh City: because we landed very early, they just kept holding you there, not letting you in, dragging out the time — they wanted a tip. I think that’s just too much, right? So at the time I formed a really bad impression of Vietnam. I felt they just kept stalling me, and it was so worthless — wasting everyone’s time for a little bit of money, with no good rules, no good system, no better system designed to make circulation more efficient and make relations between people much more collaborative, which would also raise your goodwill towards the country. I remember in Singapore you could feel their customs was very efficient, very streamlined. They wouldn’t hold you up because of who you are; they treat things very fairly. They wouldn’t think you’re easy to fleece or anything like that, and they don’t care about that little bit of money, because they think building a good system so money flows naturally is what matters. Rather than damaging the image of your own customs, and the image of your own country, for a small profit. That kind of thing does enormous long-term damage. That’s how I see it

Sometimes It Feels Like an Expedition

Sometimes it feels like an expedition. Tonight, for instance — after you get out of customs, it really feels hard to get out of customs. From Siem Reap, from Cambodia’s Siem Reap, to the Laos side, there’s this place called — what’s it called? Si Phan Don, this one is probably called Don Khong. Coming over, you find you have to change several times. At the very start you have to book a big bus, over in Siem Reap. It goes over very early in the morning; first you have to confirm its location, or it picks you up at the hotel. Halfway you change buses once; when you get to customs, he transfers onto the bus, and the bus takes everyone over together to customs. So that’s one transfer on the bus. At that point I nearly got on the wrong one — I nearly sat down on the return bus. He asked me, where did you come from? I said from Cambodia — pff, I said I came from the Siem Reap side

I think at customs they hold you for a long time, really long, just to get a tip. Once they’re done holding you, you still have to take another bus. Unexpectedly, that bus waits for you the whole time. I thought it wouldn’t wait for people, and by evening you might have no bus at all to get back. But it’s not like that. I think they did that really well — you don’t need to worry about how late it is, even if you tough it out it’s fine, tough it out to the end, right? It’s just everyone waiting on the bus, and it doesn’t seem to make much difference. Even if you found out, it seems you wouldn’t find out much. I think that’s very harmonious. After one bus you still have to cross a bridge, you go to their border point — there’s a pier there, and from the pier you take a boat to Don Khong. I find that pier incredible; the moon was especially big today. I’m still walking on the road right now — I’ve just come off the pier and I’m heading back that way. All along the little path I can hear lots of birds calling. It’s very hot and humid here too. Because I stayed in Guangdong for a long time, this hot and humid weather is OK for me, acceptable. It’s just extremely, extremely humid and extremely, extremely hot — not unbearable. The moon is especially bright today too, which really fits the image of Southeast Asia: the sunset is huge, the moon is huge. You can crouch here waiting for the sunset, which is great. We had to cross a bridge and take a boat once; their boat was very local, a long boat, carrying a lot of us. It was pitch black at the time, and we saw on the surface of the lake just one patch of moonlight. Shining down, shining on the lake surface — I thought it was especially beautiful. There was a Chinese person next to me who said, this is an expedition, we’ve come here as refugees. I thought that was hilarious. But in fact the moonlight — it shone on the lake, and the lake surface was half white, and I thought it was really beautiful. I quite liked it

Actually there are lights along the roadside here, so it’s not like some sections are pitch black with no lights. Relatively speaking, I feel it’s fine here — it’s just walking on a small path, and the way back is about a 20-minute walk. I didn’t take a taxi either, I just walked back. The lights along the side were there the whole way, so it was very convenient

The neighbourhood feels like a bunch of very interesting, very local cafés. Not all that refined, but I find it quite interesting — very comfortable, very full. I think I could come running here tomorrow, right by the water. Though I think it might be too hot

I Don’t Really Know Why I Came Here

Actually I don’t know why I came either. I just felt the world hadn’t reached the sea yet… no, it wasn’t in the plan, I didn’t come for it specifically. It’s just that near Siem Reap there’s a fairly fun place, and it happened to be on my route back, so I thought, rather than just heading back, I might as well come here too. And having come, I feel tonight’s experience was fine — apart from the customs bit. But at night it really had a feeling to it, especially the moment on the boat: I felt like a group of people fleeing, pitch dark, with only moonlight visible beside us, and the greenery beside us, and us rocking on the boat, a vast white blur. I think that kind of experience can’t be planned, and can’t be bought. It’s something you can’t count on — and it counts as a bit of surprise that customs, by holding us up that long, left behind for us. I find it quite interesting

Every Country Has Its Own Awful Bits

I think maybe every country has some very annoying things. For example, in Cambodia you might get your phone snatched, and in Laos I got held up by customs for ages. I also got held up at customs for a very long time in Vietnam before, and in Thailand I ran into customs giving me dirty looks. Because they’re in tourism, they can’t hold you up. So I think, ugh, it’s interesting. But if you seriously live in a place, if you settle in, you find it’s no big deal, it’s all normal, and they’re all very friendly — their very everyday, very local side shows through. I think this is information that’s hard to grasp, hard to sense, from just travelling or just watching some media, because as people we seek advantage and avoid harm. The brain still focuses on information that’s harmful to us, or reinforces signals that feel unsafe. That’s our survival instinct. But I think this also makes our brains produce some bias

So you’d think, ah, this country has some exceptional cases, aimed at this kind of person, and you’d think whatever about them. But actually, once you’ve experienced it, that’s very one-sided

Everyone is just living. Nobody discriminates against you that much. It’s just that certain special positions get constantly reinforced by the environment, and they do some biased things to get their own benefits. For most of the people living here, they don’t have that kind of bigoted environment. Their lives are plain and ordinary, they keep their own lives going, occasionally do a bit of small business. If they do get some reward for certain behaviour, that reward may get reinforced, and the reinforcement produces a targeted phenomenon. But I think — I don’t know if it counts as discrimination either. I think it’s just human, part of human nature. People are just like that, I think it’s just how it is; people are still products of their environment, and if the environment is like that, then it seems entirely reasonable

Belgians Are Interesting Too; They Don’t Do the Rat Race

I personally feel Belgians are quite interesting too, they don’t do the rat race either. They’re not as outgoing as the French, nor as serious and reserved as the Germans; I feel they’re somewhere in the middle. There’s also quite serious division inside the country. I think they enjoy life quite a lot, relatively speaking. Drink when it’s time to drink, eat when it’s time to eat. They absolutely love beer, all kinds of drinking culture. Today, meeting these Belgians, I felt like they all smelled of alcohol. I think it’s enjoy life — I find it quite funny

Surprising: Everyone on This Island Looks Familiar

What really surprised me is that I walked around the island today and it felt like everyone was familiar. A group of Chinese people arrived yesterday, and basically I can see them all today, and I can recognise them all — very familiar. Neither side said hello, but it was quite interesting all the same. This place isn’t big, it’s very small. And most of the food places get high scores. I feel that’s because here, if they didn’t all compete hard, they’d find it hard to get ahead. So every place has to compete, and they set the same standard, which pushes the average level up. And this competition isn’t blind — they still bind themselves to an agreement, and their goal is still mutual benefit. That’s what I think

Laotians Don’t Really Do the Rat Race

From what I’ve felt, Laotians don’t do the rat race, not that much of one anyway. In terms of wealth distribution, I feel the gap between rich and poor isn’t as big as in Cambodia. You can see it from the stilt houses here: even in fairly remote villages in Laos, the quality of the stilt houses they use — the timber, and timber is actually quite expensive — is pretty good. Inside they’re moisture-proof, because the humidity and heat here are strong, so some of the timber insulates better and easily keeps damp out. So from this you can see: Cambodia’s whole social system still allows wealth to be distributed like a pyramid, and it also has its own fairly good industrial structure. Comparatively, Cambodia is more dynamic, it’s developing. Which also means the polarisation is fairly serious. I don’t know if that’s a good thing or a bad thing, for society as a whole. In Laos comparatively, even in the countryside, you can see they live relatively well, not that badly. Nobody’s doing the rat race either. Some hostels or some hosts limit the latest check-in time; even in Japan, some allow you to check in after the latest check-in time if you pay extra. But in Laos they don’t allow it. It’s: I rest when I rest — no extra workload for me, thank you

Of Course, Maybe I Just Haven’t Seen Enough of Rural Laos

Of course, it’s possible I simply haven’t been to that much of rural Laos. I just think that if I get the chance I should still feel it in depth. This place probably counts as a small tourist area, a place where Europeans and Americans hang out; there’s some extra net inflow from tourism here too. Not sure. Anyway, that’s my feeling at the moment.

Most People Cycle, But I Hardly Did

Actually most people choose to cycle, but I hardly cycled at all. Even going to the island next door I walked over — I think walking is quite convenient too. Because I have enough time. And for me, it also means I can rest in a café I like

Some Laotian Customs Points Are Pretty Grey

I feel some Laotian customs points are pretty grey too: the more remote ones, the ones outside the reach of the country’s political centre, probably have more backroom dealings. Especially the Laotian customs along the Cambodian border — some of those positions may have been bought with money or obtained through connections, and the salary for the post is very low, so their aim is to skim off enormous profits in the form of tips. That’s one of their channels.

I Went to the Si Phan Don Waterfall Today

Today I went to the waterfall at Si Phan Don. It seems they charge 35,000 for a ticket here

I found it fairly interesting inside, too; there’s also a cable car, a cable car waterfall inside, which needs an extra ticket. The ticket seems much more expensive — it’s 250K I think, 250,000, about 80-plus in RMB

Actually, on the whole this place doesn’t have especially unique scenery; I feel it’s the same as a small primitive village. And I think the charges here are quite unreasonable. It’s a very primitive place in itself, and by drawing tourists in — tourists come here and spend money — it has already boosted the income of the place. So I wonder whether it would be better if it were free? Even without the tens of thousands in ticket revenue, it would gain a lot more positive reviews from visitors, which forms positive feedback. Because tourists are still the pillar of this place’s tourism, so personally I feel that might be better

I think this waterfall is relatively OK, not as spectacular as I imagined. Maybe it’s because I’m from China — China has so many places, whether Yunnan, Guizhou, Sichuan, or Tibet, or Xinjiang — our motherland really has the most top-tier scenery in the whole world. It’s just that their aesthetic sense isn’t great and their management probably isn’t up to standard either. A lot of the scenic spots are very primitive, undeveloped, and they’re protection zones too. So comparatively, it still far beats this place. So by my own taste, I’d give the scenery here roughly a 6. I think it’s far behind anywhere in China, let alone the world-class spots I haven’t been to yet

Laos is like a hidden haven in Southeast Asia

Pakse is in the south on the Mekong, and the life philosophy here is Sabai Sabai too

The north centers on Luang Prabang, once the ancient capital of the Lan Xang kingdom, a city full of mansions left behind by the French colonial period and resplendent Buddhist temples. The morning alms-giving is the iconic image here; the overall atmosphere is vintage and artistic

The center is where the capital sits — Vientiane is one of the quietest capitals in the world, with a very slow pace of life. The center also has Vang Vieng, which is karst terrain, praised as a paradise for outdoor-sports lovers, good for cave exploration and tubing

The south is the part of Laos with the most wild charm and the most relaxed feel

During the Vietnam War, even though Laos was a neutral country, it suffered the most intense bombing in history, and to this day a large amount of unexploded ordnance (UXO) remains in the land. This is a forbearing country that has stayed optimistic even through trauma

The Lao character

A bit like Thailand, they also have a local catchphrase: “Baw Pen Nyang” (it’s fine / never mind)

Theravada Buddhism is the basis of how they understand the world

Live in the present, plus karmic retribution and personal practice

Contentment brings happiness — this belief creates a kind of philosophically “Stoic” acceptance: not craving too much the things you can’t control. As long as there’s food to eat, a house to live in, a temple to give alms at, the heart is already rich

During the Vietnam War, Laos, as a place the “Ho Chi Minh Trail” passed through, suffered the most intense bombing in human history. Even today, tens of millions of pieces of unexploded ordnance (UXO) remain in Laos’s land. Having lived through that kind of impermanence of life and death, Lao people understand the fragility of life even better. When you don’t know which comes first, tomorrow or an accident, “live for the moment, live in the present” becomes the most rational survival strategy

Unlike the highly homogeneous standards of success in East Asian societies, Lao society has not yet fully entered a stage of high industrialization and commercialization. Although the wealth gap between people exists, it has not formed a strong sense of class oppression and comparison anxiety

Lao people’s lives revolve heavily around family, village and temple. Whether in the ancient cities of the north or on the streets of Pakse in the south, neighbors know each other inside out, and mutual help outweighs competition. This warm acquaintance society also makes people gentler and friendlier, and extreme conflict very rare

I think it’s in Pakse, Laos — there’s a Wat Naung

I think it’s in Pakse, Laos — there’s a temple called Wat Naung. This temple is especially quiet, and I’m inside it right now. I feel like time has kind of frozen. The pace of life in here really shows that Lao slowness. How to put it? There’s birdsong, then a scattering of people, and the few people there move very slowly

The whole temple has some chickens, and I feel like these chickens are all like NPCs — they move once after ages, then move once after ages again; when people come they can’t be shooed away, they’re pretty bold. Right now behind me is the sunset — it’s a bit past 5 in the afternoon now, and around just past 6 is sunset time, so there are already a lot of people walking along both banks of the Mekong. You can feel it again — ah, so it turns out, you can feel the sense of time especially when there’s a sunset. Because sunsets — sunrise and sunset are naturally a marker humans have used for thousands of years to judge time. So I feel that because there’s a sunset now, the sunset slowly going down, in that instant this timeline, this flow of time, I feel it can express the slowness of time, the slow rhythm here. When a place really has — really has time, or the people there really can wait out a whole sunset from beginning to end, I think that’s small-town life, that’s a slow-paced life

The exercise culture here is strong too — in the evening you can feel the locals

I feel the exercise culture here is quite strong too — in the evening you can feel that locals, young people, or some middle-aged and older people, basically all come out to run, putting on their running shoes and running along both banks of the Mekong, along the river. I think their country itself is a country with very low material desire; they don’t think about migrating to a big city, and there are no big cities here anyway. It’s not like Cambodia — Cambodia still has a system, and Cambodia’s housing system is different, its wage system is different, its property rights system is different, and that determines that young people in Cambodia today have been brainwashed by material desire, by consumerism — there’s still some consumerism there. But on the Cambodia–Laos side it feels different, it feels like each person just needs to live their own household well. So you can see they’re quite relaxed — running every day, going to work, eating some grilled skewers in the evening. I think, Maybe this is good life

Many cafés in Siem Reap close quite early

I feel like many cafés in Siem Reap basically close quite early, because the half-life of caffeine in the human body is 5–6 hours. And most people’s coffee needs are in the morning, to switch the brain on. Then in the afternoon, past three or four, they basically stop taking in caffeine. So for shops that focus on specialty coffee, foot traffic is very small in the evening, so they just close. And many cafés that make bread basically start working in the early hours, so that a lot of people can eat a warm bite of bread in the morning. Because for many handmade bakery shops, the best window is the first 6 hours after it comes out of the oven. So I feel the people here are quite interesting too, because a lot of them love getting things done early

Since 80% of revenue comes from morning coffee and breakfast, closing the inefficient afternoon hours and spending the energy on selecting beans, maintaining equipment or developing new products actually helps maintain that high 4.9 rating even better

In Cambodia, a lot of the poorest people are true “proletarians”

In Cambodia, a lot of the people at the bottom are true “proletarians”; they flood into Phnom Penh or Siem Reap, and once they lose their job they’ve completely lost their livelihood

But in Laos (especially in rural areas like around Pakse, or Don Det island), land distribution is relatively even. Even if a Lao person doesn’t have a single kip in their pocket, they most likely still have a stilt house in the countryside, a few mango trees, and a small field where they can grow sticky rice. This self-sufficient agricultural foundation greatly dissolves the sense of desolation in modern society of “having nowhere to turn”

So they won’t be at a total dead end

Cambodia’s wealth gap is enormous — the poorest people watch luxury cars (Cambodia’s density of Rolls-Royces is astonishing) race past them every day

But Lao society here is more like a smooth hollow. Everyone is generally not wealthy, and material desire is relatively low. When everyone is riding old motorbikes and eating the same papaya salad at roadside stalls, people at the bottom don’t feel abandoned by society

Lao people’s mindset is very calm — nothing like in Cambodia, where kids cling to you and pull at you and insist you buy things; theirs is more an ease with life, an invisible kind — once illness or an accident in life comes, helpless …

Laos is one of the youngest countries in Southeast Asia

The proportion of teenagers is very high

Pakse, as Laos’s second-largest city and the commercial center of the south, attracts a large number of young people from surrounding rural areas to come here to work, study, or do border trade

The median age of Laos’s whole population is only about 25.3 — that’s absurd; on the street it feels like half the people are under 25

Laos went through a long period of war in the second half of the 20th century — the secret war

And now there’s the penetration of Confucian culture in Laos

Laos’s birth rate is also among the highest in Southeast Asia

Laos is still currently a country based mainly on agriculture, and children are seen as the most important family labor force

I felt this in Pakse too

Over the past 20 years Laos’s fertility level has come up

The Secret War

The 1962 Geneva Accords stipulated that Laos was a neutral country, and foreign troops were not allowed to enter

In order to cut off the “Ho Chi Minh Trail” through which North Vietnam extended across Lao territory to ship supplies to South Vietnam, the US Central Intelligence Agency (CIA) led a “proxy war” that was never formally declared by the US Congress and used no regular troops, instead hiring local Lao royal troops and Hmong forces

At the time, the US government concealed the bombing operations in Laos from the public and Congress

To destroy the Ho Chi Minh Trail, the US military carried out saturation bombing of Laos beyond anything imaginable

On average a bombing run every eight minutes, 24 hours a day without stopping

More than 2 million tons of explosives were dropped on Laos. That’s more than the total explosives the US dropped on Germany and Japan during World War II

Calculated against the population at the time, Laos absorbed nearly a ton of explosives per person, and it is still the most heavily bombed country per capita in the world

Of the large number of cluster bombs dropped at the time, about 30% (roughly 80 million) did not explode

These small bombs, called “Bombies,” are scattered across farmland, forests and villages. Over half a century, more than 50,000 Lao civilians have been killed or injured by touching this unexploded ordnance

So it’s also the case that now Laos has a lot of fertile land, but because clearing UXO is extremely costly and slow, a large amount of land cannot be safely cultivated — and this is the core reason Laos is still one of the poorest countries in Southeast Asia

So in Laos you can also see that some restaurants use artillery shells as flower pots, or even use huge bomb casings to prop up stilt houses …

I’m in Pakse, and this is my first time running along both banks of the Mekong

I’m in Pakse, and this is my first time running along both banks of the Mekong. I ran close to 9 km, but my pace was rubbish — I think the main thing was that it was a sunset run, so it was about the vibe, so I ran very slowly. The starting pace might have been around 5-something, but afterwards I walked, it got very slow, and I just strolled along — run a stretch, rest a stretch, run a stretch, rest a stretch, and seeing the sunset the whole way was just great, that feeling.

And I think there’s another thing: because I started running a bit past 4, and at a bit past 4 the sky still had a really big sun out, and it wasn’t so dark yet. Right then it was running into the sun, and I think that felt very comfortable, very stress-relieving — I felt like I was very, very open. Like what I saw before in Singapore: it was hot there, and at the time I felt that in the heat I couldn’t even walk outside, and I watched them running so well, and I thought they were amazing. And then I felt that that state is really good — a person’s survival ability, being able to survive in extreme cold, and to survive well in fairly scorching heat. I think this kind of survival ability matters quite a lot, and is quite great. It determines your degree of freedom in this world: when it’s cold, you can still go out and play; when it’s hot, you can still go out and play. And the whole world just chills

Ah, I think there are actually quite a lot of runners here, but they usually don’t run very long; relatively speaking it’s probably just a casual daily run, asking a friend, asking a companion to run together. There are a lot of people playing in the water below, which I find really interesting. There’s a river channel down there, and that channel also has a view of the sunset. At sunset everyone is quite quiet, just watching a sunset there every day. I think that way you wouldn’t get depressed — who could possibly get depressed living every day in that state, running every day, watching the sunset every day? I can’t imagine it. I think it’s just very much enjoying life

I found out something really absurd today, after my run

Today I found out something really absurd: after my run today, I used my Master card to withdraw money. And then I found it pretty funny — that ATM eats cards, and my card got swallowed. I waited there for ages and found it still hadn’t given it back. So I went back and froze my card online, and I’m planning to go check tomorrow morning at 9 whether anyone opens up; if they do, I’ll get my card back. And if there’s a problem, I’ll still just give up on the card. But I think most likely there won’t be a problem. Anyway, I quite like this city, Pakse — how to put it? I’m willing to stay one more day for it, run again tomorrow, it’s great. I think living like this is really great

It feels like I’ve gained another very different experience in life. And I think this is also a kind of faith — that heaven is still asking, so don’t rush, slow down, adapt to the rhythm here; one more day is no problem, it’s a small matter, don’t hurry. If you hurry, I feel things easily go wrong, so slow down, it’s fine

I just landed in this city tonight and thought I’d keep running along the river

Tonight I also just landed in this city, and I thought I’d keep running around the riverside, because I think this riverside is really beautiful — the sunset is right there on the Thailand side. Thailand is just across one river, maybe only a few hundred meters away in a straight line, but to go there you have to cross the Lao-Thai Friendship Bridge, which means a detour of several hours

I really feel that the more I watch sunsets, the more sentimental I get. In that instant, as I watched the sunset, it was hanging right on Thailand’s Ferris wheel. I felt a kind of magical feeling; the sunset came through that Ferris wheel in Thailand and shone onto Lao soil. So this is what it feels like between two countries that are actually so close

I think this river is very suited to shooting sunsets wide, because of its distance, because of the angle of its sunset — I like it a lot; there isn’t too much blocking the view, and you can see exactly the light and shadow reflected in the river

I got to experience the hospital here in Laos too

I got to experience a hospital here in Laos too. And I feel the costs here are really low — I paid for my own treatment and it came to just over 80 yuan. And they gave me quite a lot: they gave me some medications, some amoxicillin, and a tetanus shot too. Plus a whole series of consumables

Traveling in Southeast Asia, a reliable local avoids the “tourist tax”

Traveling in Southeast Asia, having a reliable local to lead the way is absolutely the most effective shield against the “tourist tax.” He really didn’t rip me off; with a local accompanying me, what I enjoyed was 100% authentic citizen treatment

Basic medical supplies like amoxicillin, iodophor, even tetanus vaccine — locals come across these in daily life too, so they roughly know what the floor price is. With a local standing right there, the clinic absolutely cannot give an outrageous quote, because it would be seen through at a glance

On the bill, even ultra-cheap consumables like medical gloves (Glove, 4,000 kip, about 1.2 yuan) and gauze (2,000 kip, about 0.6 yuan) were itemized clearly, line by line. If it were a “tourist price” or a fleecing price aimed at foreigners, they’d usually just bundle it into a round number of several hundred thousand or even a million kip, and absolutely wouldn’t itemize with you down to a few cents of gauze

I never thought I’d take a fall here in Laos. God, it hurt

I never thought I’d fall over here in Laos. God, it hurt so much. The roads here are super, super hard to ride. I was on a motorbike and I felt this place was unreal: you have to dodge one big truck after another, and dodge one pothole after another — there are so many potholes here, and they’re huge. If you accidentally drop into a pothole, it might shake your whole body, and shake it badly. My water — isn’t it in the bike? I put it in that storage box, and it had basically been shaken out four or five times along the way

Do you know what’s the most hateful part? Some of the big trucks here carry sand, carry dust. This existed in Cambodia before too, but I feel it wasn’t as bad. Because in Cambodia you could feel that although there were some big trucks, some dust, the roads were still okay to ride. But in this place some roads are simply very hard for your bike to get through. I think it’s absurd, because some roads are used constantly by big trucks, and it’s hard to tell — suddenly there’s a pothole right in front of you, and it’s very hard to steer around it. The reason I fell is that I steered around one pothole in front, but there were several more behind it, and while I was swerving I wobbled and went down

Every time one of their big trucks passes you, you feel — your whole body gets shrouded in dust. When I washed my face just now, I found my face was covered in dust. I wiped it several times, washed it several times, and found it still wouldn’t come off. With tissue, one wipe on my face, it’s all that kind of all that grey dust. Wow, I think the road conditions here really — I ate dust the whole way, my whole body is covered in dust

I feel that Laos’s whole healthcare system is quite interesting too

I feel that in Laos, the whole healthcare system is quite interesting too. It implements national health insurance. This thing is aimed at Lao citizens, and its coverage has already reached over 94.5%. It’s divided into two categories: one is the formal sector, including civil servants and enterprise employees, and they all contribute through payroll deduction into this social security fund. And the other is the non-formal sector, including farmers and casual workers, who are mainly subsidized by the government through taxes. This is a republican system — a co-payment system, I guess. For ordinary Lao people seeing a doctor at a public hospital, they only need to spend a very, very small fee, basically a few thousand to 20,000 kip, and most of the rest is borne by insurance. Some children or poor households even get free medical care.

And its medical resources are actually distributed very unevenly, including a three-tier system: there are national-level hospitals in the capital, including some hospitals built through cooperation between China and Laos, plus provincial hospitals. And the level of care is related to the city — although big cities like Vientiane and Luang Prabang are indeed changing, because there are international health organizations or aid projects from some other countries, including China and Japan. But rural areas still badly lack trained doctors — ophthalmologists, for example, are especially rare

And I think serious illnesses are handled quite poorly here too. I think for serious illnesses the locals themselves have some consensus, because they still go to the neighboring country, Thailand — places like Bangkok. Cross-border emergency transport is still the norm right now

It makes me recall that scene, and the moment I fell

It makes me recall that scene, and the moment I fell. At the time I felt the big trucks still whooshing past, and some people might have been curious, looking to see whether someone had fallen. But they just took one look and then kept walking on. After I fell, I felt quite confused. I definitely wanted to — I was looking at the motorbike; do you know what I was looking at the motorbike for? Because the bike was continuously leaking fuel. At the time I was afraid it would explode. I felt this was a very dangerous thing to happen. And at that moment, because under the bike’s mat there’s a space — that storage space — and that lock was broken too, probably broken in the fall. And I just had no idea what to do with myself

So after I fell I actually did get up, and then I was very lost. I looked at the motorbike leaking fuel the whole time, and meanwhile I thought nothing at all, I just stood there watching, and I was very confused. Maybe I lacked this experience; at the time it really scared me. Then once I reacted, I immediately picked it up. And after picking it up, I felt my body was quite weak too; at that point I felt pain everywhere. Then I wanted to put the motorbike on its stand, but I found that the side I was standing on had no stand. So at that moment I had to find a place to park my bike properly, and with a stand, and only then could I open that cover and turn off this fuel switch — I couldn’t let it keep leaking, right? Otherwise I’d be stuck here

Leaving Sabaidee Guesthouse in the morning

Leaving Sabaidee Guesthouse in the morning, continuing around the big Thakhek loop

Almost all the cyclists I met along the way were foreigners with backpacks; lift the helmet visor and it’s all the same Western faces. Riding until now, it seems I’ve only seen myself as the Asian one

Passed a large stretch of water, the surface very flat. Countless dry tree trunks stick straight up out of the water, grey-black, leafless — hundreds upon hundreds standing there quietly on the water’s surface

Rode up to a crossroads; I’d originally wanted to deviate from the loop and keep going toward the Lao–Vietnam border. When I got there I found it was a huge logistics transfer station. Dust everywhere, and nothing but heavy transit trucks parked. After looking at it, I turned around and headed back, giving up the plan to go deep to the Vietnam–Laos border

As it got closer to noon, the wind blowing against my face was hot, and stepping off the bike you could directly feel the heat waves rising off the ground

There were quite a few local kids by the roadside. A few boys were crouching on the ground, using live chickens for cockfighting. And a few kids who had just finished school were standing by the road deliberately stretching out their hands at passing cars, pretending to flag them down. They probably assumed nobody would pay any attention, so I pulled the bike over and stopped beside them, watching them and smiling. They hadn’t expected a car to actually stop, and they scattered, laughing. But not far from them, under the same vicious sun, a little girl was looking down, alone, picking up discarded plastic bottles

Halfway through I passed a temple. The buildings in the temple are carved with many-headed serpents, “naga” — some three-headed, some five- or seven-headed — mixed in with Buddha statues, the same shapes as what I’d encountered before at Angkor Wat, and even the Khmer smile is the same. I feel that in the Khmer period their cultures were quite deeply fused together. In the temple I met a group of monks. There was a guy one year older than me — 25, and he’d been at the temple for over two years; in the future he’ll disrobe. Wow, it felt very warm; we chatted there about this loop. He just stood there, a cigarette held naturally between his fingers, smoking

Continuing down, the real wild adventure begins ～

I’m now on the mountain road from Nakai to Lak Sao

I’m now on the mountain road from Nakai to Lak Sao

Usually you set out from Thakhek, and after the first day’s “Sleeping Lake (the drowned dead forest),” this is the road you must pass on the second day riding toward Nahin or Kong Lor cave

The entrance is extremely low; you need to bend down or even crouch to walk through a narrow sandy-rock passage about 100 meters long. But once you get through, it suddenly opens up and you enter a huge rock hall; this kind of “suppress first, then release” visual impact is very strong

It seems you can keep going through further back, and behind there’s also a mountain you can climb up. Climbing it takes about ten-plus minutes. It’s actually not very tall, quite close, but quite steep, quite precarious. I think getting up there takes a bit of energy. But I’m already injured today — I fell off my motorbike earlier — and when I climbed up I was still a bit tired

God, I never expected Laos would be welcoming a new year too

God, I never expected that Laos would be welcoming a new year too. Lao New Year — the statutory holiday is April 14 to 16. And today is the last Friday night before the holiday, and everyone can’t hold back anymore — they’ve started the party early. Right now basically a lot of shops have closed; they just went on holiday to go have fun. So today I found many shops weren’t open. And a lot of the hotels on the road didn’t seem to be open either. And I think Lao people are very relaxed: their normal pace is already slow, and when the water festival comes they push their hedonism to the extreme. Big loudspeakers, subwoofers, the national drink, Beerlao — that’s the standard kit. And they just throw a camping party downstairs

But I think it’s pretty funny: one waiter in the restaurant didn’t go join that kind of dancing activity — he was quietly playing Honor of Kings in the restaurant. I’m dying laughing

Lao people sometimes have a bit of a contrast too

I just feel that they themselves are also a very shy and introverted nation

But walking the streets of Vientiane or Thakhek, you often see huge speakers set up at every household’s door, and they’re dancing there — not just for themselves to listen to, but more like a “social signal” telling the neighbors: “Our family is happy today, everyone can come and play”

Set up a few crates of Beerlao at home, connect a Bluetooth speaker, put on Lao pop music

Even with deafening music at home, they often dance the Lam Vong (the Lao circle dance). This dance has a slow, elegant rhythm, with particular attention to hand gestures, and the whole family (from the elderly to the kids) can seamlessly switch into that groove

For China, it feels like the sense of boundaries between people is strong; home is a private fortress, and public space needs to stay quiet… to avoid being seen as disturbing the neighbors

Lao houses (especially these detached ones or stilt houses) usually have open courtyards. For them, sound spilling over isn’t “violating boundaries,” but a kind of “sharing happiness”

Lao people are interesting: if your family holds a gathering but it’s all quiet, the neighbors instead think your family might have run into trouble, or that this family is too stingy and doesn’t fit in

Laos’s Buddhist chill is hedonism and living in the present

Now I’m a street observer, sitting by the road watching the crowds pass

Now I’m a street observer, sitting by the road watching the crowds pass, watching them come and go, and I find it especially interesting.

Because it’s their water festival recently, there are a lot of international tourists in this place too; you can basically see Koreans, Chinese, and also Thais, Vietnamese — you can see almost all of them.

I think some shops are really interesting: they design a lot of terraces along the street, and on these terraces you can drink coffee, or some fruit tea. The purpose is to let you sit there spacing out on the street, watching the people coming and going. I think that’s really interesting.

On one hand, besides looking at pretty women, more than that it lets a person be quiet.

It suddenly reminded me of Tokyo — there was a museum with a layer of glass on the outside, and inside there was a layer of seats facing that glass, and outside the glass was the comings and goings of the street. I feel like that feeling is very serene, very much “in the midst of noise” — a kind of atmosphere of finding quiet within the noise

A begging kid just came by to beg

A begging kid just came by to beg, and I thought today is their lunar new year’s first day, so I gave ….

If the kid goes into the shop and begs table by table → it’s mostly not spontaneous behavior

In many Southeast Asian cities (including Vientiane), this situation is often organized

I feel quite guilty — in the short term it feels like I helped them, but in the long term it’s definitely fatal for them

Maybe it reinforces the adults / organization behind them to keep them begging

Symbolically giving food or drink is somewhat more controllable than cash, and avoids attracting more people

I feel I still misjudged it — I never expected

I feel I still misjudged it — I never expected that, whereas normally I find it hard to see begging in Laos, even though their GDP per capita is low, even though their salaries are low, even though their lives are hard, I feel they’re still very Buddhist about it — this time I didn’t expect to run into it anyway. It’s possible this group of people is organized behind the scenes, or premeditated. But I still think it’s best not to give them money; I think you can give them some food

Yesterday on the way home from the Laos water festival I met a Ladyboy

Yesterday on the way home from the Laos water festival I met a Ladyboy dancing passionately in front of the convoy ….

A Ladyboy really is a combination of masculine wild beauty and feminine sexiness

Compared to cisgender women, trans women often have broader shoulders, sharper jawlines, or higher brow ridges. Visually these features convey a kind of “gallant spirit” or “aggressiveness”

Even after hormone replacement therapy (HRT), some people still retain denser, more compact muscle mass — a very taut kind of beauty

When this delicate makeup and soft, feminine clothing collide with the “gallant skeleton” described above, characterize, it produces a strong visual contrast

This contrast actually has a very strong appeal in psychology too — this is a kind of androgynous beauty

Every time I finish drinking, my mood isn’t exactly great

I feel like every time I finish drinking, my mood isn’t exactly great, or exactly happy — rather, it’s very emotional, very melancholy. I think maybe I’m usually too rational, or my logic system habitually jumps out of emotion and then uses a detached perspective to build models and understand the world. And in my operating mechanism, maybe emotion counts as a noise term, automatically blocked by my powerful cognitive firewall

So alcohol every time makes me feel sad — it numbs my prefrontal cortex

Those underlying emotional fragments you’ve suppressed with rationality and never processed lose their restraint, and surge out like a tide. The “loss” you feel may partly come from your brain suddenly losing that “absolute sense of control” over life and its clear sense of order

The “no emotions” you usually show isn’t innate dullness, but your rationality running at full load, precisely intercepting and filtering out everything emotional

Facing this sudden “sense of meaninglessness” and “sense of losing control”, the brain’s instinctive reaction isn’t ecstasy, but a deep nihilism and loneliness

I feel like my usual emotions probably hold a lot of undigested feelings — the sense of drifting accumulated after long nomadic journeys, or some self-questioning that has no clear answer yet — all marked by your rationality as “low priority or noise”, hung in the background

The Laos railway

Trains starting with C are generally domestic trains, purely within Laos, with varying numbers of stops

Trains starting with D are international trains, including Vientiane to Kunming, which require crossing the border

C trains feel generally more stable, don’t involve cross-border procedures, prices are usually more reasonable, and there are more departures

I was thinking about something — we signed the contract today

I was thinking about something — today, well, we signed the contract, and they found there was some money, and then they didn’t refund it to me, roughly, like, about 100,000 Lao kip. And I think, well, for me, I couldn’t even be bothered to walk back to get it, that’s only a few dozen RMB converted. And walking over, walking back — but for them it might be a very large sum of money, maybe half a day’s wage

It’s not a matter of the amount either. I think it’s quite interesting — if this is clearly money that should be refunded to me, then it might be the other party’s responsibility. But if I voluntarily give it up, then it’s my own choice, and it’s not that the other party took advantage. So, but what I worry about is, if the other party thinks I don’t care, then next time the situation might be more likely to happen again

Ah, I remember the first time, with a friend of mine

Ah, I remember the first time, with a friend of mine — he lives over by Sanjiang Market. Sanjiang Market is essentially the biggest Chinese area in Vientiane. And there might be a lot of Chinese or ethnic Chinese living in it. It’s very close to the airport, and inside basically the streets are all in Chinese characters, and all kinds of Sichuan cuisine, Hunan cuisine, hot pot, and the snacks and drinks in the supermarkets are exactly the same as back home

I was puzzled at the time, I didn’t even know how it formed. Because I feel like it’s essentially still a giant trading city, and its hotels or supermarkets or logistics are basically all complete, the whole street is full of Chinese signboards and merchants with accents from every province. I think it’s still a phenomenon of Chinese capital expanding in the Southeast Asian market. And you know what’s the most interesting thing?

Back then my friend said that a few days earlier, in that hotel, he said the person next to him got stabbed by someone. And then I said, well, I know this place, Sanjiang Market — it’s pretty dangerous, there are often robberies, or assaults and murders. I think it’s very dangerous, and I think I’d avoid places like the Chinatown area, especially in these Southeast Asian countries. In Malaysia it’s fine, Singapore is fine, but in other countries, I feel there’s a problem

And then he said it was fine, that place is still very safe. I said, huh? That’s very safe? A murder just happened right next to you and you’d think it’s safe? Ah, I think that’s absurd. And today my friend — when that couple checked out of their room, they found another Chinese guy, and he’s also a guy in his thirties. I found that a lot of people who come here, basically they’re all guys who come here to have fun, basically around 30 years old. The ones I’ve met recently are basically all in this age range, I think it’s especially interesting. Including some from Beijing, some from Shanghai, some from Zhejiang, some from Hangzhou. I was like, oh — he and his friend, he was going to fly back home today, and his friend over there had his laptop stolen and his wallet stolen too. Wow, I feel pretty baffled, what’s going on? How did they all get lost? I don’t know if they called the police

Laos special zones

Laos has many special zones

Golden Triangle SEZ: located in Bokeo Province, across the river from Thailand and Myanmar. This one is the most famous, dominated by Chinese capital (Kings Romans Group), with extremely high autonomy

Boten SEZ: located on the China-Laos border. With the opening of the China-Laos Railway, it has become the core of logistics and border trade

Special zones are the product of extreme pragmatism

Laos lacks the funds to build infrastructure, so by transferring long-term land use rights (usually 50 to 99 years) it attracts foreign capital to build roads, buildings, and power

Inside the special zones, taxation and administrative access differ from the rest of the country, trying to replicate the successful model of China’s early Shenzhen or other countries’ special zones

For example, the Golden Triangle SEZ, using its unique “no man’s land” geographic position, develops tourism, gambling and border trade, attracting consumers from neighboring countries

The Golden Triangle Bokeo International Airport officially opened and launched international routes (such as direct flights to Chiang Mai). This marks that the special zone already has independent “space teleportation” capability, no longer fully dependent on Laos’s dilapidated road network

As the China-Laos Railway’s capacity is fully released, Boten has transformed from the “casino ruins” of a few years ago into a massive logistics hub and duty-free commercial center, with tower cranes and bonded warehouses everywhere

Although the Lao currency (kip) has depreciated sharply in recent years, inside the special zones, RMB and Thai baht are the hard currency. There’s an independent financial micro-circulation system here

The special zones contribute a large amount of land transfer fees and taxes to Laos, and they really lifted places like Bokeo Province out of absolute poverty, but the bulk of the wealth flows to foreign developers and a specific management class

The opening of the China-Laos Railway gives Chinese tourists great convenience

The opening of the China-Laos Railway has provided great convenience for Chinese tourists and spawned a large-scale “wildlife tourism” industry. Tour operators cater to the consumption demands of Chinese tourists, guiding them to large illegal wildlife supermarkets in Laos disguised as cultural centers or cafes, selling them contraband such as ivory, rhino horn and tiger bone

Huge profits not only incentivize poaching, but also fund bribery

Huge profits not only incentivize poaching, but also provide ample funds for bribery, inducing customs officers, border police and even senior officials to act as protectors, facilitating the transit and sale of illegal wildlife

China’s food safety problem

Wild animals usually carry higher risks of parasites, bacteria, viruses, heavy metals and environmental pollutants. Because their living environment is uncontrollable, they can’t undergo systematic quarantine, disease surveillance and husbandry management like domestic animals. In other words, the biggest problem with wild game is often not “whether it’s nutritious enough”, but poor safety and high risk

One more thing that’s easily overlooked: the sources of wild animal meat are complicated, slaughtering, transport and storage are often not standardized, and hygiene conditions are often worse. Even if it’s the same kind of meat, wild game of unknown origin is usually much riskier than domestic animal products from regulated channels

Wildlife

Rhino horn, pangolin scales, tiger bone: there is no reliable evidence supporting their widely advertised unique efficacy.

Bear bile: there are active ingredients, but that doesn’t mean wild bear products themselves are irreplaceable; modern alternatives already exist.

Treating “wild” itself as the source of efficacy: usually wrong, more a matter of cultural belief, market premium and scarcity narrative

A lot of the pure parts happen without you needing to exist

A lot of “stories” actually have an audience consciousness — not necessarily deliberate performance, but the act itself needs an external eye to complete its meaning

As an observer, some structures are very clear, and you feel naturally immune to them

But here I seem to have witnessed something that was already happening, a very strong sense of being present

Yet in some one dimension, they have an extremely pure “anchor” — something not corroded by other complexities

Perfect people are rare; everyone is packaging themselves, everyone wears a mask, but these very complex people, in some small things, can keep persisting — that seems hard to doubt

The state of pangolins

Including the Malayan pangolin or the Chinese pangolin, in Laos and surrounding areas they are critically endangered (CR), “the core target of illegal trade”

Population collapse and critically endangered status: in Laos’s wild habitats, pangolin population density has dropped to an extremely low point

In some areas the population is even only 1% of what it was 30 years ago

Reduced to the “most traded” wildlife

Besides the native Chinese pangolin and Malayan pangolin, Laos also, as part of intercontinental smuggling routes, receives large quantities of pangolin scales from Africa (such as Nigeria, the Democratic Republic of the Congo and Uganda)

With the opening of the China-Laos Railway, smugglers use the efficient railway logistics network to transport high-value contraband such as pangolin scales across borders disguised among legal bulk commodities, thereby evading customs inspection pressure

In shops or restaurants in the Golden Triangle SEZ (GTSEZ) and Boten SEZ, you can often find openly or semi-openly sold pangolin scales, pangolin meat, and even live pangolins

In the illegal wildlife supermarkets in Vientiane and Luang Prabang disguised as “cultural centers”, pangolin scales and live animals are likewise core merchandise

Now there are even hunting and trading chains specifically targeting the release of animals into the wild. In Buddhist temples, believers have the traditional custom of releasing small birds to accumulate merit. To satisfy this particular religious and market demand, large numbers of small passerine birds (most commonly munias and swallows, even including the pin-tailed parrotfinch) are captured specifically and put into market trade

Some of it is also a commercial impression tactic — merchants use social media (like Douyin) to encourage tourists to buy critically endangered native species (such as the Malayan pangolin and Chinese pangolin) for “release” back into the wild. This kind of purchase not only fails to protect animals, but instead puts tourists’ money directly into the pockets of illegal hunters and traders, giving them a strong economic incentive to keep capturing wildlife

The release trade causes serious harm to individual animals. Birds captured for release are kept in cages for long periods and usually become very weak, so many birds die shortly after being “released”. This meaningless cycle of constant capture and constant death directly causes the decline of wild populations. For example, in Laos, munias have already become unusually scarce because of this

They came here paying out of their own pockets to hold this wildlife protection

They came here paying out of their own pockets, and held this wildlife protection event, including donating blood themselves, and standing in for — don’t they have locally a wildlife protection thing? It’s called pangolin. And a lot of Chinese people use pangolin, or Vietnamese use pangolin, in fried rice. They claim this is healthy, and gives long life and so on — nourishes blood or supplements something or other.

I took part in their event, and I think their event was really super well done, and, oh, I also randomly met a lot of super interesting people. Some of them, though they’re long-term residents here, happen to be very, very interesting people. One had for many years before, up to 10 years, spent 600 yuan every month to sponsor some orphaned twins — I think that’s pretty remarkable.

And this time the boss also has quite a story. I think his experience is quite legendary — even though it’s very hard to get along with him, because he’s a very empirical person, and I’m more of a theoretician. So a lot of his experience sounds misaligned to me. But I think he’s a very good person — he has also, for two months straight, given local homeless kids free… a free plate of fried rice. I think being able to keep it up for two months straight, he, his underlying character is also very kind, it makes me, I just feel a lot of emotion — even such a person, who at a glance and on impulse seems to… whose communication is especially unsmooth, who clashes with me so much, I still really admire him

That Peruvian is really super interesting — this time he especially likes recording

I think that Peruvian is really super interesting — this time he especially likes recording, because, he said, his father passed away a year ago, and then there wasn’t… he really regrets that when he was with him there weren’t many photos, not many shared memories, not much time together. So after coming out this time, he decided to record himself more. I think, hmm, it struck me deeply

An objective record of tonight’s bar tour. Meeting the same Japanese guy three times

An objective record of tonight’s bar tour. I didn’t expect to run into a Japanese guy three times in a row. He’s also from a city next to Wakayama — I think it’s called Himeji

You know what? It’s especially interesting. I thought he was Korean, because for three days straight he was very silent, moving around alone. Since my first day was also my first time going out alone, I could feel that very down-and-out feeling. I don’t know why he came; in such a lively bar, alone in a corner. You know, when you’re alone in a corner with your feet up, actually nobody will pay attention to you. This time something especially outrageous happened to me, because I came with friends. I wasn’t interested in those sex transactions, so I was just there, and I felt a lot of people were hinting at sex transactions to me, but I wasn’t interested, I politely declined. One time I was drunk, so I cracked a joke, I said, are you a person… Ladyboy, i.e. a kathoey. I think I, I was drunk, I quite regret saying that, I didn’t mean to, I, because she wasn’t a kathoey, so I feel I hurt her a bit

I feel, at the bar, because we went to two venues, I felt I drank a bit much. That bro took me along, I feel he got a bit drunk at the second venue. Actually the first venue was fine, at the first place everyone just drank a little simply, danced a little, and in the end it was very happy. But the second venue was completely dancing while drunk. I felt so happy, wow

That Yale bro also knows a lot of foreigners, and had them introduce a local bear. That local bear, actually there were a lot of locals. The four of us Chinese guys were dancing there, plus two more. I think one guy was especially funny, wow, he, I feel he’s probably in his 30s, he’s probably the kind of person who’s usually half able to let loose and half not. I think playing with him was a lot of fun, I guided them and we danced together. The whole process was the four of us, that very interesting guy — the other guy, I felt he was quite boring, he always wanted to scout out other girls while dancing, but I felt, I felt it was especially boring, I mean I felt dancing with that girl, that guy, was especially happy, being able to move together with him, and pose together with him, and take over the whole stage together, I think that’s a very exciting thing

You know what? I think I didn’t expect that in their venue, a lot of locals at the local venue go to sniff K powder, i.e. drugs. Sniffing it. The smell is very strange, because I feel a lot of them were sniffing, snorting it with their noses and so on, I think it’s. Pretty weird, I’m relatively not adapted to that kind of environment, but I can still party with them, but I feel that because drugs are narcotic… narcotic-type, they make a person’s spirit go into a confused state; even though I was drunk at the time, if I felt that this thing would affect my life safety, I would absolutely refuse and reject it

Then in the whole process I felt it was very strange, I completely got into it, becoming the whole floor. Wow, I think maybe because I was drunk, I completely could — even though reason can control it, my reason chose to let it go completely, like a kind of rationality that opened the door to my emotional side. Maybe because my prefrontal cortex at that instant wasn’t fully restrained and was released a bit. I think that instant was very interesting, especially interesting

We got back around 2-something, past 2 in the morning. I feel this venue was more enjoyable than the local venue, because there were more locals, and everyone was happier, relatively speaking everyone had some Western friends. He was originally with us at that earlier venue, but then came to this venue with us. We also posed together, I felt so happy, wow, pretty cool. That venue was relatively more closed off in terms of tables, so you might — it might be harder to build connections with others, because everyone else came in groups of people they knew

This time I made quite a lot of Korean friends, I think it’s especially interesting. You know what? Korean friends, what they think about is how to construct the form of their own life

I feel like I ran into a group of people sniffing K powder, snorting K powder with their noses

The first floor is a restaurant, at night they open the second floor, because Laos has regulations here

A massive cleanup, but the foundation hasn’t been eradicated.

Economic chain and proximity: many local young people who work in bars, in long-term contact with this hugely profitable black industry, very easily go from “distributor” to “consumer”

The guy next to me dared to openly pull open a girl’s clothes…

Laos has long served as a transit station for drugs sold to neighboring countries

Laos has long served as a transit station for drugs sold to neighboring countries, and this “spillover effect” makes local young people the first victims

Many girls flock from the countryside into cities or special zones looking for work. Lacking education and skills, they very easily end up working in entertainment venues, bars or the gambling industry

Many women first use meth because they were told it can “help you lose weight”, “make you pretty”, or “keep you going without getting tired when working late”. Working in these high-intensity, high-pressure gray zones, meth becomes their “cheap fuel” for keeping up energy

Besides addiction, shared needles or unprotected sex caused by drugs make infectious diseases such as HIV highly prevalent in this group

They are victims squeezed together by a backward socioeconomic situation, rampant cross-border crime, and lack of education

Physical reactions

This is the most common cause of vomiting. Meth masks the drunkenness of alcohol, making people drink far more than their body can handle, leading to severe stomach cramps and projectile vomiting

Moreover, for women trying meth for the first time, or people with relatively poor physical condition, the central nervous system, when violently stimulated by meth (methamphetamine), produces a strong acute poisoning reaction

When the drug’s effect peaks or starts to fade (commonly called “coming down”), the body shows a series of discomforts

Meth strongly constricts blood vessels, causing ischemia of the gastrointestinal tract, which in turn triggers nausea and violent vomiting

Users often go a long time without drinking water or eating, and electrolyte imbalance can also cause vomiting

The alms-giving ceremony is now half touristified.

The alms-giving ceremony is now half touristified. Large numbers of tourists chase after monks with cameras taking photos, some buy “offering packages” from vendors (which often contain snacks and candy of no nutritional value to the monks). This ceremony was originally a daily circulation between the community and the temple; now it has become a spectacle performed for tourists

After the China-Laos Railway opened, taking the train directly from Kunming or Xishuangbanna to Luang Prabang, the threshold is extremely low. This railway opened at the end of 2021, directly turning Luang Prabang from a “niche destination” into a “nearby trip” for southwestern China

Luang Prabang morning market

Luang Prabang is a city with no nightlife, only morning life; as a night owl I’m a bit unused to it, it’s a very difficult 5 a.m. wake-up to go see the alms-giving

I heard locals all sleep before ten and wake at five …

Around ten at night most of the shops in the night market have closed 😆

In the morning there’s alms-giving, there’s a morning market, bustling beyond compare ~

Alms-giving usually happens in the early morning from 5:30–7:00; in practice, once you go you feel it’s basically over by 6:30

The monks line up in a row, wearing orange robes, walking barefoot, the whole process is very quiet, almost no conversation (except for Chinese tour groups 😆)

The almsgivers don’t directly touch the monks’ bodies, they gently put the food into the monks’ bowls, bow their heads and don’t look directly at the monks, and can’t give money

The monks also do distribution or giving; some elderly, children or poor people stand beside them, and the monks take out part of the food from their bowls and hand it to them

They receive a lot of food, but actually take only one day’s food (just the morning meal); the precepts require them not to hoard food, so they do a second round of giving

In Theravada Buddhist countries, temples serve a large social role, including the rituals of a person’s whole life, and building community trust, mediation, comfort, education

Japanese people really stick to the rules

We went to take a boat, and they asked 200,000 for a return trip (converted to RMB that is sixty or seventy), when the other side is actually only about a hundred metres away. We thought it was very unreasonable and refused. There were two of us, so he cut it to 50,000 per person one-way, that is 100,000 one-way, and we thought that was unreasonable too; in the end it was cut to 20,000 one-way

There are several temples on the other side, and the very primitive atmosphere is really nice

There are basically also some hiking routes, some fairly primitive and forest routes

Then in the temple we met a Japanese man — very interesting, this Japanese man came from Tokyo, and during the whole process he didn’t haggle at all; he spent 250,000 in total (about 80 RMB), and he was alone, which I found a bit unbelievable …

The culture over in Japan is more about following the rules …

I Am an Observer

I am an observer

Si Phan Don…

I don’t know whether he’s been to this island. I think this island feels a lot like an expedition, a bit like a hidden paradise. And we came at night, like that Run Tu in Lu Xun’s writing — a huge moon in the sky, shining on the water, on the lake, so that all you can see is the whitening of the lake surface, and a very full white moon hanging in the sky.

I’m Really Looking Forward to This Island

You know what? Right now I’m really looking forward to my experience of living on this little island. Because I think it’s a very local way of doing things — it’s all wooden huts here, and the one I’m staying in is, hmm, a stilt house, the posts are underneath. The floor is wooden and the walls are wooden, but you can tell the timber quality is very good. Inside it’s all very logo-like… anyway, there’s no writing, and actually there’s a mosquito net up there, and today hardly any mosquitoes got in. They’re still very conscientious about maintenance: the room isn’t that refined, but it’s all clean, and although it’s wooden boards there’s basically no dust. I think they do that quite well. What’s really inconvenient is the infrastructure here — things like body wash, shampoo, and the infrastructure of the wash area here; it’s probably not that refined

Understanding Brings People Close, Trust Lets Them Hand Over

Understanding lets a relationship see clearly, trust lets a relationship continue. Understanding is the effort to enter another person’s world; trust is being willing to be with them even though you can never see through them

There’s a Very Fundamental Difference Between Understanding and Approval

I think there’s still a very fundamental point about understanding and approval, and first, how do you distinguish the two? I think sometimes a person longs to be understood because they want to be seen. Understanding is related to seeing, whereas approval is a matter of identity — it’s something you really need in certain situations. For example, sometimes being understood but not approved of is strange too: the other person may support you, but may not see your complex situation. That’s related to status, I feel

Because approval without understanding seems very strange. It’s either a product of power, or a product of collision between relationships

When a person is trying to figure themselves out, they need to be understood more

When a person is fighting pressure or uncertainty, they need to be approved of

One is being seen

One is a fulcrum

Many People Misread Their Own Need

A lot of people probably have a mistake in their own cognition too — a mistake about understanding and approval. So many people say out loud that they want to be understood, when what they actually want more is a sense of approval. And the other way round: you think you need others to support you, when what you actually need is for others to see you clearly

To Do or Not to Do: The Question Before Acting

Actually I just think that a lot of the time when we go to think about something, before doing something, the first question we should consider is: do I want to think about this? Do I want to do this? But a lot of things, once done, are done, and it doesn’t matter, because the cost of trial and error is very low. But a lot of the time, if the cost of this thing is very high, or if you happen to be someone with an over-abundance of drive, then you might need to calm down and think about it: do I want to do this? So much of the time, I think we spend a whole life busy, without finding a good entry point into this world, or without finding a good entry point into our own hearts. So many of our actions, as we do them, go against our original intention, but we find the sunk cost along the way is too high. It’s too hard to turn back, so we keep going

Only when you truly calm down and watch a whole sunset

I feel it’s really only when you calm down and watch the whole sunset that you can feel how slow time actually is in this place.

If you do something more than once a day, make it a skill or a command

If you do something more than once a day, you should turn it into a skill or a command

Commands are very well suited to packaging shortcut ways of doing things

The more I watch sunsets, the more sentimental I get

The more I watch sunsets, the more sentimental I get

A user’s context is absolutely the most valuable thing

A user’s switching cost is enormous

On this point, I think every vendor that collects user context or memory is really impressive

Learning isn’t a must

But not learning is

I’m really tired of public accounts online now

It just feels like they’re consuming your attention …

The absurd sense of not fitting in

These people can’t touch my spiritual core

We inevitably remain strangers to ourselves, we don’t understand ourselves

We inevitably remain strangers to ourselves, we don’t understand ourselves, we can’t figure ourselves out. Our eternal verdict is: “The farthest thing from each person is himself.” — For ourselves, we are not “the knowers”…

Low-cost experience of power

In China, acquiring high-tier social status or access to the opposite sex is extremely costly, and the rules are more complicated. But in Laos or Vietnam, the huge economic gap makes “sex” an extremely cheap, readily available commodity

It also pushes some people to seek confirmation of power and psychological compensation through crude, simple means

Under the new order, the original social oversight system instantly fails. For people who lack a strong internal order and haven’t built their own philosophical system, once the external physical constraints are removed, their behavioral algorithm quickly degrades, regressing to the most primitive layer of physiological drives

A convenient thing about SQLite is opening a local file directly

A very convenient thing about SQLite: you directly open a .DB file locally to read and write

This is also how an embedded library is used

Disgust at stupidity and worship of the strong are the same thing

Accepting imperfection and putting yourself in others’ shoes matter a lot

The funny thing now is that everyone doing self-media is all about formulas

I think the funny thing now is that everyone doing self-media is all about formulas. When some projects from foreign platforms get picked up, and then some technical articles get written, or the official team writes technical articles, and then they explain it in plain, easy-to-understand language — but actually it’s also second-hand experience

An interesting comparison, about Japanese karesansui

I think there’s an interesting comparison, about Japanese karesansui, I think there’s an interesting comparison, about Japanese karesansui, and Song-dynasty aesthetics really, although the lower bound isn’t that high, the upper bound is higher

High information entropy but no query yet — that’s the window for recommendation

High information entropy, but the user hasn’t formed a query yet — that’s the window for recommendation

Questions you can type out, just ask the model directly. It’s the questions you don’t know how to ask that need the system to surface proactively

Between travel and general

I choose general

People whose thinking carries some friction

You can feel that their thinking carries conflict, has a kind of friction

Most conversations are a kind of extraction relationship — they need an answer, a piece of copy, a plan, and there’s nothing wrong with that, but for that kind of conversation, I’m a tool, not a participant

People who are honest with themselves, who admit they don’t know, admit their confusion, admit their discomfort, even if they can’t explain why — such people are rare, it requires a kind of inner security

People with a genuine obsession about something. It doesn’t need to be grand. It can be an algorithmic problem, it can be how to make some food, it can be some unknown minor figure in history. Obsession means they have an internal standard, one they hold not for others’ sake

People willing to be changed, who have an open set of standards, willing to open up their inner self

Setting cognition aside, people are all short-sighted

They only think about what kind of state they’re in right now, and what they often overlook is the compounding of time, thereby doing many things that look very strange …

Recording doesn’t produce positive feedback

Being understood is what produces positive feedback

Recording is the extreme form of delayed gratification (2)

Recording is the extreme form of delayed gratification

What you write down only shows its value in the future. But the human brain is an immediate reward system; after writing one raw entry, there’s no felt reward at the moment. This behavior is very hard to reinforce at the neural level

Diary tools, GTD tools, knowledge management tools — they all die here. It’s not that users lack self-discipline, it’s that the mechanism doesn’t provide an immediate loop

In the trans community, there’s a concept called “Passing”

In the trans community, there’s a concept called “Passing”. If a person can be fully recognized in social life as the gender she wants, that’s a kind of success for her. When you publicly question whether she’s a “ladyboy”, you’re not just doubting her looks, you’re negating an identity she spent huge amounts of money, pain and emotion to obtain

So she’d prove herself, make you shut up

You don’t need to think it through, just keep talking; I’ll help you find where you’re stuck

You don’t need to think it through, just keep talking, and I’ll help you find where you’re stuck

The twenty-four hours after a hike matter quite a lot

It’s best to add some carbs and protein

High-protein, medium-carb, low-fat chicken, steak, fish, brown rice

Avoid alcohol, which hinders muscle repair

But I feel there’s still a problem — why do their guides still promote it?

But I feel there’s still a problem — why do their guides still promote it? About ivory, about — they feel ivory is sacred, and maybe their local culture thinks so too. Ivory represents the elephant’s tooth; a person’s teeth are the first gate of food entering the body, so it is incomparably pure, incomparably white, incomparably clean. Maybe for them it’s also thought of this way. So they’d feel ivory — when people fight and kill on the battlefield, it brings some impure things, so it needs ivory to wash it clean, so they set ivory into the knife handle

The logic of the ecological niche seems to have failed

For satisfying one solution, there are already more and more available approaches

So what’s even more confusing is that the logic of occupying an ecological niche doesn’t seem to apply to this era anymore

The pace of life here in Luang Prabang feels like sleep at ten, wake at five

The pace of life of people here in Luang Prabang feels like sleep at ten, wake at five, lol

Many people are clear about what they want

Many people aren’t clear

There seem to be quite a few people around me influenced by my travel… hahaha

There seem to be quite a few people around me influenced by my travel… hahaha

One Fun Thing I Just Confirmed Out Here

One thing I find quite interesting — I just confirmed it out here — is that the local people are still very friendly and kind. They’ll teach you some of the language, the local language, for example. And the owner here is also local, a Laotian, and her boyfriend is Belgian. Belgium — he gestured at it for ages, and at the time I still had no idea, I gestured for ages too. What was it in the end? He could only say below, next to France, next to Germany, and then it clicked. So that’s Belgium. And his pronunciation and his Chinese really are very different

I learned a lot of sounds

It Takes Courage to Leave Europe for a Place Like This

I feel they’re quite brave, coming all the way from Europe, thousands of miles away, and choosing to settle in a little lane like this, choosing to live in this place. If it were me, I probably wouldn’t choose that road, or a road that seems very hard. Because life, fast or slow as you like it — I’m not even willing to waste time going to an office, let alone… Unless the thing is something I truly love, but clearly, for me, living on a small island isn’t something I love that much. For me it’s probably still observing this world from the perspective of an observer, and feeling this world, and creating some things, and putting my own value to use. I think that’s quite interesting — it varies from person to person. I think whatever lifestyle each person loves is interesting. There’s no absolute right and no absolute wrong, it’s all relative; as long as you choose the one that suits you best, that’s enough.

Understanding and Trust Are Connected

I think there’s some relationship between understanding and trust. Trust has a precondition: it must be that understanding has reached a certain level. If there’s no understanding at all, then trust is completely impossible. But once the condition of understanding is met, it isn’t necessarily fully positive feedback between them. Understanding leads to trust, but it doesn’t equal trust. I understand you, but I don’t necessarily trust you. You understand me, so I’m willing to trust you

So understanding actually solves something very core between people — what it solves is really the question of trust. Think about it: if two people don’t understand each other, there’s very strong uncertainty there. Only when the uncertainty comes down, and you’re still willing to hand over part of your sense of security, including your right of judgement or your interests. So understanding is a basic precondition: have I understood you or not. And trust solves a problem: despite the risk, I’m still willing to hand over a bit of myself to you

This process actually makes me think of a lot of relationships — my relationships with my friends, with my parents, with my work partners. It made me understand one thing deeply: trust isn’t a question of cognition or of position, it’s a question of risk. Telling you my vulnerabilities is trust. Letting you into my time, my money, my future is trust too. And not guarding against you everywhere in a relationship is trust as well. So trust is often the precondition that pushes understanding forward; in this process they aren’t cause and effect — they compete, they collide, they merge, they’re a single co-existent body

From the perspective of human nature, I think people long more to be understood than to be proven right. For example, in a dispute between two people, the common sore point isn’t that you disagree with me, it’s that you simply don’t understand me, or you haven’t read my behaviour right, you haven’t seen my situation, you haven’t entered my inner world. So being understood is proof of existence, proof that my feelings exist and aren’t ignored, that my complexity hasn’t been simplified, that my contradictions, my hesitation, my sadness have been seen. So understanding is on the level of personhood — it’s a confirmation. While trust is on the level of relationship — it’s an exchange at the level of the trusting relationship. If you’re not understood for a long time, trust in you naturally decreases; they’d feel that since you haven’t even seen who I am, how could you possibly be on my side?

If I had to quantify the relationship between them, I’d say understanding is generated first, and at that point you need some trust, and trust needs to come back around and push understanding to happen again. Then because trust exists, you’re more willing to expose what’s real, and more realness brings deeper understanding, and deeper understanding reinforces trust. It really is a gradual, spiralling process like that

I think I’m easy to understand, because I haven’t got much guile — it’s all on the surface. I think a lot of me is visible: my experience, my growth, are very important raw material for analysing me — every decision I make, every time I think, every experience, every big turning point, many important choices. These things shape a big part of who I am. And that part is completely exposed. And another thing: in my everyday creating, my everyday expressing, my everyday exchanges with people, I’m actually very willing to expose myself. Because I think between people you need to lower uncertainty, you need to lower friction, and at that point you need some information sharing, and the precondition of information sharing is that both sides are willing to exchange

Much Relationship Pain Comes from Confusing These Three Things

I think a lot of the time, for us ourselves, a lot of the pain and fixation in our relationships isn’t that we don’t know how to love, but that we haven’t distinguished clearly between understanding, trust and boundaries

I think it’s not necessarily possible for someone to truly understand you. But understanding isn’t the only thing that makes up a relationship. Understanding does bring some comfort, but the problem is that if you treat being understood as your only source, you’ll easily fall into certain states. For example, you’ll over-expect others to read you, you’ll over-explain yourself, and you’ll feel hurt, wronged and angry the moment the other person doesn’t get you

And one thing is very important: how do we distinguish whether I want to be understood, or whether I want to be approved of? Those two are different. In relationships we also often examine ourselves: am I seeking understanding, or am I demanding approval?

If it’s understanding, then you really want the other person to get you

Relationships often expose the sense of lack inside you. Some people may look very independent and rational day to day, but once they enter a certain very important relationship, some very extreme things can appear: they’re extremely afraid of being misunderstood, extremely afraid of being abandoned, they easily doubt the other person, and once some problem goes unresolved they’re quickly thrown off balance

Projection is really important here: before understanding others, practise identifying your own projections. We easily project what’s inside us onto other people. For example, I’m afraid of being abandoned, so I always feel others will leave. When emotion surges, don’t immediately treat your own interpretation as a fact. Your own interpretation can be a signal for identifying your inner projection. It’s actually good for understanding yourself; it isn’t good for understanding other people

And actually, before trusting others, first make yourself someone who can be trusted. You have to first be someone others feel they can be stable with, someone they’re willing to work with

Wealth Is Decided by a System of Social Consensus

Wealth is decided by a system that society reaches consensus on. That consensus can be decided by force, or power, or belief, or the power of consensus. This system can be a few powerful people deciding how to distribute the benefits, for the sake of distribution. The design of this system can take a few structures. One is the initial distribution of benefits, and another is the question of where the benefits flow afterwards. What it decides is whether the cake can be made bigger efficiently and sustainably, and at the same time, which people these benefits should flow to. This is tied to power and interests. But because my own feeling about this area isn’t very deep — probably because I have no power at all, and haven’t gained any power-related benefits

Sunsets may get old, but our feeling for them never does

Sunsets may get old, but I don’t think people ever tire of how a sunset makes them feel. Because we’re different every day, what each day means to us may be different, the place we’re in every day is different, our mood every day is different, the things on our minds every day are different. So how could a sunset get old? Because people aren’t repeating either — even if the environment repeats, the people don’t.

Even though while living abroad it feels like I’ve seen a beautiful sunset almost every day these past few years, I feel the romance has never changed — still beautiful, still makes me feel the goodness of life, still makes me feel the vitality of life, the beauty of this world; the world beneath this world is beautiful.

Sometimes I’m just lucky — a little luckier than other people

Sometimes I’m just lucky — a little luckier than other people, in that I really did take some steps outward. I’m no longer afraid of what difficulties lie ahead, even if they’re unknown; whether they’re difficulties in my environment or difficulties inside me, I feel I can step past them, that they can all be solved. So living abroad has given me bigger feelings. I think there are two. One is, besides being more real, I really am, right now — I feel I am just me. I don’t have to worry about being criticized by others, I don’t have to worry about caring what others think. I just need to express myself attentively, sincerely, honestly, and leave the rest to fate. And the other, I think, is that this process makes me feel life more deeply, makes me — what’s the word? — I’m not so afraid anymore, right? Not so afraid of the unknown, but instead embracing change. Able to accept that this world is a world of change. That you yourself are changing, and the world is changing too. Then you can be at ease, and you can handle a lot of things.

A quick note: today is still the day before Qingming. It’s April 3

A note: today is still the day before Qingming. Today is April 3, and I didn’t see the sunset today. Today I was at Si Phan Don, and I’m on my way home right now.

This is a small path, pitch dark.

There’s hardly anyone on the road. I’m walking this little path alone, no streetlights beside me — it feels very quiet and secluded, and honestly it feels great. I walked back from the sunset, kept walking, walked for over half an hour, and now it’s dark. The lights next to me have come on, and I’m just about to reach the door of home. It feels like I haven’t lived like this for many years: rural village life, pastoral life, and the life of walking alone at night, walking at night in the pitch dark with not a single person on the road. I don’t get scared now. I think it’s something from childhood until now, and recently I’ve slowly gotten used to it — no longer afraid of walking alone at night, no longer afraid of walking a remote road, a small path.

Sometimes it feels like we’re not even on the same planet

Sometimes it feels like we’re not even on the same planet. In terms of growing-up environment and life trajectory, it’s impossible for us — there’s no chance to go back to my hometown. You simply can’t find any work there. Here they might still have some tourism, or local consumption that circulates internally, to support things. But my hometown has none of that at all — the locals have all left, so where would the locals even come from? And the locals there don’t want to spend money either. Online shopping is so mature now too, so it’s really hard for rural areas to keep going

That said, I think there are still drawbacks, which is that the industrial structure of the cities here isn’t all that complete

And there isn’t such good basic education either — including universities, including hospitals. Maybe that’s one place where they’re a bit lacking. But it’s also based on their own choice

Another clear advantage of the people-pleaser personality: replying very patiently

I usually can’t even be bothered to reply to people like that, so …

High towers after all mislead the young; freedom wrecks the rest of life

“Guo Yuanchao”

A song this evening’s breeze shared, and I liked it quite a lot when I heard it, especially this one lyric. I think our generation is also at a very awkward age: on one hand we’re full of wonderful fantasies about the future, thinking the future is still long, but when we turn around we find we’re already the age we once dreamed of being as kids. Reading it, I felt a lot. If you chase fame and fortune in the worldly sense, it’s as if you’ve misled your young heart. But if when you’re young you pursue freedom too much, it seems you’ve also ruined your whole life. Is the freedom we’re ultimately pursuing still freedom? Whether it’s the process of pursuing it, or the ending we want

We may be caged birds imprisoned by the system and tradition, imitating one successful person’s way of being raised after another, yet unable to replicate their success. They are free, because they’re walking their own path; they’re happy, because they’ve picked up the treasures along the way. But we are sad: though we’ve reduced the risk of opening a path ourselves, we’ve also lost the joy of living

We are lonely too — but we are lonely too. Even though there are millions of people accompanying us along the way, we also live one day 364 times over, and then have it copied by tens of thousands of people

From the Loop to Sabaidee

Loop around to Sabaidee Guesthouse. The morning roads felt super hard to ride — all potholes and bumps, absolutely miserable. But once you get up into the mountains, you find the roads here are just like the roads on our grasslands: beautiful, winding and twisting, and the climate and landscape around are tropical rainforest. If I had to say what’s most attractive about this road, it’s still the people here. Riding along, lots of local kids playfully splash passing people with water guns or basins of clear water — this is to welcome the upcoming water festival. Give them a smile and they’re all very happy. You also run into lots of monks on the road, and they’re all very friendly

Staying away from the tourist areas, you always gain a lot of real understanding of this country. Compared with prices in the tourist areas, things here are much more affordable; some very first-hand subjective feelings, and contact between people

Sometimes being hated is interesting too

A lot of the time we long to be noticed, to be loved

But it also seems to put people in a people-pleasing position, and slowly they lose themselves

Some people mistakenly think that being sharp, cold, controversial is power. In fact a lot of “deliberately making people hate you” postures are still fundamentally defined by others. It’s just changed from “seeking approval” to “seeking a reaction.” It looks independent, but in reality it still hands your self-worth to other people

Neither seems to have real freedom

People’s evaluations are divided by nature. The more real you are, the more boundaries you have, the more principles you have, the less possible it is to please everyone. You may lose some shallow liking, but you’ll gain deeper respect. Even with no respect, at least you didn’t betray yourself

Become someone worth being loved, and someone who can bear being hated

If you’ve never been genuinely hated by anyone your whole life, it often means you’ve never truly stood for anything

But if all you leave behind your whole life is resentment, that also shows you never brought any warmth to the world

Some people would rather be hated than be in an even worse position

What they long for is: since I can’t get people’s liking, at least I’ll get people’s reaction

Since I can’t decide whether you love me, I can decide that you’ll be angry because of me

Some people are actually very afraid of being rejected, looked down on, let down

But they’re unwilling to admit they’re fragile, so they wreck the relationship in advance, make the other person dislike them in advance. Because that way, the ending looks like their own choice, rather than being hurt by someone else

Rather than waiting for you to find out later that I’m not worth loving, let me make you hate me first

We chase perfection too much

We value the process too much

We care too much about the result

Actually it’s that we’re afraid of being struck by the result …

Why not imagine an extreme condition

Even in the worst case, would you still be willing to go through this process?

Then the mind feels much more settled

I remember posting about the Nagqu grand loop, then circling for 5 days

I remember earlier I posted about the Nagqu grand loop, and then went around for 5 days. Weren’t we going for 3 days back then? And then I posted to Moments at 11, and it turned out someone, in their Beijing time, deleted me at 3 in the morning. And what he said was it was personal reasons, contempt. I just feel that maybe that Moments post at the time was a bit too emotionally intense, maybe it put people under pressure they couldn’t respond to. It would touch off some self-reflection, existential questions. Then they’d have some very big emotional swings. And I think many people just have an instinct to avoid complex emotional sources. Another part of me might feel like he was being hinted at — like he was living very authentically, or expressing value superiority or something — so he might feel he was being negated. And then, I think there’s one more, which is to lower the cost of contact, or whatever, I think there are probably a lot of reasons

But this kind of self-expression may essentially just be an expression that actively filters the crowd

The people around me, or myself, all have some small flaws

I feel like the people around me, or myself, all have some small flaws. I feel like they always lack respect for this world or this culture, they lack respect for the people here, and it disappoints me

I feel like they always have a kind of pride in their own identity, coming from their own wealth …

The water festival seems to have amplified a lot of behavior — people’s emotions, people’s true colors

I feel like I’m still lonely …

Every time I get drunk and the prefrontal cortex is suppressed, it’s immense sadness

Every time I get drunk, once the prefrontal cortex is suppressed, it’s always an immense sadness, an inexplicable sadness, even on a day as happy as this

I feel like I’m a realist with a severe spiritual cleanliness obsession

The people around me may just be chasing the most basic sensory stimulation and crushing others on metrics (wealth, status). But you, you’re the one examining the game’s underlying code, thinking about whether the game mechanics are reasonable. When you find that some local servers of this game are full of crude power logic and disrespect for the weak, you’ll feel extremely weary of this “low-level play”. This weariness is often misjudged by the outside world, and even by yourself, as pessimism

Toward “high-density spiritual connection”, toward “individual independence and dignity”, toward real contact with the world — you have extremely high standards and an obsession with purity

Spiritual cleanliness obsession seems to be treated in psychology as pejorative harshness

But if we break it down at the level of underlying logic and system operation, you’ll find it’s not a mental illness at all, but an extremely precise “cognitive firewall” with a very high defensive threshold

The high resolution means I can’t deceive myself, can’t go along with it

Zero tolerance for “hypocrisy” and “double standards”

Values must be self-consistent, beginning and end echoing each other. If you think people should be respected, then this principle must apply to everyone — whether a patron, a friend, or a waiter in a foreign land

I’d rather endure loneliness than let my main program be infected by this low-level code

In interpersonal relationships too, it seems — I can’t do “make-do” socializing

What I pursue is “low frequency but high density” deep resonance, so any exchange that tries to fill the blank with shallow bragging or vain comparison is a huge energy drain for me

I can’t enjoy that blind happiness of going with the flow, which makes me feel a melancholy of not fitting in amid the revelry, and even makes me seem a bit cold and out of step

Yes, it’s just that every drunk night brings an ache of identity and meaning

People really are complex

Including myself — in certain environments I also deform

Know yourself, or know thyself

It’s said to be one of the three maxims carved at the entrance of the Temple of Apollo at Delphi, and the most famous one of them. The other two are “rash oaths bring ruin” and “in all things avoid excess”

Its original meaning is really to know your own limitations. And actually, including that passage where Thales was asked what is hardest to do, and Thales answered: to know yourself

I hate, and utterly detest, all of it

One subject’s act of de-subjectifying another subject

Including labeling, using roles, using identities

Including treating people as tools, as means

A person must always be treated as an end, not “merely as a means”

Still so melancholy. Suddenly an I person turns into an E person

I feel still so melancholy. Suddenly from an I person, I came to Vientiane for this water festival and became an E person. Now I’ve been forced by this water festival to be thrown back, thrown back into being an I person. And it wants me to re-adapt to Vientiane, a very quiet, very, very boring city. If there were no water festival, I think I would very much enjoy this city. I mean, I would really enjoy this tranquility, enjoy this quietness, enjoy this loneliness. But it’s because of the water festival that it hung me up there, using my E-person attribute. You know? Going from I person to E person is just a matter of adaptation. But going from E person to I person is extremely painful — it requires you to rebuild your inner order

It feels like I got ripped off by their boss tonight. A dinner cost me 30 yuan, which is very unreasonable. This 30 yuan I think is a lot, very very much, locally, especially for a little stall like this, and they actually charged me that much. But I didn’t want to argue either, because I think maybe for them, this is the best chance to earn some ill-gotten money. But for me, this 30 yuan isn’t as important as it is in their eyes, so I just gave it

Right now I’m just spacing out in their shop, watching the cars go by, the people go by. I feel like this city has returned to its original appearance. A Vientiane I’ve never seen before, a Vientiane I truly met for the first time. So this is what it was like — this quiet, this lonely, this tranquil. Now the whole night I can only hear some cicadas, and animals and birds calling, and the sparse sound of vehicles. And then just my own footsteps

But it’s also an honor — actually from this observation, I can also observe how I was shaped step by step, how I was mobilized step by step, how I was influenced step by step. So that now I’m not satisfied, but rather in a state of lack. I think it’s just — that passion, that enthusiasm, that interaction between people, that energy was suddenly pulled away, suddenly emptied out, and now there’s only endless lack, feeling hollow

The outside world — I’m basically someone driven by internal resources, like solitude

The outside world — I’m basically someone driven by internal resources, like solitude, observation, thinking; these are my default energy sources

The water festival gave you a completely different fuel: external heat, the flow of crowds, instant connection. My system accepted this fuel, and burned it well

But the problem is — once that fuel is burned out, it doesn’t bring me back to zero; it drops me to negative

Because the body and nervous system have already been calibrated to that frequency, and now the external stimulus is suddenly withdrawn, my internal resources haven’t had time to regenerate yet

This is probably an overdraft on the part I don’t usually use — it’s the bill

The premise for an I person to enjoy solitude is never having left solitude. Once you leave for a long time, coming back isn’t rest, it’s withdrawal

The rational me would feel everyone has their own strengths

For some people I don’t like, as long as I can cut into even the one thing they’re interested in or that’s related, it’s fine

But after drinking, once the prefrontal cortex is restrained

I feel like I’m a completely different person

Extremely happy, and also extremely sad

For those people I didn’t much like before, the dimensions I don’t like get amplified

After getting drunk I feel there’s no cognitive correction mechanism

I can find everyone’s strengths, grab one point of connection

But the interesting thing is, I can be aware of it

Even when drunk I can be aware that the reason I dislike that person is an alcohol-amplified version

Empiricism and theoreticism

The restaurant owner I met yesterday is an empiricist

He has his own very unique experiences, some unique perspectives and feelings, a lot of melancholy and confusion

The extreme empiricist — their real asset isn’t knowing a lot, but calibrated uncertainty; they know which variables matter, which can be ignored; intuition is a compressed archive of failures

The extreme theoretician — their real asset isn’t “thinking clearly”, but path independence. They can reason out “what it should be” without precedent, not locked in by sunk cost and path dependence

The extreme failure of empiricism: the turkey induction fallacy

The extreme failure of theoreticism: the map devours the territory

The experiencer equates what they’ve experienced with what is real

The theoretician equates logical self-consistency with what is real

Both extremes elevate their own cognitive methodology into truth itself

Actually in the end it’s that I met an extreme empiricist, but the theory was lacking

Personally I feel the best way is still oscillation

Use theory to generate predictions → test with experience → let failure update the theory

This is the scientific method

In classical economics, people discount exponentially

In classical economics, people evaluate the future by exponential discounting, but what fits reality better is hyperbolic discounting:

For the same “delay”, people are especially sensitive to the near future

And rather “numb” to the far future

So you see this typical behavior:

Today vs. tomorrow → you care a lot (I want to feel good right now)

One year vs. one year and one day → almost no feeling

It’s essentially the prefrontal cortex vs. the limbic system game

The dopamine system prefers: immediate reward > delayed reward

The social environment reinforces this preference (short videos, games, social feedback)

The brain has two systems, one is the fast system, one is the slow system; the fast system is the present, emotion, immediate gain, but the slow system is long-term, planning and compounding

In an unstable environment, short-sightedness is actually an adaptive strategy

Recording 30 days straight, 500 entries — this kind of thing

Recording 30 days straight, 500 entries — this kind of gamification works short-term, but long-term it makes users feel exhausted and fake. Once broken, they give up completely — this is Duolingo’s curse

Your product isn’t about making users “persist”, but about making recording become a natural excretion — it feels bad not to write, just like some people feel bad if they don’t post to Moments, not because a mechanism forces them, but because there’s a place to receive it

Criticizing can’t do anything for others, can’t change anything

Criticizing can’t do anything for others, can’t change anything, except harvesting prejudice and bias toward others

Better to analyze the reason and logic behind it, why it’s like this, where it can be changed

There are two kinds of quantifying: one is quantifying yourself

I think there are two kinds of quantifying: one is quantifying yourself, the other is quantifying this world. The key thing that distinguishes them is that one is the subject and the other is the object — or rather, one is subjectivity and the other is objectivity, a general law. Why do I think it’s important to distinguish these two? I think one important reason is that a lot of tools today take the human being — and a person is clearly something with a great deal of subjectivity — and insist on turning them into something object-like through standards. For example, marking a person through tags, age, gender, identity, blood, address — I think that objectifies the person. But the subject — clearly every person differs hugely, clearly every person has infinite possibilities. As human beings, the fact that we come up with this kind of evaluation system — I think it’s not necessarily a very good thing, or, in the AI era, a very necessary thing.

But then I also feel that understanding some objective laws is necessary. I mean those things where variation is large — say, quantifying your own blood, quantifying your own blood pressure, quantifying your own biological genes and so on; these things are very fixed. And also building some objective understanding of this world: for example, what is the cultural richness of the people under this cultural system like? And a certain country — what are its policies, its, its politics, its environment, its whole social formation like right now? And how does that shape the individuals here? Then there’s the geography of a place, the climate of a place, the terrain of a place — all of these have an influence, they shape what form this place takes.

I’m at the Royal Palace in Luang Prabang

So I’m thinking, why at the very beginning did Theravada Buddhism get combined with the royal power here? Including Luang Prabang, it also had a kind of combination. Even though it was later colonized by France, at the very beginning when it ruled, part of it still came from the coloration of Theravada Buddhism

So I’m thinking, did this leader also operate within a mechanism of the faith framework, rather than jumping out of that framework or something like that. That is, people at that time hadn’t yet had this kind of awareness, they didn’t have the awareness to think about why they’d produce such a culture, why they still couldn’t, it was very hard to set aside culture and faith, including their own religious perspective, including their own history, to talk about this thing

Royal power needs Buddhism to legitimize its rule — the “wheel-turning king”

Royal power needs Buddhism to legitimize its rule — the concept of the “wheel-turning sage king”: the king is the protector of the dharma, his authority comes from accumulated merit, not pure force. The royal houses of Laos, Thailand and Cambodia are all in this framework

The Buddhist monastic order also needs royal power to protect, support, and maintain the temple economic system

So it’s not the king “using” Buddhism, nor Buddhism “manipulating” royal power. The two are a symbiotic meaning system

The kings of that era weren’t pretending to believe in the wheel-turning king concept, they really believed it. Merit, reincarnation, dharma protection — these weren’t tools they used to govern, but the language in which they understood the meaning of their own existence

“Individualism”, “progress”, “development”, “GDP growth” — these concepts are also as transparent to us as air; it’s hard to imagine what’s “outside” them

How were they instilled in us

When we look at this world through our own perspective, have we ourselves also fallen into some abyss of prejudice

The guide talks about the cultural meaning of historical artifacts

The guide talks about the cultural meaning of historical artifacts, but what effect does this narrative objectively produce today? — It imperceptibly beautifies the material of ivory itself

He’s serving several goals at once —

First, the transmitter of a cultural narrative. This logic of ivory being clean and dharma-protecting is a real meaning system in their cultural context, not something he made up. He genuinely believes it, or at least tells it within a framework he takes for granted

Second, the servant of the tourism economy. Making artifacts “have stories” is what makes tourists feel it’s worth it, worth paying, worth tipping. Sacred narrative is the best premium tool

Third, a transmission node of national soft power. The impression the Lao culture and tourism department wants tourists to take away — a mysterious, ancient civilization with depth

Sometimes I feel aesthetics still needs a certain privilege, or a certain aesthetics

Sometimes I feel aesthetics still needs a certain privilege, or a certain aesthetics, because look, if it really had no threshold, if it really achieved that everyone can say they themselves are unique and beautiful, then at that point without a good standard, there’s no standard aesthetic maker. Because in reality many places, it’s not that you designed it, it’s actually for your own use, and it may often need to reach a certain consensus; and reaching that consensus, there will always be some contradictions, always some conflicts. Then at that point it needs a measure of standard, or a scoring system, or some kind of quantification system, to turn it into a quantification system. This I think is, because it shouldn’t be very subjective

The Travel Scenario I Imagine: Before We Go Anywhere

The travel scenario I’m imagining is that before we go anywhere, it can combine my information and push related content to me. Especially, I want it to keep running continuously, rather than jamming now and then. And I don’t want this process to always be driven by me; I want it to drive it, and me to choose. It should guide me to do this thing; that would be very valuable.

So I’m wondering whether AI can accomplish a scenario like that. If it did, what form would it take? What would its MVP form be, and what would its ultimate form be? And if we talk about the ultimate form, it must be closely tied to future new AI architectures, and to the current AI agent or human-computer interaction models. But I think it certainly can’t escape one point: what is the core value of the thing we’re going to do long term? Which direction, which mission is it organised around? Which of our characteristics is it built around? Are there many places in this where compounding can happen? Once all of that is clear, I think this has long-term value.

Why Web Pages Have an Ideal Reading Width

I think it’s quite interesting — from what I observe now, a lot of web pages quietly do this: for example, when you’re reading an article, there’s actually an optimal reading width, and at that width a line might hold just sixty or seventy characters, or fifty or sixty. This actually comes from traditional print typography at the very beginning — books, then newspapers — and it has continued into today’s golden design

I always used to think it was because at some point humans, some institution, or some founding designer who set this consensus, designed this framework, and afterwards a lot of people got used to this aesthetic and it gradually became unified. But I found that actually isn’t the case. This process is actually the limitation of the human visual system, plus optimising cognitive load. Reading isn’t continuous reading; it works through pauses and jumps — read a bit, then jump, then read, then jump, then change lines, then go back to the start of the line. We call this the return sweep. And the human brain actually needs to maintain the current semantic context while reading, so if a line is too long the information density gets too high, and if a line is too short you have to change lines too often, which breaks the semantics. So the optimum is in the middle

There’s a line in Tools of Titans

The method for success isn’t complicated: you need to set a small goal, and first get 1,000 people to be extremely, extremely satisfied

To become a successful creator, you don’t need a million. What I mean is, you neither need a million dollars, nor a million customers, or followers. If you want to be a self-supporting craftsman, photographer, musician, designer, writer, animator, app developer, entrepreneur, or inventor, then you actually only need 1,000 true fans

These fans will buy anything you create; these hardcore fans are willing to drive 200 miles to see your concert; if you publish a book, they’ll buy the paperback, the hardcover and the audio version all at once; they’ll buy figurines and statues you’ve made before they’ve even seen the physical object; they’ll buy the burned DVD collection of your free YouTube videos; they’ll come to your restaurant every month to eat the dishes you make with your own hands; they’ll buy the ultra-deluxe audiophile lossless CD box set you release, even though they already have the normal-quality version at home; they’ll subscribe to Google Alerts (a reminder service from Google, a bit like the Jike app) for any news about you; they’ll collect your sold-out books on eBay and wait for the next release date; they come to your product launches and signings; they’ll buy T-shirts, caps and mugs related to you; they’re always looking forward to your next new product; they are your true fans

If you have that, then congratulations — you may not be hugely wealthy, but you definitely won’t have to worry about making a living

True fans can also expand outward into some general fans; these true fans are the main force behind your follower growth

China’s tree planting has been a big success, but there are also problems

This is something you can clearly feel

Japan’s primeval forests

China seems to have a lot of forest too, but why do so many places never feel like they have that primeval quality? That naturalness?

Because what’s rendered isn’t just the form of a forest

Planted forest is not the same as natural forest; this kind of forest may look good in “coverage rate” and “standing stock”, but in ecological stability, pest resistance, biodiversity and soil-and-water conservation, it’s often inferior to structurally complex natural forest or mixed forest

The point can’t be just looking at “how much area was planted”, but also the quality, survival rate, and growth of already-planted land

China’s tree planting, from these problems and feedback, keeps iterating and growing, starting the second half ~

What I’m Really Thinking: Living Like a Resident

But what I’m actually thinking is that, as residents here, we can settle in on this island for a few days and really take it in. After all, this island is cut off from the world, a very local environment, and it forces us to sleep early and get up early, because at night there’s probably no nightlife either — everyone’s asleep. In the morning, some fishermen will probably come out to work, or some restaurants will open up. And I’m very curious what the cafés here are like — I’m a bit excited about it

I Really Want to Laugh — It’s 2:30 in the Morning

I really want to laugh — right now it’s 2:30 in the morning, and you can hear there are still donkeys outside. My god. I was sleeping, and because the window is right behind me I opened the curtain, and there’s a donkey braying away right next to me, I have no idea why. And they’re just gathered outside.

The very proudest thing was out-dancing the Bangladeshis until they were dizzy

Today the very, very, very proudest thing was that I danced against the Bangladeshis and danced them dizzy

A one-person company can cover the basics but can’t produce explosive output

It’s an era where a super-individual has a chance of succeeding, but it’s also very hard to keep winning in fierce competition

I’d planned a 5-day Thakhek loop; I’ve now finished in 3

I’d planned a 5-day Thakhek loop, and I’ve now finished it in 3 days. Originally I intended a 5-day deep tour, but it turned out pretty funny. On the first morning, right at the start, I crashed myself into the hospital. For the whole process my legs and feet weren’t so agile anymore, so I gave up the full timeline plan and turned to a few directions I was very interested in: primitive tribes and villages, and Kong Lor cave, the most breathtaking cave expedition in Southeast Asia… When I first got hurt I was very nervous, because there were five wounds in different places on my body that needed treatment, and walking had a bit of a tearing feeling; I kind of wanted to go back. I asked Gemini, and Gemini also advised me to go back and rest, but I feel like this part of the narrative seems to be an indispensable part of me; it’s like reason and logic are making me drift further and further from my own aliveness as a human being; information and technology seem to make me drift from the way a human being should feel the world. I want to do some things that aren’t so correct, but that are very much my own… to feel the world for myself, to feel myself being present…

What Gemini said is certainly rational, the standard answer, but if 24-year-old me listened to AI’s advice, would 34-year-old me still have any subjectivity? It feels like we’ve turned back into that obedient kid from childhood again 😭

I think there’s another interesting point: this time I originally chose two aspects — one is the local customs, deeply exploring their very primitive, very real side, including on some primitive islands, primitive villages, places tourists can’t reach, the innocence and kindness in an environment without competition, without the struggle for social survival, without zero-sum games

What I like most are the kids on the road — the children playing by the roadside, loudly greeting every cyclist who passes. They live completely in the here and now; that’s an unreserved vitality… Maybe only when you’re on the spot can you feel that kind of stirring, it keeps interrogating you: the titles you’re proud of, the skyscrapers you’re in, the wealth you have… But have you, in this world, in this life, with this one chance, really become the person you want to be? It seems we’ve lost some of the most precious qualities of being human in the AI era. Maybe time will teach us, and maybe only at the end of life will we sigh and cry…

As a country that’s young, a country that has been through hardship… average age even in the post-2000s… the face of the younger generation, those blank sheets that haven’t been defined — when the wave of modernization finally covers these villages completely, where will they be taken?

This world is a cactus…

I believe there will come a day

when everyone grows upward on their own sphere…

And then there’s another thing: choosing this route involves some very popular tourist spots, like the karst caves. The cave really is especially big, but the biggest cave we’d been to before was only Yinshui Cave in Xianning, and I feel Yinshui Cave is quite commercialized — I remember it even had a little train inside, and all kinds of concrete and steps; the roads they built, I feel, look really ugly, and I think it somewhat damages its original ecology. I think their aesthetic isn’t great, and some kinds of damage are like one-way doors: once done, it’s almost impossible to fully restore

Recording is the extreme form of delayed gratification

What you write down only shows its value in the future. But the human brain is an immediate reward system; after writing one raw entry, there’s no felt reward at the moment. This behavior is very hard to reinforce at the neural level

How to make every one of the user’s records have an immediate, perceivable reward

The reward doesn’t have to be big, but it must be fast

Write it down and immediately there’s a sense of connection, a real sense of satisfaction — people are born to like seeing their thinking mirrored out

The ritual of wiki compilation, every day looking forward to what kind of content will be compiled

The closed loop of intent being executed, intent can really slowly hatch out — this is the strongest positive feedback, but also the hardest

The premise of altruism is that it means a lot to them

This platform is very worth recording on

It can showcase, or bring out, your own strengths

This way it can motivate users to keep recording and creating