May 2025 Thought Notes: Self-Knowledge, Engineering, AI and Agent Systems

Rather endure suffering than leave the comfort zone that makes you suffer

Out of fear of the unknown, people would rather endure a familiar pain.

Human nature has a stronger preference for the familiar, even when those preferences aren’t very good.

With a known pain, people seem to know better how to face it.

To know others is wisdom, to know yourself is enlightenment

Most people measure the world by their own thoughts and viewpoints. The meaning (or viewpoint) we assign to a thing ultimately determines how we feel about it.

It’s also a person’s worldview, outlook on life and values that determine their viewpoint on a given thing. Even for work: some people find work very happy, others find work very painful.

If you think about it, who would you become if the thought “I hate work” were no longer there?

A lot of people have never thought about this — as metacognition, understanding the existence of emotion itself.

When you’re angry, reflect on the reason for the anger, on what the essence of anger is, record it objectively, and it will naturally resolve itself.

To know others is wisdom, to know yourself is enlightenment (2)

To know others is wisdom, to know yourself is enlightenment.

Most people measure the world by their own thoughts and viewpoints. The meaning (or viewpoint) we assign to a thing ultimately determines how we feel about it.

It’s also a person’s worldview, outlook on life and values that determine their viewpoint on a given thing. Even for work: some people find work very happy, others find work very painful.

If you think about it, who would you become if the thought “I hate work” were no longer there?

A lot of people have never thought about this — as metacognition, understanding the existence of emotion itself.

When you’re angry, reflect on the reason for the anger, on what the essence of anger is, record it objectively, and it will naturally resolve itself.

Thoughts on the evolution of the brain

The brain’s complex computation is aimed at the instinct for survival and safety awareness since human evolution. But when it comes to the happiness and joy of the mind, it’s powerless.

The brain’s task is to maintain survival; the mind’s task is to create a sense of fulfilment.

If you want freedom, happiness, calm and overflowing love, you can’t just let the brain give the orders and settle for physical health; you have to break this limitation and pursue a higher spiritual health.

Thoughts and ideas

Thoughts and Reflections

What flomo records is the instantaneous cognitive fragments in the process of thinking.

Ideas are psychologically driven, and they’re the material with which humans construct a sense of time. Ideas are the premise of all human experience.

We can reconstruct our own process of a thought popping up — this definitely isn’t controlled by the brain; it forms spontaneously, and the human brain can’t even control the emergence of a thought.

Thinking is the processing of ideas. Thinking needs the human brain involved and consumes resources, so thinking is very difficult. Humans would rather bear pain than actively think.

But it seems that thinking is also the root of suffering.

But it seems that an optimistic mindset often isn’t arrived at by reasoning, but is a more instinctive, more intuitive state.

So often the first reaction to something is your own state of feeling, which is closest to the truth, rather than the later decision after thinking, after reason or after emotions.

Even so,,, isn’t thinking really important? This seems a bit conflicting? Thinking leads to unhappiness again?

A person lives, has to grow up, has to figure something out.

Growing up always involves thinking; you need it to go beyond the intuitive and superficial, to figure out what you actually want and how to do better.

So stopping thinking & actively thinking is itself a kind of balance.

The essence of experience

Trekking experience, slow-travel experience, social experience, all human experience.

If humans learned only one thing, it would be enough to change the world, and that is: don’t be afraid of experience.

People are born a bit afraid of this and that. Afraid of failure, afraid of losing face, afraid of the unknown, afraid of hassle. This fear keeps us in the comfort zone, afraid to take a step forward. But the thing is, all growth, all change, is hidden behind “give it a try”. If you don’t try, you have nothing.

Fear makes things feel terrible. A lot of people now get a headache the moment they hear “AI” or “programming” and say “it’s too hard, I can’t”. And what happens? If you don’t dare try, you really won’t be able to, and you gradually get left behind.

There’s another kind of fear: fear of facing yourself. Afraid to admit you were wrong, afraid to see clearly what you want, afraid to be different from other people. Too much of this fear and a person lives cramped, and getting along with others becomes awkward too.

Don’t always shrink back. Dare to try, dare to feel — that’s what makes life interesting.

Some new thoughts on flow

“Anxiety is thinking without restraint. Flow is restraint without thinking.” — James Clear, habits researcher

I’ve discussed a finding before: I feel I reach a flow state more easily when I’m slightly tipsy, or when I’m sleepy. This phenomenon baffles me.

Humans perform at their best precisely when they enter a thought-free, self-forgetting state. Japanese has a beautiful adjective for this state: mushin (no-mind).

No-mind means no distracting thoughts in consciousness, no resentment or anger, no worry or fear, to the point of no self.

Only with the mind free of distractions can you escape the limits of the self and create all the variety in this world. I don’t want to force this belief on you, I only hope you’re willing to try it and experience it, and grasp it through your own experience.

It may be the same as that teacher earlier who forgot his lesson plan between classes — the same way, he reached a flow state more easily.

So the essence of anxiety is thinking too much, and the essence of flow is having no distractions.

So back to an earlier question: why do I more easily reach a flow state or a focused state when tipsy, or in a tired state?

When tipsy, the prefrontal cortex (PFC) — the region in charge of planning, criticism and self-monitoring — has its activity mildly suppressed by alcohol, which is a bit like the “ego dissolution” effect after meditation.

Tiredness may be a bit different: executive function declines -> less criticism/distraction, similar to the flow effect, but actually it’s because resources are insufficient and the brain automatically processes in “energy-saving” mode, only focusing on the current task and no longer handling multi-tasking interference.

It may “look” easier to enter flow, but in substance it’s entering a “narrowed attention” state, not genuinely efficient flow.

Usually early morning is the golden time when it’s easier to enter genuinely “high-quality flow”: the brain has just restored its resources, and executive function is at its most complete (including working memory, suppressing distraction, emotion regulation).

This may be why artists all need an environment, music and a little wine to activate their artistic cells.

Achieving goals in inspiration or in desperation

Goals have two sources: triggered by inspiration; born from despair.

Let inspiration trigger dreams, rather than letting despair birth goals.

Driven by pure inspiration (rather than despair), there is a deep and strong impulse to create something that makes the whole world see it fresh.

But how do you tell them apart? A simple way is to remember the difference between ideas and thinking. If a goal or dream is born in a flash, it was triggered by inspiration; if you think it over and over and finally settle on a certain goal or dream, it was born from despair.

As we keep saying, earn enough money and then go travelling — goals like this are means serving an end, not the end itself. We always want to achieve these goals out of all kinds of motives, so we always feel there’s a missing piece deep inside.

A goal itself isn’t about right or wrong, only about whether its source is inspiration or despair. It depends only on how you feel inside. Once you realize the difference between these two kinds of goals in origin and in form, you’ll feel incomparably happy that you can create things that make people see fresh.

We don’t create in order to attain fulfilment of body and mind; rather, because we feel that fulfilment, we want to create, to give without expecting return.

What do you really want?

If I had money I could never spend, had already been to every corner of the world, nothing could worry me, and nothing I did could be criticized — then what would I want to do, and what new things would I want to create?

Trusting intuition

I’ve always had a sort of rejection of intuition. Then it suddenly occurred to me: why the rejection? Because to a large extent my reason tells me intuition has a certain error rate.

If you try to seek approval from the outside world, you will almost without exception receive only strong opposition and contrary opinions. Don’t look for answers from the outside world.

Be loyal to your intuition, your sixth sense and your inner wisdom, and trust the universe. If you do this, unexpected miracles will appear in your life. People who have such faith and courage, while experiencing the miracles of life, also find the calm, love and joy they have always yearned for.

Have the courage to follow your heart and intuition — they somehow already know what you want to become. Everything else is secondary.

Intuition may whisper to you that you should go start a conversation with some stranger in a café, opening a beautiful friendship, or that you should call a friend, only to find he’s been having a rough time lately and needs a friend’s encouragement and comfort.

A lot of things don’t necessarily need reason to hold your own heart down; you must have the courage to face your own intuition.

If you want to see scenery you’ve never seen before, you must dare to throw yourself into the unknown; first of all, put down your worrying and listen to your inner intuition.

Thinking about answering every message, again

Is answering every message really so hard to do?

Why can’t I do it?

Why do I ask other people to do it?

When others don’t do it, whose fault is it?

I don’t think every message needs an answer, but what if the other person needs every message answered?

Replying to everything feels very stiff to me.

The other person seems to have got angry because I didn’t respond.

But I feel I wasn’t wrong.

Why is that?

There’s a very good method called tracing back to the source

When we watch how someone else outputs a viewpoint, what they’re often doing is tracing back to something distilled from their own experience.

The interesting part is that what I’m looking for is the other person’s experience — how they drew that viewpoint out of their experience.

Everyone has viewpoints, but experience is precious. The more first-hand the experience, the more it can form distinctive thinking.

Human nature before and after reform and opening up

Gannan Jishi, Ten Years of 100 People.

It feels like human nature never changed? But why can we understand the evil of human nature after reform and opening up, while the evil of human nature before reform and opening up seems hard for us to tolerate or understand?

The essence of human nature hasn’t changed.

Longing for a better life (self-interest).

Fear of losing resources and status (a tool of the survival instinct).

Pursuing fairness yet not refusing privilege (contradiction).

Breaking your moral bottom line at any moment in the face of temptation (fragility).

In extreme survival conditions the evil in human nature is more easily amplified; it seems more like a survival instinct driving it.

People who have been through pain better appreciate how hard-won happiness is

People who have been through hardship often better appreciate how hard-won happiness is, and better know to cherish what’s in front of them.

It’s also really about your own inner sense of fulfilment.

We’re actually already very happy; I’m already satisfied.

At least every day is a happy life.

It’s the life I wanted to live, the things I wanted to do.

How nice.

Inheritance

It seems the times and technology are progressing, and tall buildings give us a lot of illusions, as if the world really were different.

What is the essence of change? Technology’s relentless advance.

Values are changing too; every generation will rise up and create some better values for a new era.

My grandfather’s generation spoke of “the collective”, “endurance”, “sacrifice”.

My parents’ generation spoke of “striving”, “making something of yourself”.

Our generation pays more attention to “the individual”, “freedom”, “meaning”.

Every generation’s worldview is changing fast too.

But we still seem to want to escape our family of origin, just like our parents’ generation, the struggles they went through as children too — it’s just that they did it for a living.

Nothing has changed either.

Our longing for love, the deepest need of human nature.

The search for belonging, only constantly changing form.

The search for meaning; every generation searches for its own meaning.

It feels like every generation has its own troubles, but people of their era can never quite understand the previous generation’s ideas, even though their qualities are the same. They all have a kind of courage that doesn’t yield to life, a breaking of traditional ideas, a pursuit of the life they long for, a pursuit of happiness. Human nature is always like this: we always think about what we haven’t got, and don’t cherish what we already have.

Generational conflict

As I said last time.

Every generation has its own troubles, and every generation has its own mission.

But they are essentially “the same kind of people” struggling in different worlds.

Every generation challenges the “order” it lives in in its own way: our parents bravely smashed the feudal thinking of the generation above them, we pursue freedom and independence. It looks like we don’t understand the previous generation, but we’re actually repeating a similar struggle — only the script and the stage are different. So understanding generations isn’t about what they did, but why they did it.

Criticizing the past seems to have become meaningless. The reason I think I’m smarter is only that I’m closer to today.

Every generation is reconstructing “what happiness is” and “what success is”.

Before my grandfather died he warned me to act carefully, to have food to eat, someone to rely on, to live with dignity — that was his definition of happiness.

My parents are the reform-and-opening-up generation, the era of the fastest economic growth; that generation’s happiness was making something of yourself, having a house and a car, and their children doing well — that was their greatest wish.

My understanding of the essence of happiness is that among too many choices, happiness is living into the shape you like, living as yourself, cherishing the present.

Happiness is being alive -> happiness is success -> happiness is being real.

Breaking traditional social expectations

It disguises itself as “taken for granted”, and generation after generation replicates its own fate through “should”, “must”, “you’ll understand when you’re older”, projecting its own unfulfilled expectations onto the next generation.

Just like the ama divers whose fate is to hope the next generation will be ama divers too.

The fate the parents’ generation couldn’t break free of — to make “the price meaningful” — makes them subconsciously demand that their children repeat it.

“I went through exactly this hardship back then, and you have to endure it too” → rationalizing repression;

“Girls should be obedient and take care of the family” → internalizing gender roles as norms;

“Only a civil service job is stable” → packaging social fear as “for your own good”;

“What marriage isn’t full of grievance” → treating endurance as a sign of maturity;

“I raised you this big, not so you could just do whatever you want” → treating children as a means of self-continuation, rather than as independent lives.

Remembering home

Maybe one day I’ll go back to the hometown I once wanted to escape, and the memories will reappear before my eyes bit by bit, but I’ll never be able to walk back into them.

If I had known that the spring of that year was the spring of my life, I would have dared to live it more thoroughly and exquisitely.

What we love, what we hate, what we want to return to, what we want to leave — all of it is that self of ours that needed to escape or to pursue.

When young we long for distant places, when old we miss the way home.

Saying goodbye to the past is also accepting yourself.

I heard a line: if parting is destined, then what is the meaning of meeting? The meaning of meeting is that the part of me you changed stays by my side forever, in your place.

So once again I felt ichigo ichie and mono no aware. Parting isn’t a break, it’s internalization.

Nothing truly exists; only some things we believe to be true

Everything you go through is really just your interpretation of the world.

The world isn’t real; it’s only constructed by our perception. Our subjective consciousness is what’s real.

So this also explains that line: it isn’t the thing itself that causes pain, it comes from our judgment about the thing.

So the more you resist something, the more it persists; the more you want to control something, the more easily you get controlled.

Real freedom isn’t doing whatever you want, but having the power to not be driven even when facing an inner impulse.

Metacognition can help us become aware of such impulses, and has the ability to help us balance such impulses — that is real freedom.

Freedom isn’t “following your feelings”, it’s “knowing you have feelings, and still being able to decide how to go”.

After a child dies, they grow up in the parents’ hearts

Losing a child often brings deeper trauma that’s harder to recover from: crying at the thought of it, seeing them again only in dreams, the repeated tearing upon waking — guilt, self-blame, powerlessness, repetition.

Losing a child is very painful, your face washed with tears — but what else?

In reality the child’s growth was cut off, but the parents will keep imagining, remembering, even “constructing” in their hearts the life trajectory the child never lived.

They will imagine: “If he/she were still here, how old would he/she be now? What would he/she be doing?” This is a continuation of emotion, and also a way of mourning.

The year I was born, my second aunt’s only boy also died by accident. My second aunt was especially good to me from childhood — no matter how much trouble I caused, my second aunt forgave me.

Being self-consistent without depending on any identity

Society needs labels to identify and classify.

But there’s a problem: once you start depending on those labels, you start living inside other people’s expectations.

Identity brings a false sense of certainty, but once these things waver we start to panic — without them, what do I still have?

Who you are isn’t what’s written on your business card; it’s the “self” that can still exist calmly after losing those labels.

And so you fear leaving your job, you fear society’s definitions, you fear what the people around you think of you, and you end up lacking the courage to act.

Ask yourself a question: would you rather be seen as your whole self, or as your labels? If today you lost every title, every bit of income, every connection — who would you be then?

I’m willing to keep creating, I’m willing to keep exploring and enjoying this world.

A sense of realness

I long more and more for a sense of realness, and yet I couldn’t describe the feeling in a single moment.

Is it telling the truth? Is it being the kind of person I am?

A sense of realness, I feel, is more about knowing my true feelings, motives, limits and longings in the present moment, neither denying them nor beautifying them.

1. Awareness — what am I doing, why am I doing it, do I really want it?

2. Honesty — I don’t cover up, and I don’t pretend to be an “ideal person”.

3. Consistency — my words, my choices and my inner state line up as much as possible.

Get out of performance, and make something that truly belongs to you.

Step back from “how do I want people to see me” and gradually turn to: “what do I want myself to look like in my own eyes”.

What’s behind a relationship is value exchange

Material value, emotional value, spiritual value.

The reason to try morality comes from reflecting on the selfishness of human nature.

So in relationships, a person should stay clear-eyed and separate interests from feelings.

We don’t deny that a lot of feelings are great, selfless and noble, but if we take this sacred feeling as the truth of human nature, we’ll very likely end up far from the truth.

The fear of amplifying the level of detail

Sometimes I feel the same: when you first meet someone it’s like looking at a painting from far away — you think it’s beautiful and simple, but once you walk closer, it really isn’t the same. Like you said, after you zoom in, the whole world changes appearance. It isn’t that the person changed, it’s that I see more: the details, the small flaws, even some complex emotions hidden very deep, all come out. Sometimes I wonder, is this a good thing or a bad thing?

Actually, I think it’s a lot like adjusting a camera’s focus. From far away the picture is blurry but comfortable; up close, clear is clear, but also messy. You discover the other person isn’t only the way you thought they were; they have many sides, and sometimes even they haven’t fully figured themselves out. I was chatting with a friend once, and he made an offhand complaint, and I suddenly realized that under that cheerful look she usually wears, a lot was actually being pressed down. In that moment I froze, thinking, have I never really understood her? Once this “level of detail” in a relationship changes, you have to readjust yourself or you can’t keep up.

Philosophically, I think this is a bit like choosing the resolution at which we see the world. Stay far away and it’s simpler, and you live more easily; get close and there’s more truth, but it costs more brainpower. Sometimes I ask myself: how much do I actually want to see? After seeing these details, can I still get along with this person the way I used to? Because once you zoom in, you can’t go back. You have to accept that they aren’t perfect, that I’m not perfect either, that all of us are a messy tangle, and then still try to find some harmony in that tangle.

It’s like listening to music: at low volume you only notice that the melody is nice; turn it up and you find there are also noises, harmonies, even small flaws left in the recording. Same with relationships: up close, what you hear isn’t just the sweet surface talk, but the things left unsaid, things even they may not be aware of. That’s pretty scary, but also kind of interesting. Because at that point you’re really dealing with a living, breathing person, not a concept.

So I think intimacy is quite contradictory. The closer you get, the more easily a simple relationship turns complex; but if you don’t get close, you stay on the surface forever. Sometimes I think that a real relationship may just mean daring to go deeper, even knowing the water below isn’t all that clear.

The rise of Japanese anime

Talking about Japanese anime, we can clearly see that anime has become the receiver of the psychological gaps of modern people.

Why? On one hand, an economic and material downturn needs spiritual energy replenishment, and culture is a very good container.

On the other hand, fictional worlds fill in the lost spiritual coordinates and emotional needs of the real world. People want emotional value but don’t want to pay a price for emotional value, and anime becomes a very good choice.

A kind of collective psychological compensation mechanism (escaping from and repairing a failed reality).

A kind of substitute spiritual life system (the virtual world receives meaning).

A kind of “post-industrial society” aesthetic expression (cyberpunk, the metaverse, moe culture).

And there’s also a cultural factor, which is in fact one of the decisive elements.

A blend of Confucianism, Buddhism and Daoism + secular aesthetics: allowing “fiction” to carry “the real”.

The sense of beauty isn’t a big blaze, but a faint sadness, a beauty that vanishes in an instant.

The small beauties in life, the small happiness they generate — these are actually some psychological medicine for heavy reality.

The mechanism of emotional transference

I said before that China’s development needs to be viewed through a generational lens, because the gaps between each generation’s upbringing are enormous.

There’s a type of person whose childhood environment brought distrust and insecurity, giving rise to a strong desire for control, transferring inner unease onto the people around him. The core of this is really the nature of emotional transference.

The security he wants is something he tries to grip tightly by controlling other people, but that security is fake, because it’s built on other people’s lack of freedom.

What is the nature of emotional transference? Put plainly, it’s projecting the unease and fear inside onto the outside, trying to make yourself feel better by changing other people.

This pattern is far too common in human nature; all of us have done something like it more or less — losing your temper at family when you’re anxious, or gripping tighter when you fear losing something. (It just depends on whether we become aware of it; this is a bit like metacognition, the eye of God.)

Human nature is essentially chasing three things: security, belonging, self-actualization.

But ironically, security often becomes the limit; this is precisely what blocks his path to self-actualization. Real growth needs freedom and trust, but he puts shackles on himself and on other people alike.

Human nature is this complex and this fragile. We’re all afraid of getting hurt, but the more afraid we are the tighter we grip, and the tighter we grip the more we hurt, until finally we push ourselves and everyone else into a corner. The real balance is learning to accept your own imperfections, and daring to believe the world isn’t that dangerous, even if you trip now and then.

Not the beauty of appearance, but the feeling

Comparing the refined cafés in many countries and cities, artistsan cafe didn’t feel that exquisite to me, not that luxurious, not more forest-y, and not that photogenic. I even found it hard to feel how to shoot the look I wanted. But that’s exactly it: not that good-looking, yet unusually full of feeling.

The moment I walked in, I felt invited into the slow rhythm of life of the Renaissance.

I thought of a spiritual core shared by Japanese wabi-sabi and the aesthetics of old European film: not pleasing, but resonating.

An aesthetic isn’t the aesthetic the lens produces, but the aesthetic the “head” behind the lens discovers. Not seeing, but feeling.

Some spaces have a story even when they say nothing. That’s art; art comes from life and rises above it.

The traces of time, ichigo ichie, mono no aware — old wood, cracked ceramics, peeling paint, an imperfect, asymmetric yet simple beauty.

I’ve been to some other very distinctive art cafés before: abandoned factories, Western-style houses, manors… it seems the breath of culture isn’t piled up, but “grows out”.

The emotion of this moment is real and deep.

Culture and art really don’t need extra embellishment; they are the flesh and blood of daily life.

Observing my own process of researching travel guides

There are three types of people.

The go-with-the-flow type: this kind of person goes wherever the mood takes them, completely unbound by plans. They want to travel, they pack a backpack and set off, and may even buy the plane ticket at the last minute.

The semi-planner type: this kind of person does a bit of homework before setting off, like looking up the destination’s culture, geography and food online.

The plan-controller type: this kind of person makes an extremely detailed checklist, from what time to get up in the morning to where to sleep at night, everything calculated clearly. They pursue a sense of control; for them travel is like executing a task.

I seem to be the second type, though I actually bounce between 2 and 3.

I don’t demand that I follow the plan checklist strictly, but I will get a deep understanding of the destination in advance, learning as much information as possible — transport, culture, local life and so on. I’m the sort who both understands planning and knows how to let go, having found a balance between preparation and spontaneity.

Slightly drunk, I’ve become an INFP

I drank some alcohol last night. At the time I was thinking that a tipsy state lets me concentrate better on flow, creation, work; then I got sleepy, and fell asleep at half past eleven.

I woke at five in the morning and felt very excited.

It feels like last night and today both my emotions and senses are amplified — my perception of the environment around me, of relationships, and of nature.

The sun passes through the trees and over the window onto the wall, with some shadows of the trees.

I can’t sleep, but it feels good to be up early, so I get up. I’m hungry so I order breakfast delivery. The soy milk in the morning is super tasty; I wonder whether my body especially needs soy milk, or whether in a tipsy state the taste nerves are more sensitive? But these buns are much worse than before…

I start working in the morning and feel something is off. I can’t say what’s off, it’s just off. Dizzy, and hard to concentrate or want to work. I think my body wants to go back and lie down.

I go back to lie down, thinking I’ll sleep for an hour. I set an alarm. It feels like not sleeping isn’t good; I remind myself not to overthink, to fall asleep. But I just can’t sleep, and again I can’t say what’s off.

I try putting on Mayday’s music, but I’m afraid of being woken by the music if I fall asleep. Oh, then I understood: what’s off is that my own subconscious is gaslighting me.

Treasure your time; not sleeping is wasting time. And then senses and emotions are amplified again.

Metacognition observes the self, and I become aware of it again.

I cancel the alarm timer. If I overthink, let it overthink. I feel a bit like crying; then just cry.

I think of how flomo’s design philosophy mentions objectively recording your emotions, not criticizing or evaluating, just describing them like that — because you just want to describe, which is doing what you want to do.

It suddenly occurs to me that a lot of the time when my emotions are off, the cognitive reason comes down to not knowing myself well enough.

Why is it hard to keep loving high-pressure work? The root isn’t actually the high pressure, it’s that I need a healthy emotional state to face work.

I really want to cry. What exactly was off before? Was it confusion about my own emotions? It seems I still don’t understand myself well enough. Then what meaning does understanding myself have? It seems I’m still controlled by my emotions.

I’m so moved. I’m observing myself, saying what I want to say, doing what I want to do, loving whom I want to love. I feel like I understand myself better again. I get it: what I love is my own growth, the process of understanding myself better, not meaning itself.

Then is this high-pressure experience also a kind of experience, a way of observing and understanding myself? It opens up again.

I wanted to lie in the park and soak up the sun, but it turned out to be all dew.

That can’t stop me; I’ll go home and get a sleeping mat and come back to lie down.

No compromise, no internal churn, hahaha.

At the very end: the auntie in front was practising baduanjin, there was also an auntie practising basketball in the court on the left, and I was lying flat in the park.

Addendum, June 1, 2025:

While trekking I suddenly thought of the group of people with depression — emotions out of control — and then connected it to the state I’d been in before of being carried along by my emotions, and I had objectively recorded the whole feeling. It isn’t a manifestation of depression, but it’s still a good opportunity to feel the amplification of depressive sensation and the pull of emotion.

I can feel that my emotional regulation ability and my self-observation ability are very strong. From confusion to final self-acceptance, I’m good at understanding and observing myself.

Aotian is back

I feel I lack experience being around small animals.

Do you only cherish something once you’ve lost it?

Or maybe I just couldn’t bear it, so I adopted it back.

Seeing the cat still made me excited.

I was never good enough to Aotian before; I should respect a cat’s life more. I need to be responsible for myself, and responsible for my relationship with the cat.

Observe the cat: what does it like, rather than observing what I like and imposing that on it. That’s actually a manifestation of the desire to control.

Same as with intimate relationships: get to know the cat more, get to know my relationship with the cat, and don’t hurt it (regret).

Independent development in the AI era: the front end is the main battlefield

AI models have turned core services into commodity services. The backend gets simpler, with no technical moat.

SaaS and PaaS providers wrap up foundational capabilities; using cloud services directly solves a lot of problems, without needing to implement them.

The front end becomes the key place where “user value is delivered”, and it’s easier to focus on the MVP and a single-person product loop — with Next.js/Tailwind/Firebase/Vercel you can complete a full-stack delivery. You can quickly validate the chain of “feature → value → growth”.

Databases, permissions, storage and model calls are all heading towards platformization (Supabase, LangChain, Replicate, Vercel AI SDK); independent developers don’t need to write the backend from 0, they only need to “integrate”, not “build”.

AI has decoupled the “intelligence” part from the backend; the front end is the battlefield closest to the user, changing the most, with value most concentrated.

Simplified design in the AI era

A lot of design isn’t necessarily needed. For a simple team, preparing two designs up front is enough.

Product design: valuable to users, good features, MVP features.

System design: system modules, architecture, interfaces, deployment.

These two parts need to be designed in interaction with AI and in combination with your own business team’s situation.

Then combine the resulting documents and output a task list note, and manage it with project management tools like github projects.

Generally you output a product feature list (Feature) and a technical implementation task list (Tech Task), corresponding to GitHub’s issues and sub-issues, and at that point a project is simply under management.

An extreme user experience matters; how to save WeChat Official Account articles

An extreme user experience matters a lot. How to quickly save a WeChat Official Account article is also, I think, a very valuable area, including a whole series of divergence and aggregation around articles, and the ability to find similar Official Account articles.

Product design

Product design techniques. At this stage it’s still a user-centred design perspective.

Background

users: our users’ profiles/personas

goals: our users’ main goals

pains: the obstacles blocking users from reaching their goals

Value proposition

product: what needs to be built to help our users achieve their goals?

alleviates: how does the product relieve the pain?

advantages: how will the product create gains?

Goals

Break the product down into a series of goals.

Solution

Core Features: the main features to be developed.

Integration: how the product integrates with other services.

Alternatives: alternative solutions that should be considered.

Constraints: limiting factors to be mindful of.

Out-of-scope: features not being developed for now.

Feasibility analysis: the project’s feasibility.

Thoughts on Elmo Chat’s marketing

Elmo Chat is like deepwiki — neither needs a login to use.

Which lowers the barrier to use. Also, cost management matters a lot; a large volume of simple-model examples is very good for driving traffic.

Even for promoting an open-source business model.

The most fundamental reason Elmo Chat can be free is: they don’t use expensive large models, they use low-cost open-source models + a local or lightweight inference architecture, plus a growth-oriented business model.

And one more thing that matters a lot: lightweight and speed. Simplicity especially matters; the beauty of simplicity is more like an aesthetic culture of a brand.

Thinking through RAG and long context

It’s undeniable that long context is still a big challenge now and over the long term. Today long-text processing speed mainly comes down to the computational complexity of the attention mechanism. And there’s another problem: even if long context is big enough, can the model truly remember and use all the information?

Accuracy — even if long context is big enough, truly remembering and using all the information, and finding the key information, is extremely difficult. For complex logical reasoning or information task integration across a very long span, the model’s accuracy still needs a big improvement.

RAG has two parts: retrieval speed and generation speed.

And one more thing is RAG’s traceability. In the short to medium term, RAG is probably the trend.

Over the long term, the boundary between the two may blur.

The goal, in short, is to be faster and better.

FOLO, some methods for an efficient information entry point

Collecting information has always been something that gives me a headache. I really dislike phones because a phone’s recommendation algorithm gives you information that has already been filtered. Folo is a way of actively subscribing.

There are only three cases of getting information:

recommended information

subscribed information (which also depends on recommendation at first)

deeply relevant information (depending on recommendation or active search)

The way people collect information varies from person to person and comes in many forms; redefine and organize your information sources according to your own needs.

One factor to consider is supporting subscribed information, and alongside that, how to integrate some of the deep tools you use daily, folo for instance.

Scott Scheper’s approach to reference cards

The aim is to make the reading process systematically merge with your personal knowledge network. He thinks a reference card isn’t a simple bibliographic recording tool, but the first link in a knowledge “access system” — the “gateway card” that organically brings outside thought into your internal thinking universe.

Three elements:

Information: author name, book title, subtitle (if any), year of publication, publisher, page numbers and so on.

The goal of the reading source.

An overview of the source.

The approach to reference cards for the internet age.

The characteristics still worth keeping, plus a few optional fields to add.

Processing and collecting information

Processing information quickly matters especially; for people in the information trade the speed of processing information is especially critical.

For me, I open a large number of tabs every day, including all kinds of web pages, articles, tutorials and so on.

For an efficient information-processing tool:

The density of the information really matters a lot too, the right amount of abstraction + an engaging way of parsing it.

Subscribing to the collection of information on a topic over a period of time

Core requirement: for instance, I’m researching the development of a direction (LLM Long Memory), or subscribing to a direction — the related papers, literature, blogs and so on.

Another possible need is that I’m about to travel and want to know all the information about a place I’m going to, and whether it can be recommended to me in real time. The weather, the latest posts and so on.

It can be provided as an agent capability, even as an open-source agent service.

Subscription trigger (schedule/event) → Query builder → Multi-source fetcher → Content parser → Importance filter →

Structured summary (optional clustering) → Rendering and notification (or saved to Notebook) → Learning from user feedback

The future of LLMs

claude now focuses on running a task for as long as 6 hours; jules can independently complete an engineering task.

Within a year it’s expected to reach the user level: AI independently executing a task for an hour.

That is to say, within three years users can basically shield themselves from code details at the business layer.

Programmers will collectively turn into engineers afterwards.

How to schedule which LLMs and agents perform what operations, how to set engineering rules, and how to accept the result.

Before: human (intent, code) ⇒ execution. Now: human (intent) ⇒ AI (code) ⇒ execution. Humans “package up” the underlying complexity and “push complexity down” one layer, so they can think at a higher level. Mastering “How” used to be the core competitiveness; now mastering “What” (knowing what you want, judging whether the result is good) has become the key ability. Competitiveness has undergone a turn.

The era of short text

Some people say the new generation is poisoned by short videos.

They’re obsessed with fragmentarily consumed video and no longer willing to read, so they lack the ability to structure things.

But a person’s structuring ability is ultimately limited. In fact, writing papers and writing books is also a process of constant reshaping, repeatedly drafting how to structure things.

What AI can empower most is, in fact, the people who are least good at expressing themselves.

In the AI era the most important thing is no longer writing ability, but willingness to express and life experience.

In the AI era you don’t need to read long texts. Facing a ten-thousand-word article, in the past you might have chosen “read later” out of fear of its length, and the result was usually “never read”. Now, AI can help you extract the core points, generate a summary, list the key arguments, and even turn complex concepts into plain explanations. You can grasp the gist of an article within minutes, and if you’re still interested, invest more time in studying the original.

The speed of collecting information has exploded.

Literacy with words is still the “privilege” of a minority. We see countless vivid stories and profound insights lost in obscurity because of a lack of expressive ability.

A good article is absolutely not because the writing method is outstanding, but because of the soul (life experience).

Some fragmented words, as experience accumulates, keep resurfacing, occasionally welling up like ideas.

The question of data sovereignty

Data sovereignty really is a fairly big problem.

I looked through some of the issues around NotebookLM’s data storage.

One is user-level privacy; preserving the privacy of content as much as possible matters especially.

notebooklm is essentially document-upload-first, rather than endless searching.

Depending on your geographic location and Google’s infrastructure, your data may be stored in data centres in a specific region. Data localization and sovereignty issues are becoming more and more important, which is also why it can’t be accessed domestically or in certain countries.

Thoughts on using the NotebookLM product

I was preparing to go to Jeju Island in Korea, but had no time for deep research on Jeju Island. I thought of NotebookLM. Let me describe my method.

Since it’s travel, I care more about how current the information is. I downloaded Jeju Island’s official documents, blogs and books locally.

I used grok to do a deep search on Twitter for some of the last month’s information about Jeju Island, and events coming up.

With gemini’s deep research I gave it my own information and plan, and my motivations, and deep research put together a complete planning document for me.

I quickly found some valuable posts on xiaohongshu.

On NotebookLM I used the explore feature to search for information sources related to trekking on Jeju Island.

Because trekking on Jeju Island isn’t very popular domestically, on Perplexity I searched in English for detailed guides related to Jeju Island trekking, searching with English keywords and having it reply in Chinese.

Everything above was gathered into NotebookLM, in a new workspace.

Quickly generate an audio overview so I can quickly grasp the core key points, with a prompt as the guide and a more conversational tone.

Then quickly generate a mind map; the mind map serves as the final structured output.

Based on the mind map, once I’ve gone through the basic chain, I click on nodes I don’t understand, expand them, and let notebooklm explain by calling the rag knowledge base, with citations. If I’m interested I get a deeper explanation, or click the citation to the original text.

Then comes the important travel stuff, which I can ask Notebooklm directly. For instance I asked it about phone cards, or about getting the Olle Trail passport; basically it gives an answer and finds the original text.

When I need to read the original, NotebookLM generates a source guide (summary) for every original text, plus the key themes.

These key themes aren’t just for categorization — I often Discuss theme to quickly understand a category of theme or terminology, and if I’m interested I can have it explore the corresponding theme further, pulling together important information and literature from wiki and other sites.

I can’t imagine it: in the past, organizing and absorbing literature might have taken a day; now the full information collection and structured organization takes less than an hour.

Thoughts on NotebookLM’s product design

What attracted me most is the mind map. A big problem with RAG seems to be that the main reason many people think RAG is useless is that the information density RAG extracts isn’t high enough. If the collection step is done by yourself — quickly understanding an article from its summary, quickly choosing what to save based on the summary, for instance.

Both the saved content and the original text get sent to the workspace as context. A workspace is essentially a working space too.

At this point, doing RAG produces a lot of inspiration that interests information workers.

What really surprised me most about NotebookLM is the audio and the mind map. Audio is actually a very good tool to help you get up to speed quickly. What matters is the mind map: it makes the various characteristics very clear, and after you click, rag explains it to you in detail.

Today’s AI state can’t reach a good flow effect

It always feels like something is missing. I thought and thought about what was missing. Ah — I realized the situation with me hasn’t changed either; it’s just that before I was turning screws next to them, and now I’m supervising AI turning screws next to it, busy like a robot either way.

Where’s the flow state in that? It’s like being a CPU constantly switching processes — now directing this task, now that one. Each task’s cycle is short too, it ends in a moment, and then I have to go track and debug it.

I thought about flow states a bit. What I personally like best is sustaining a task for more than 20 minutes; if it doesn’t exceed that, it just looks busy.

But today’s AI obviously can’t do that. A sustained task lasts five minutes at most; cut off once every five minutes, and that feeling really is uncomfortable.

A few insights:

If there were an AI scheduler that interacts with a human to schedule the various AI agents, that might be a form that appears in the future. Note it’s not a general-purpose agent; a truly general-purpose agent still has a long way to go — most people wouldn’t even need to work.

The task cycle of LLMs, the duration of sustained tasks, will definitely get longer. As it keeps getting longer, humans explore the boundary: what is the limit of what AI can do.

As AI’s sustained duration keeps increasing, two things change: one is that I spend more time thinking and designing (how to set tasks for AI agents, thinking about AI’s boundaries, my own projects, system design), and the other is that I spend more time reading and learning (what AI outputs, what AI searches).

You no longer need to focus on code, but you still need engineering. In the future there will be no code farmers, only engineers and designers; the intermediate state of product manager will disappear too. An AI product manager is just a virtual concept; in form it still moves closer to engineering and design.

Fragmented memory matters too. When you occasionally switch AI agents you need to know your goal and the context — though that can be supplemented by the interactive agent platform from the first insight.

Some practical engineering suggestions

These days I actually pay very little attention to code details; what I basically pay attention to is directory structure, design patterns, and the detail parts of the code — at most a glance.

It’s really just a human guiding AI. I’m maintaining two repositories, one for cursor and one for windsurf. The windsurf scenario is mostly used after jules finishes a PR, to supplement and correct it.

So every evening I organize some tasks, the important tasks, list them out, put together the design and the acceptance criteria, and then hand them to jules to implement. The next day I keep making changes based on jules’s PR requests.

In cursor it’s more about the main tasks; some tasks jules still handles too slowly, or the context isn’t sufficient. In cursor practice I’ve actually accumulated a lot of engineering techniques.

More and more time goes into reviewing and proposing design requirements.

Even the review and the design requirements are obtained by interacting with an LLM bot or LLM chat.

You can pause, think a bit more

But try not to use filler words, like “then, and then”.

It feels as annoying as saying “whatever”.

Focus on the value of the content itself, not the rhetoric.

The neural-circuit trap of hedonic pleasure

Short videos → rapid dopamine release → threshold rises → need stronger stimulation → pleasure poverty

The cognitive construction process of eudaimonia:

Effort → Understanding → Connection → Internalization → Growth → Deep satisfaction

Eudaimonia has one precondition: without resistance, there’s no real growth.

In the fast-consumption era, as we consume information, on the surface information becomes easier to digest, more interesting, more accessible.

At a deeper level, we’re losing the ability to “chew” cognitively. Reading a book, we increasingly want to find some shortcut method.

Isn’t that essentially cognitive degeneration?

The long-text era: deep cultivation → sowing → waiting → harvest (farming mode)

The short-video era: discover → gather → consume → forget (foraging mode)

It’s like — if even thinking has to rely on AI, do “we” still have any reason to exist?

We should strengthen the user’s cognitive sovereignty, not replace it.

Create an “anti-entropy” reading experience — one where the reader’s cognitive entropy goes down rather than up.

We’ve talked before about the difference between intrinsic and extrinsic motivation; cognitive gamification is similar.

Gamification: external rewards (points, badges)

Cognitive gamification: internal rewards (the thrill of understanding, the surprise of connection)

Most people scrolling short videos belong to the first kind.

We should think and reflect on how these good human qualities can be made more readily evident.

How do we make deep thinking as addictive as scrolling short videos?

How do we make knowledge accumulation as visible as leveling up in a game?

How do we make cognitive growth as responsive as social interaction?

However good the tech, if it can’t do it in a convenient, effective way

However good the tech, if it can’t solve the user’s (or customer’s) real problem in a convenient, effective way, it will struggle to succeed. Developing a product should start from user needs and the overall experience, rather than having the tech first and then fishing around for an application scenario.

The wrong approach: develop a cool piece of technology first (an AI API), then expect other people to rework their existing products to fit your technology.

The right approach:

First think: “What should this product (or rather, the user value you want to provide) look like end to end?” You have to conceive the complete user experience and product shape first.

Then consider: “At which points should we add AI technology to solve a specific problem or improve the experience?” AI should be used to assist and optimize the overall product, not to be the product itself.

LingoWhale deep analysis

The basic model service is dedicated to handling summarization.

Text preprocessing and chunking cleans and formats the raw text; once it exceeds the length limit you need an intelligent chunking strategy. Continuity also has to be considered, and cross-chunk mapping has to be supported.

The basic model service generates the text summary, summarizing according to the text’s type and length.

The mapping mechanism between summary and original text feels like transformer’s attention-weight mechanism.

Summaries generally come in two kinds: abstractive summarization and extractive summarization.

Extractive summarization mostly uses sentences from the original text, but abstractive summarization can’t be mapped that way — it needs support for pretrained models and access to attention weights, to achieve mapping between summary and original text. It allows the model, when generating each summary word, to attend to the relevant part of the original text. In sequence-to-sequence models, for instance, attention weights can show which part of the generated summary is most relevant to which part of the original.

Hugging Face Transformers provides pretrained models (BERTSUM, T5, BART) that use the Transformer architecture with built-in attention mechanisms. Users can access these weights by setting output_attentions=True, achieving mapping between summary and original text.

Thoughts on WeChat MCP

Getting book information and details.

The main thing is being able to get the notes from books — in that case MCP should be able to handle book notes very easily.

Quickly generating reading notes — this feels important to me. In that process a person only needs to do the filtering of information.

Integrating it into a knowledge framework and knowledge map — that’s AI’s capability.

Intelligently hiding

Simplicity of core functionality needs thinking: how to make features simpler rather than more complex. The greatest way is the simplest — one simple chat in chatgpt completes more than 80% of tasks.

You don’t need to go learn anything extra; the cognitive burden is too heavy.

Interview with the founder of Lovart, who has no product manager

The generalist product manager of the future isn’t much use. I myself have over ten years of product experience and lived through the complete mobile-internet cycle. I think the core value of a future product manager lies more in depth of industry knowledge, in teaching a specific industry’s know-how to AI in a systematic way. So the product manager role will be more divided by industry, by vertical. I think the generalist internet product manager is a very dangerous profession.

General-purpose Agents are very likely to be internalized by foundation models in the future. I think what will really flourish in the future is the Agents focused on vertical applications. The core competitiveness of these vertical Agents is the vertical domain expertise behind them. For the generalist internet product manager of the previous era, I do think the positioning has become a bit blurry.

If general capabilities are absorbed by the underlying model layer, then all that’s left is professional design capability.

When tools are smart enough, “the person who manages requirements” is no longer needed, but “the person who defines requirements” becomes even more important.

AI really has massively liberated productivity, and may also make top designers’ ways of thinking more egalitarian.

cursor has more of a data advantage than GitHub. GitHub’s data is largely internalized data, but cursor records how users use AI tools and how they combine AI with traditional methods for creation, which is very valuable. ComfyUI’s workflows are essentially stacking various models and traditional function modules together. These things need AI to learn them.

Group pushes

The issue with LLM pushes is that, like recommendation algorithms, they may also be questions other people have asked or saved.

The questions are valuable, so for saved questions, they’re necessarily questions a recommendation algorithm especially deserves to recommend.

Thoughts on Claude’s projects design

As I remember it, a product’s projects are usually there to categorize content or separate modules.

After buying the subscription to Claude, I found projects actually have some use. ChatGPT has memory; I thought before that ChatGPT would also support long-term memory — and mem0 also supports categorizing stored memories or adding some tags, separating work memory from life memory. But in real practice at this stage there are still situations where we’re used to doing tasks in the same window, or not deliberately managing multiple windows. That brings problems: one is muddled memory — and for a memory mechanism, memory is also inaccurate. Before, all conversations were mixed together: today chatting about work, tomorrow about study; chatgpt easily gets its wires crossed and confuses the information from different topics.

Imagine Projects like different folders on your desktop. Each Project is a dedicated conversation space, where you can:

upload related documents and files

set specific working rules and background information

have continuous conversations; Claude will remember all the context inside this Project

For example you can create a “paper writing” Project, upload your research material, and tell Claude your writing style preferences; then create a “code project” Project, upload code files, set the coding conventions and so on.

It feels like a magic tool, similar to Claude’s mdc. Claude is actually a set of rules after reinforcement learning.

It’s just that everyone is looking for ways to improve the corresponding input quality. projects essentially affects the input and lowers the burden of explanation, so in business you can think along similar lines: distinguish the corresponding scenarios — at the current long memory stage that helps a lot.

Better output quality isn’t just about making the model “smart”, it’s about making it “understand what you want to do”.

The best AI products aren’t planned — they grow up on their own from the ground layer

Anthropic engineer David Soria Parra, while using Claude Desktop and an IDE, found that the two had no effective integration, which made the workflow cumbersome. He realized this was a classic “M×N” problem — the complex relationship between many applications and many kinds of integration. That’s how MCP came to be.

Because the need existed, MCP evolved out of it. MCP isn’t icing on the cake; it was born because the moment demanded it.

If an AI product a year from now still only amounts to “I can answer your questions, and occasionally I even offer suggestions proactively,” it will lack differentiation and appeal.

Thinking every day about what users need really is meaningless. A visible need is obvious anyway, and an invisible need may just be something you invented out of nothing. “I hope there’s a product like this, I hope they use a product like this, I’ll educate users with my own cognition” — that really is a foolish way to behave.

Applying real technology to solving things that genuinely have value — that’s the kind of company with more creativity and more staying power.

In the short-video era, what’s the essential change in pleasure?

Building an information-consumption product — the core question hinges on how you understand what information consumption essentially is.

There are two essential types of pleasure:

Hedonic pleasure: immediate, sensory, shallow — like the easy laughs, the good-looking people or the special effects while scrolling short videos.

Eudaimonia: a deeper satisfaction tied to meaning and value, like being absorbed in reading, completing a challenge, building a deep relationship with someone.

Tiny pleasure, brief pleasure, pleasure that vanishes in an instant — it feels like a kind of illusion of pleasure, hard to turn into deep satisfaction. But can that kind of thing last?

I used to think that even in the AI era people don’t need the ability to handle structured, full-length content — that a bit of fragmented information is enough to cover the input part.

Paul Graham’s “writes and write-nots”

“Writes and write-nots” is a pun Paul Graham used in an essay titled “Writes and Write-Nots,” published in October 2024, borrowing the ring of “the haves and have-nots.”

“In the long run there are only two kinds of users — those who write and those who don’t; being able to write is a scarce ability, and those who can’t will outsource their writing to tools or to other people.”

Some thoughts after reading it.

You can clearly feel a distinction between active and passive.

Active is: you want it yourself, and AI helps you get it.

Passive is: the drive is to get the goal finished, so you let AI complete the task as much as possible. For example, you don’t even know whether the code AI wrote is right or wrong. Whereas the former has a design or a line of thinking; AI just helps you write it faster.

Essentially this is a question of consciousness and cognitive autonomy.

Judging whether a person is a “write” comes down to whether they have an original perspective, a coherent cognitive structure, and a responsible attitude toward their language output. I feel this is also a very good angle for discussing whether AI will replace people.

Thoughts on the evolution of reform and opening up

Reform and opening up brought an economic miracle, people’s thinking changed drastically, and thought was no longer simply consistent.

Reform and opening up is an economic miracle, but there is a fundamental tension between economic liberalization and political control.

Political reform was limited, and basically stalled after 1989; stability was emphasized as the precondition for economic progress.

From the ideological fervour and repressive social-economic environment of the Cultural Revolution, to the emphasis on materialism and rapid social change in the reform era, to the changes in technology — China’s change has become the most intense change in the world.

War

Politics is ruthless. Those who forget a history of suffering will surely be victims of suffering.

Anyone who hasn’t wept in the deep of night isn’t qualified to talk about life!

Love is seeing your own giving in another person’s need.

ref: Chai Jing in conversation with survivors of the Chinese civil war: “Those who forget the history of suffering will surely become the victims of the next suffering” https://www.youtube.com/watch?v=Db17u8E1t0g

Thoughts on the development of Japanese aesthetics

The Japanese feel spirits in all things in nature, to the point that Japanese people have a natural sense of closeness to the natural world.

Japanese aesthetics were early on built on a sensitivity to natural beauty.

Shinto encourages viewing excellence of craft and refinement of art as an offering to the spirits, which cultivated a pursuit of craftsmanship and refined aesthetics.

Zen emphasizes simplicity, emptiness, meditation, mindfulness and direct experience of reality; these ideas penetrated deeply into Japan’s artistic practice and aesthetic values. One of Buddhism’s core teachings, “impermanence” (mujō) — that all things change and nothing is constant — and the acceptance of transience in ideas like wabi-sabi and mono no aware.

Understanding beauty, the essence

On the universal side: people are born with a fondness for melodic music, and a liking for symmetry.

There are things human perceptual mechanisms have in common; the golden ratio in architecture has been praised in both ancient Greek and East Asian cultures. This universality may come from humanity’s shared cognitive and emotional response mechanisms.

On the other hand, the definition of beauty and the way it’s appreciated are deeply influenced by culture, history and personal experience.

The Western aesthetic tradition emphasizes form, structure and symmetry, while Eastern aesthetics (Chinese painting theory, for instance) pays more attention to mood, ethereal space and the beauty of nature.

What is the essence of beauty? Is beauty subjective or objective? You can only say there’s a set of universal standards with differentiated ways of presenting them.

Is beauty only sensory, without containing logic? The resonance the implied meaning produces (cultural resonance) is also a kind of beauty (beauty isn’t only the heart, it may also be the brain behind it).

A sudden insight into beauty

I’ve been thinking about the various domains of aesthetics: the aesthetics of architecture, of art, of nature, of appearance.

Could there be universal “shortcuts” to a breakthrough?

Some commonalities, as before: perceptual mechanisms have things in common. Symmetry and proportion, colour matching, for instance.

So there are some basic principles: composition (the rule of thirds, say), colour theory, contrast and balance — all useful in photography, art and UI design.

Ask a few more questions. When a design moves you, think about why it moved you.

A large volume of input, and your own creation — it still has to be gradual.

The beauty of minimalism, in products

I hate colour now; colour just adds complexity. Even images shouldn’t be too large.

With black and white as the base palette, starting to use cursor rules to define the rules properly really matters.

Why are young Koreans so caught up in the rat race, while young Japanese just “lie flat”?

Why do the Japanese lie flat? Since the “bubble economy” burst in the 1990s, Japan entered the “lost thirty years”: low economic growth, prices and wages stagnant for a long time, young people without hope for the future. The workplace is fiercely competitive and promotion is hard. Some young Japanese choose a low-desire lifestyle — “no marriage, no house, no consumption” — that is, “lying flat.”

But young Koreans more often show a passive kind of striving, or rat-race striving. “You can’t lose” and “make something of yourself” are the social norm.

This society places extreme weight on credentials, looks and job title. Celebrities and influencers are often treated as repeatable templates for success. Can’t lie down, can’t keep rolling, but still fighting.

“N-time exam takers” (repeatedly sitting civil service or graduate school exams)

The “Sampo generation” (giving up dating, marriage and children)

“Rat-race burnout culture” plus self-mocking subculture (for example the term “Hell Joseon”)

A fear of failure, the inertia of a culture, a struggle that hasn’t entirely given up hope.

Young people in India

I met quite a few Indians on Jeju Island too. The two countries seem so far apart, and this is the world’s largest group of young people. Anxious about the future, yet eager to rule the world?

“The world’s largest youth population” — more than half of India’s population is under 25. This is the largest group of young people of any country in the world.

Their hopes, plans, worries, dreams.

India’s 2025 youth numbers.

Total population: 1.464 billion.

Youth numbers: in 2025, India’s under-25 population is roughly 600–750 million — the largest youth group of a single country in human history.

Do you believe good food is universal, or that everyone’s taste differs?

It’s hard to please all tastes. I ate a meal in Jeju Island recommended by xhs: Korean grilled fish, Korean fried chicken BHC, and seafood shin ramyun. I found them all pretty bad. Out of ten, the first one gets a 5, the second and third only a 3. They were all pretty expensive. These were all recommended by Xiaohongshu, and the ratings on Google Maps were quite high too.

I kept wondering whether my appetite was off lately, or whether my taste differs from Koreans’. But in other countries I’ve never had this situation — local popular restaurants are still something I really like.

Until tonight I ate at Vibe 230. The owner is Korean and doesn’t speak much Chinese, but he knows basic numbers — one thousand, two thousand and so on.

The decor is quite distinctive, using brick walls, antique ornaments and lots of CDs, giving a unique and relaxed feeling.

It serves pasta (especially fern pasta, shrimp pasta, basil pasta and so on), pork cutlet, gambas al ajillo and baked cheese. Most reviews think the food tastes good, and some even call it a “pasta food shop”.

The owner puts real care into it.

Good food is universal. Truly delicious things, most people understand after one bite; some flavours are just the flavour of humanity. Like music and art, it’s inherited in human genes — we just love this taste and this melody.

When you plan a trip, what do you pay more attention to

A. imagining the interesting experiences that might happen on the trip, exploring new ideas or cultures (looking forward to the local art, history or unique experiences, for instance).

B. focusing on the concrete itinerary and practical details (booking hotels, arranging transport, packing essentials, for instance).

Choose A, intuitive type; choose B, sensing type.

I still subjectively think it isn’t too late to do self-media now

I still subjectively think it isn’t too late to do self-media now, it’s just very hard to find a track that matches me well. Looking at my own experience and hobbies, on one hand there’s the profession, on the other hand the outdoors — which is mostly trekking and climbing mountains, then living in various cities and countries.

But on the professional side I’ve always struggled to find a very well-matched approach. I wonder whether it’s because the direction is wrong. Today I saw a video; actually I see a lot of information sources online, and then share and discuss within those information sources. There wasn’t a particularly good way before, but now there seems to be a faintly decent way, and I think this counts as a general solution too.

I’m thinking about how to use this approach to produce something out of what I normally see and think about. Because a lot of content now is invisible, but I think it’s good for a person to have some desire to share and some sense of creation. At least having a bit of a sense of creation is good.

So later I may feel I should still go do it. The general solution right now is to share what I have professionally — for instance, which good methods I see, record them, record them with some short-video-matched tools, and then think about how to sync them to the various social media.

Actually as for running social media accounts, right now it’s more that there’s no time, but that doesn’t mean the AI era hasn’t given creators excellent empowerment. I think I should slowly accumulate my own first-mover advantage; this is growth for me too.

Sunset in Jeju

This reminds me of flying from Bangkok to Nepal.

The first time I saw the spectacular Himalayas, I was sitting next to the aisle.

I was delighted back then — it was the first time I’d met Nepalis. They seemed as lively as Indians, and on the plane it felt like they all knew each other, wandering around swapping seats.

The guy sitting by the window also liked taking out his phone to shoot video. There seemed to be people he knew elsewhere, so I often had to let him through. I found it a bit comical; my first impression wasn’t great, it felt very noisy to me, but I’m an introvert.

Seeing the Himalayan range and the sea of clouds, I was thrilled. I hadn’t expected Everest to tower like that above the sea of clouds, looking down on the other ranges.

I kept pulling out my phone and zooming in five or ten times, trying to recreate the magnificence of the Himalayas I was seeing.

The Nepali guy by the window seemed to understand what I needed, so he gave me the window seat. That kindness stayed in my memory, and this Nepali stranger I met by chance is still deeply etched in my mind, still affecting me.

🇰🇷jeju day2 solo olle 1

🇰🇷jeju day2 solo olle 1 16.9km✅

While I was waiting for the bus, an elderly local grandma started explaining something in Korean — the gist was probably about the bus. When she realized I didn’t understand, she pulled out a bottle of beer and tried to give it to me; then she just twisted the cap off and downed it in one go. So free and easy.

And my first instinct, subconsciously, was that drinking this is bad for your health. That instinct contrasted sharply with the grandma’s free-and-easy attitude, revealing a collision of two life philosophies: one is indulgence and freedom in the present, the other is rational restraint and long-term calculation.

I really liked that grandma, because I’ve always loved and practiced living in the present. It’s an affirmation of pure enjoyment of the now, unbound by age, social norms or health considerations.

What I love about the purpose of travel is that we’re shaped by our own culture and experiences, but encounters with others give us the chance to appreciate different ways of living, even when those ways challenge our instincts.

Every now and then, put down the excess rational analysis and embrace the purity of the present. Sensibility and perception are the root of happiness, not rationality.

You have to become yourself. What you admire are the people on the road to becoming themselves, embracing the present while looking toward the future.

The history of the Olle Trail

Tonight at dinner I had fried chicken, and I kept researching the Olle Trail on my laptop while eating it.

The owner got curious too and asked what I was working on, saying he wanted to look into it.

I said I was researching the Olle Trail. The owner said the trail was built with crowdfunding from the islanders. I didn’t quite believe it — the islanders’ kindness and simplicity are certainly real, but it also rests on the original founder.

The creation of the Jeju Olle Trail has a very moving story; it began with the dream of a rather special journalist.

The founder is Suh Myung Sook, born in 1957 on Jeju Island, Korea, a graduate of Korea University’s department of education. After graduating she became a reporter at the well-known Seoul weekly Sisa Journal, entering the media world, and in 2001 she became the weekly’s editor-in-chief. Tourism Research and Information, 2013, Issue 2. Later she also became editor-in-chief of OhmyNews, the world’s first citizen-journalism website.

One day in September 2006 she made up her mind, ended a 23-year media career, put down her work, wanting to relax completely and give herself a long vacation. After a month of preparation she flew to Spain alone and chose the thousand-year-old Camino de Santiago pilgrimage route for a thorough, quiet, restorative journey.

Suh Myung Sook spent 36 days on this pilgrimage route. Walking the ancient path, the scenery, local flavor and historical sites along the way, and the people she met on the road, all left a very deep impression on her, and she was greatly encouraged and uplifted. Tourism Research and Information, 2013, Issue 2. Toward the end of the trip she met an English girl, Henny. When they talked about how the trip felt, the girl said: “If this journey matters so much to us, why don’t we go home and build a ‘Camino de Santiago’ of our own?”

Inspired, Suh Myung Sook returned to Jeju Island, determined to create a similar walking path in her hometown. She named the path “Olle.” “Olle” is the Jeju dialect, meaning “the narrow path connecting the main street to the gate of a house.”

My understanding of the essence of growth

I once saw a formula — someone said the essence of growth is (happiness + sadness) / ordinariness.

At first glance I strongly agreed, because for a moment I felt a sense of belonging. I’m also the kind of person who doesn’t like ordinariness — choosing an uncertain dish means far more to me than choosing a repeat of something tasty. The longer you stay in an ordinary state, the harder it is to break the environment and break through yourself.

Then I thought about it: is ordinariness really unimportant?

Pain wakes you up fast; happiness sparks creativity and a sense of direction.

But looking across a whole life, stretching the timeline out infinitely, there aren’t that many ups and downs — it’s just that what we remember more deeply is change.

Real growth, very often, is precisely “the ordinary.”

This slowness is more like hiking: one step at a time, no good scenery, keep walking; tired, take a rest. Feet on the ground is the root.

Human growth needs emotional experience on one hand, but more than that it needs the settling of time.

Growth ≈ (awareness × action) × time

Happiness or sadness is the starting point of awareness.

Ordinariness is the field of time.

Without action, only feeling, it’s “experience”; with action, it becomes “growth.”

Does LingoWhale do its mapping through the attention mechanism

Setting rag aside, the modeling_lingowhale.py file of LingoWhale-8B defines the model’s attention layers, and the system can extract these attention weights while generating the summary.

But it seems this capability can actually be achieved with rag as well.

I’m wondering whether it’s possible that, at the application layer, business logic could do it just as well.

Art is real

He was mid-sentence when his head drooped onto his knees and he died. I’ve seen that way of dying in films; I always thought it was artistic exaggeration, but ever since Dong Jianyi died I’ve believed it: art is real.

A film’s plot also comes from life; art comes from life.

The absurd events in life are exactly the source of art.

Jiabianguo.

The essence of bargaining is resource scarcity; it depends on who can flip the table

The essence of a bargaining game is resource scarcity; the essence of a game between party A and party B depends on which side is more able to flip the table. If party A is more able to flip the table, party A has a lot of say. So actually there’s no point at all in trying to please other people.

What’s truly valuable is raising your own say, raising your own scarcity, making yourself the kind of person who can’t be replaced, copied or bypassed — and making what you do something that can’t be ignored or replaced.

So the most core thing is to make yourself stronger.

People shouldn’t casually use tools they aren’t familiar with that don’t help much

This time I stepped in a pit…

Blood and tears.

She believes that as long as she keeps worrying, she can avoid the misfortunes accidents bring

She believes that as long as she keeps worrying without pause, she can avoid the misfortunes that accidents of fate bring.

Shifting the scaling law curve to the left

Shifting the scaling law curve to the left, lower training cost.

Infra frontier: shifting the Scaling Law curve to the left, lower training cost. Meaning that at a lower parameter count, compute or cost, you achieve the same or better model performance. The future:

Inference cost is low, innovation-driven.

After efficiency improves, total demand goes up instead (compute requirements may be higher).

So for compute infrastructure companies, there will still be fierce competition.

The shape of programming in the future

Asynchronous interaction: by 2027 it can work independently for an hour; by 2030 it can work independently for a day.

For AI, it can increasingly keep doing things over a sustained period, more and more like a qualified engineer.

Engineering ability.

Buffett’s last battle

Opportunities don’t show up on a schedule; keeping ample cash reserves is about being able to move quickly when a major opportunity appears in the market. He put it vividly: “It’s a bit like death — the chance of dying the next day at age 10 is very low, but if you live to 100 you’re almost certain to die.”

Japan still has plenty of opportunities.

On his successor: Abel is prepared in every respect to serve as Berkshire’s CEO; he will recommend the board appoint Abel as CEO before the end of the year.

On tariffs: the US should seek trade with the rest of the world, and trade shouldn’t be used as a weapon.

On US fiscal policy: America’s fiscal problems scare me, but this isn’t a problem unique to America either.

On big swings in US stocks: compared with historic crashes, the declines of the past few dozen days are nothing.

On artificial intelligence: I won’t build all my investments and development around AI; if there’s a choice to be made with AI, it should be made by (Berkshire vice chairman) Ajit Jain.

On the Japanese market: I won’t sell any Japanese stocks, not now and not in the next 10 years.

On cash reserves: the reason I made a lot of money is that I never invested with everything.

On timing of moves: the chance of an opportunity appearing tomorrow is minimal, but the chance within five years isn’t low.

Advice for young people: you should be around people better than you; do work you enjoy, and don’t worry too much about the starting salary.

Praising Apple CEO Cook: it’s a little embarrassing to say, but Cook has made more money for Berkshire than I have.

We’ve all heard this: to build something valuable

We’ve all heard this: to build something valuable you have to be 10 times better than the existing product. It seems to be an accepted law of startups. This law is so powerful that even Google+, backed by billions of dollars, couldn’t become a competitor to Facebook. Google+ can be said to have become a better service than Facebook in some features.

Thoughts on pension insurance

According to the latest data from the Ministry of Human Resources and Social Security, as of the end of March the number of people covered by national basic pension, unemployment and work-injury insurance was 1.071 billion, 244 million and 297 million respectively, up 4.42 million, 2.56 million and 2.71 million year on year. As a “new type of work-injury insurance”, occupational injury protection for people in new forms of employment continued to be implemented steadily in pilot regions. As of the end of March, more than 11.04 million platform workers such as food delivery riders and ride-hailing drivers were covered. (Xinhua)

The total national population is over 1.4 billion; 1.07 billion covered shows that basic pension insurance has already hit its “ceiling”, and basically everyone it can cover has been brought in.

Riders pay insurance too; the state is gradually using institutions to catch the flexible-employment groups that used to be “running naked”.

Population ageing + a rising number of retirees + a falling dependency ratio means that over the next 10-20 years, the problem of pension payouts exceeding income will become more pronounced.

As the system matures, the room for small businesses and freelancers to “hide” from social insurance will keep shrinking.

Just tipsy, close to drunk

How to describe it.

I can probably still type, but I’m tired.

I really want to sleep.

I can be clear-headed, but I can’t stay fresh all the time.

I know what kind of person I want to become, but I’m not sure whether I can keep at it to the end. Wow, so I still approve of myself after drinking. That’s enough.

Aotian (the dog) is lost, very sad, and I analyse why I’m sad.

It feels like I wasn’t friendly enough to it before….

Aotian was adopted by the neighbour next door. They’ve adopted four stray dogs already, well then.

Maybe letting go will make it happier.

But I’m still very sad,,,

I don’t want to leave you alone drifting in the sea of people • Maybe in the future you’ll find someone better who understands you and loves you • I only ask that you have a happy life.

If I fall for you, I’d really want to stay with you forever.

@人生 #人生