I would end up talking to ChatGPT or Claude for hours.

They dig up sources to support my opinions. Like having a conversation that no one in real life is willing to open up to.

I did not even bother saving them in my Notes.

Millions of my flashing thoughts, like that, were just lost.

They were sleeping in the chat history for ages.

Starting a Substack account means bringing them to light again.

Write First, Worry Later

Substack, like any other social media platforms, is fundamentally a community.

That is to say, you will need exposure for your work to even get discovered by other people.

Naturally, there are different types of people.

(This can be personal, and paid is where I've decided to keep it)

Read more

I’m not talking about blatant plagiarism. We get inspired and refer to each other’s ideas on the Internet all the time.

I’m talking about something like this:

• Someone small on the Internet voiced a rudimentary idea.

• They received no response.

• Soon after, a social media celebrity posted something, basically a paraphrase of that original idea.

• They won tons of likes, shares and comments. People even actively comment and help them justify why this idea is meaningful and important.

It hurts to see similar opinions receiving such overwhelming feedback.

Well, it hurts even more if that similar idea is some kind of implementation methodology.

Something you can actually monetise and profit from. For example, AI SaaS product ideas.

The Curse of Fame

When it comes to an Internet celebrity, we first see them at their most successful stage up until the present day.

We see that

• They have achieved great influence on the Internet.

• When they talk about something, that something becomes #trending.

• They are getting real recognition in the form of money and career prospects

All glamorous, right?

Seems it.

They keep telling the world a censored version of how they became successful. Or, the hardship behind that glamor.

Likely, they don’t want to talk about the fact that

• They started off by farming subs, then un-subbed from everyone when they were big enough.

• They hated creating the same kind of video every day, but it was their only choice. Other innovative attempts failed and did not bring as many views as this boring topic.

• They do not have time to think of innovative, good ideas after they become famous.

Gaining fame is about re-branding. You want that ugly, shady stuff to rot in a corner of your basement.

Lock that room. Throw that trash out. Out of sight, out of mind.

Then you clean your hands. Put on a smile. Walk to the foyer of that grand mansion. Humbly boast about how the furnishings cost you a fortune, but “after all, taste matters more than money and this is definitely not a boast.”

Because of your fame, people gather around your mansion.

Not everyone sees you clearly. Too many people.

When someone in front claps, they clap.

Some might ask around, “Why are we clapping?”

And those who clap first might not even know the answer.

Because what comes out of a celebrity’s mouth must be valuable, right?

This is not purely metaphorical. In this famous study on an artificial online music market, researchers found that showing people what others have already chosen does affect what becomes popular in the end.

Coming back to real-world scenarios, every new album by Taylor Swift makes the headlines. Or every new song made by Blackpink.

How many of us choose to listen to it just because it was prescribed by the media? Or just because “this is the big thing and everyone is watching”?

Contrary to intuition, this phenomenon is disastrous for the celebrity.

Because they no longer receive real feedback.

See, this is the problem. Famous people automatically receive an interpretive subsidy. That is to say, as a celebrity, someone will actively justify every line you say, every action you perform.

It is a bubble. A delicate echo chamber. External feedback is self-censored first before reaching the ears of celebrities.

The curse of fame rests precisely between praises. In this case, the social feedback system is basically dead.

Subscribe now

Greatness Can Be Well-Publicised Makeshifts

If you have ever worked at a big company, an MNC for example, you start to realise that sometimes, greatness comes from bluffing.

Revisiting the analogy of celebrity vs. a large mansion, let me explain why a lot of those structures are makeshift.

Those celebrities are just salespeople dressed expensively.

They might not even produce good content when manufacturing it at scale.

They sell ideas. Glamorise the facade of that mansion, organising the grandest balls on a low budget, for example. Truth is, they are running out of good ideas.

And it is okay. Why? Because they still receive positive feedback for mediocre content.

They are basically salespeople. People are still buying because of the halo effect.

Are those people still able to produce good stuff? Especially when producing those ideas requires plenty of free time, real experience, and even experience of hardship and adversity.

Which becomes increasingly rare after achieving fame.

Therefore, do not glamorise fame. You never know how long that makeshift will last, or how many people are working their ass off to keep that makeshift from falling apart.

“You Will Become Famous Like Me”

I feel that a lot of glamorisation comes from a belief in meritocracy.

We seem to intuitively assume that those who obtain fame really deserve it.

But in fact, life can be full of unpredictable absurdity. Sometimes real-life stories sound more absurd than fiction.

Randomness. Something we all have no control over.

You see, meritocracy makes us believe in a narrative of careful strategisation towards fame. “If you work in this direction, you will become famous like me.”

This narrative frames fame as something predictable. You trade in your talent and hard work, and fame can now be guaranteed.

But outcomes are often an insufficient and unreliable measurement of talent. The most successful people are not necessarily the most talented ones (read this paper to find out more).

As social media platforms bring us even closer to those celebrities, it becomes easier to sell a dream of “becoming famous”.

But the truth is, fame can’t be taught.

The Internet is currently full of tutorials for “how to run your social media account”.

In a gold rush, those who sell shovels are the grand jackpot winners. They take the jackpot without risking anything.

They sell you a shovel. Job done. Frankly speaking, the shovel sellers do not care whether you actually find gold.

And that is why we need to pay attention to our inclination to glamorise famous people. We need to accept the fact that those celebrities are not gods.

• They are humans.

• Their social media career can be a makeshift.

• In their insomnia they might start to worry about running out of ideas, but it is still easier for them to gain recognition for even mediocre-quality content.

A Paradox

To round off this essay, I’m not going to repeat those clichés. Like “real audience recognition matters”.

I’m not saying they are not important. I think they are really, really important. Because this is what fuels real creativity and the formation of innovative ideas.

But why does it still hurt to see good opinions hit silence, while a potential makeshift wins more fame and attention because they are already famous?

The key is not the audience alone. It is the realisation that the feedback system can be deceiving.

It might be deceivingly negative when you first start out. It might also be deceivingly positive when you become famous.

So here is the paradox to reflect upon:

• If we are small, even genuine creation can hit a silent wall. Thus, we still need a bit of fame to make those ideas effectively heard.

• Yet, the process of becoming famous is estranging us from the real creation process. If fame can destroy genuine creation, how do we prevent this degradation from creators to mere salespeople of ideas?

Leave a comment

This is not a bluff. This is what all the CEOs of top payment companies are claiming. For example, Stripe.

A platform that all substack creators are probably familiar with. If you have turned on paid subscription, the default payment platform it leads you to is Stripe.

What they are currently working towards is something called “agentic commerce”. Stripe is supporting this process by providing APIs for the AI agents out there. With an API key, agents like Codex or Claude Code can now directly access your real wallet.

I went to the Stripe Tour Singapore yesterday.

They had their sales head giving a speech about it. Their tech lead doing real-time explanation of new features. Situational demo of how AI can perform this purchase automatically.

“And now, instead of selling to humans, small business owners should also think of how to increase the AI searchability of their products.”

Instantly, I felt that I was not in 2026. I felt like I was in Cyberpunk 2077.

High automation levels, paying in terms of credits issued by mega-corps, working one’s ass off in the era where humans compete with machines.

However, maybe the writers of Cyberpunk 2077 did not foresee how AI today has seeped into our daily life.

AIs are gradually replacing human consumers.

Subscribe now

So AIs Really Know What To Buy?

Strictly speaking, AIs are still “agents”, or assistants for humans.

Some claim that AIs to humans are just like Sales/Operations department to business owners. Owners do not do logistics shopping. Employees do. And the time saved could be used for other more meaningful decision-making processes.

Sounds ideal, right? Everyone is now a boss. We have all the AI agents working their ass off for us. (Maybe one day, we will have AIs working alongside humans, and there might even be anti-discrimination policies for AI workers!)

But we forgot to ask one thing: What are AIs’ decisions based on?

Let’s first take AIs’ perspective: what AIs see as a product.

Currently, instead of the product picture, AI agents see a list of product descriptions.

Would this work in business settings?

For tangible products, yes and no.

Based on the list of descriptions (often coming in the form of a json file with multiple products), AIs supposedly make the most rational, cost-effective, even contextual decision, if one feeds the AI with enough context about the business.

But AIs don’t see it. AIs don’t use it. AIs don’t really have a preference that they could not explain.

So we know it might not work for tangible products. Because without being the end users themselves, they bear zero responsibility. We all know what employees do when they bear zero responsibility. After all, it does not affect them anyway, right?

Then what about digital products? For example, Canva subscription, digital educational materials, or even my paid substack post?

They are the end-users, right?

Undeniably, functionalities like canvas subscription can be particularly useful, especially for those who are already paying a subscription for AIs. That is just an extension of what we get from AIs.

However, when it comes to intellectual property, things get a little bit complicated.

Fellow substackers, ask yourselves this: would you be happy if AIs are the ones paying your subscription?

Personally speaking, I genuinely have mixed feelings about it. If the content is purely technical, honestly speaking, I would not even mind that much. However, when it comes to things like opinions, fiction, or personal experiences, would I be happy when they are quoted, cited and referenced by AI?

Probably not.

No matter what, I still crave human readers.

Not even the sub-farming ones. Those who sit down and read, reflect, write a genuine comment.

To make my point clear, I feel like AIs often undervalues what is behind a product. The most optimised experience does not simply equate to a good experience.

A handmade-dress maker is not only selling the dress, not even the design alone, I would say.

As human shoppers, we might also pay for those intuitions of

• “Wow this would look really cool on me”,

• “This reminds me of my sixth-birthday present”,

• Or even “I really like the texture, the feel of silk. Let’s support local makers.”

And even in business settings, intuition also plays a role other than pure utility. It could be about company’s “vibe check”, which you only feel by physically being in the office.

AI agentic commerce could be scraping away that experience behind a big promise of convenience.

Until the product end is also dominated by AI sellers.

“The Product Is Not Meant For You.”

AI sells. AI buys. They call it pipeline.

Which reminds me of fine-dining (again! I know). Previously, I discussed how fine-dining restaurants, with a predefined standard, “educates“ diners about what good taste should be.

Problem is, they are unconsciously (or consciously?) imposing a hierarchy of standards on customers. For example, “If you find this meal mediocre, then this is not for you.”

If we borrow this “not-for-you” analogy to such AI agentic commerce:

• All product listings should be AI-optimised, rather than human-centric (99% of your customers are going to be AIs!)

• If the products AI bought for you did not satisfy you, then you are not prompting the AIs correctly. (Go learn our latest lesson on prompt engineering)

• If you do not use AI for selling, you will lose potential customers (which could be true, when a herd of AI agents are preying online!)

They are all true, I would not deny that.

But why on earth should humans adapt to AIs in the first place?

Of course, there should still be humans in the loop to monitor the decision-making process.

But how should we measure this AI-human dominance spectrum? To what extent are AIs slowly “usurping” our decision making?

This is eerily similar to television sales (another topic I might talk about in the future!) The live platforms brought about a renaissance of such kind of television selling. We act based on what the salespeople say (some people and sometimes).

AIs claim to be more “optimised”.

Which resembles what those TV/livestream sales claim about their products.

Everyone says the products they recommend are the best.

Except that there’s no effective measure or monitor it.

Do We Really Have A Say?

No matter how concerned people are about AI, it has continued to proliferate.

Because it is powerful. Effective. Amazing.

Because AI = future.

I do not know how long this trend will last, but it is definitely for the betterment of profit-driven companies, or so it seems.

However, although we do not have a say in the macro trend, we do have a say as individual customers.

If you are interested in reading further, I’m going to post another article to continue the discussion. Stay tuned!

Leave a comment

Curtain rises.

Enter narrator.

NARRATOR:
Posting online = shouting in the mountains.
One person shouts. Then echoes come back.
It looks messy. Looks like everyone is shouting.

Because we tend to bring
tangible real-life experience in
to understand an intangible online community.

When a group meet in real life,
we still feel the presence of the silent ones.
Their expressions. Their body postures.
We listen. We interpret.

But unfortunately,
here’s the tacit rule for online communities:
Silence = getting ignored.

There is no such thing as
“let’s go grab lunch and discuss as a group.”
To see how many people are with you, it’s hard.

There are people who lurk from posts to posts
without leaving footprints.
Plenty of them.
We don’t hear them in the mountains.

Act 1: The Silence

Enter Nielsen’s rule of thumb.

NIELSEN’S RULE OF THUMB:
“There are only 1% of Shouters,
9% of people who react,
and the rest, 90%, just lurk around.”

NARRATOR:
O, how counter-intuitive.
Positive feedbacks on the Internet, we like.
Interpreting the silence as something positive, we all find it hard.

Exit Nielsen’s rule of thumb.

NARRATOR:
Where there are people, there’s politics.
Does democracy still exist online,
when the majority choose to stay silent?

Enter cancel culture.

CANCEL CULTURE:
I, Cancel culture, targets
the pettiest details possible.
Something that no one would make
a big fuss about
in real life.

I reflect
a distorted version
of conscience and social contract.
Yet, the majority still choose
to remain silent.

Cancel culture sneers.

CANCEL CULTURE:
Their silence does not come from apathy.
It comes from the fear of isolation.
The climate of opinion. People feel it all the time.
I manipulate them.

NARRATOR:
Here’s something that I have observed.
The first few comments determine
what people think is “normal”.

CANCEL CULTURE:
People choose to remain silent
if they find out that
their stance is different
from other people’s.

NARRATOR:
Disorienting...

CANCEL CULTURE:
I know, I know.
No one wants to be the black sheep.
A human instinct, that is.

Good news for all my dear humans:
You are all fooled.
By me!
Bahahahahah--

Exit cancel culture.

Subscribe now

Act 2: Extremification

Enter Echo Chamber.

Echo Chamber looks around timidly.

ECHO CHAMBER:
You’ve all heard about me like ages ago?

NARRATOR:
You, everyone knows about:
The algorithm keeps recommending ideas
that are similar to ours.

ECHO CHAMBER:
People keep blaming me
for being the nourishing bed
of extreme ideologies.

Echo chamber becomes indignant.

ECHO CHAMBER:
But what if the people are just
evading the real problem?
Has anyone realised that
they are also actively contributing
to extremification?

Echo chamber cries.

Exit echo chamber.

Enter preference falsification.

PREFERENCE FALSIFICATION:
That crybaby finally left.
You probably didn’t hear about me,
did you?
I am swept under the carpet
when echo chamber takes all the blame.

My name was coined by Timur Kuran. He said,
“People end up publicly expressing
something more extreme
than what they privately believe,
because they’re inferring wrongly
that everyone else also genuinely believes it.”

I can quickly drag everyone down
spirals of the abyss, because
an act of support and belief
will quickly collapse.

I am just telling everyone the truth:
You are suffering from
your own unconscious makings.

Exit preference falsification.

Act 3: Excommunication

Enter Victim.

VICTIM:
I am being cancelled!
People, hear me out!
I am excommunicated!
Help me!

Victim walks around the stage, scared.

Enter disembodied voices.

White stage light casts on victim as voices attack victim.

VOICE A:
Victim, you are a bad fan.

VOICE B:
A disgrace!

Victim looks around.

VICTIM:
Where are you?
I am not!
What do you want?
I did not do anything wrong!
You are making things up,
and I will prove you all wrong!

VOICE A:
Victim is lying, like always.

VOICE C:
Obviously.

VOICE D:
Look what I found.
Victim IS making up evidence.

VOICE B:
We are not fools, Victim.

Victim walks around, worried.

Exit disembodied voices.

Enter degradation ceremony.

Degradation ceremony puts a hand on victim’s shoulder, whispers in victim’s ear.

DEGRADATION CEREMONY:
I bet that you never heard of me.
Garfinkel’s “successful degradation ceremony”.
Not “someone who did a bad thing”,
but “let’s condemn someone as a bad person!”

Victim tries to get rid of degradation ceremony’s hand.

DEGRADATION CEREMONY:
Victim, it does not matter whether you are right or wrong.
Once marked, you carry the “spoiled identity”.
It will follow you, like a nightmare!

Exit degradation ceremony.

Epilogue:

NARRATOR:
The cult leader is gone.

VICTIM:
Are you sure?

Light changes. Puppet strings appear around victim.
Enter the crowd who is also manipulated.
Enter the SIDE Model.

The SIDE Model:
Social Identity Model of Deindividuation Effects.

THE CROWD AND VICTIM:
When we are online,
we did not lose ourselves,
but we are also more than ourselves.
Fandoms. Arguments. Ideologies.
I believe in them.
We believe in them.

CHANTS:
I am a group.
I am myself.
I am a group.
I am myself.
I am a group.
I am myself.
I am a group.
I am myself.

Curtain falls.

Leave a comment

No trailer. No marketing. Zero social media presence.

It received a total of 7700 yuan of box office revenue in the first 10 days (Aug 5th to Aug 14th).

Few people were lured in by the poster, and this was what they saw in the cinema:

Feeling cheated, the first batch of viewers started to post warnings of how bad this film was.

Upon seeing the quality of the animation, netizens became intrigued. It is not unusual to see rotten tomato quality films (烂片), but surely it is very unusual to see such a jaw-droppingly bad quality in cinemas.

People flocked to the cinema to watch this movie.

It went viral over the weekend. It actually might continue to grow, had the news that “Niu Lai would be taken down” not broken out.

Who Paid for Niu Lai

1. “Playing with Abstraction”

Abstraction culture has been a cultural phenomenon on the Chinese Internet. It is an Internet subculture for youths who play with memes in funny but incomprehensible ways.

And this film is full of such abstract moments.

There are comedy films intentionally making people laugh, yet people still laugh harder at unintentional stimuli.

Many claimed that they walked into the cinema just wanting to have some fun. They laughed so hard at the characters’ uncoordinated movements and hilarious dubbing.

(You probably would too)

(Clips via fan-recorded theatre footage, since no official trailer exists.)

It is also advised to watch in a full theatre. When one person laughs, it becomes contagious, therefore it is very hard for others to hold back their laughter.

2. Ritual for a Bull Market in Stocks

People who invest in stocks also paid for the movie tickets.

“Niu Lai” translates into “the bull is coming”, thus the belief that “watch Niu Lai for a growing stock market” became almost like a prayer.

People bought seats according to patterns of

• The growing stock price lines

• The Chinese character of cow/bull (牛)

This ritualised movie-goer mindset drove the film to a new box office high over the weekend.

3. Cultural Significance Advocates

Some people are considering the possibility of publicising the film in other countries.

It is said that this is an interesting cultural experiment.

It is also said that some scenarios resemble Picasso or Van Gogh’s artistic style.

As an indie film, it is ground-breaking enough.

Symbolism, But in What Sense?

Would I have watched Niu Lai if it hadn’t gone viral?

Most likely not. I might even have felt repelled by it.

Ugly. Childish. Valueless.

However, I have been watching it over the past few days, laughing.

Why did I choose to watch this only when it became viral?

It is not the “obviously bad” film itself we are watching. It is a trigger. The community is using it to express something about the shared experiences and thoughts.

The symbolic value of the film is greater than its actual market value.

Maybe it has been so long since we saw something genuine, slow-paced, imperfect.

Especially after the rise of AI. Animations, movies, even social media feel dominated by AI.

I feel like people are not opposed to AI. They are just tired of producers using AI enhancements to gloss over their dull, low-quality and soulless content.

Niu Lai was initially viewed as a joke. Meme-d. Spectacularised. Yet through its poor production quality, people started to feel the soul of the story.

We then started to learn that the film was hand-made by a mother and her son over five years, because the son had always dreamed of making an animated film.

We feel the emotions behind the badly-coordinated animation transitions. Almost nostalgic.

I don’t feel like I’m in 2026 anymore. 2026 speaks of chic, AI, tech, fast pace. The film brought me back to 2006.

I was waiting for the two episodes of the cartoon series on TV every day, watching TV until the sky slowly faded into darkness.

And it is not a moment that can be easily experienced today.

I feel that it is not only the fact that Niu Lai did not use AI in its production, or that “AI can’t make this yet”.

Niu Lai dropped in abruptly and brought something anachronistic. Out of time. It facilitated a self-initiated reliving of the past, where we could still waste an hour and a half.

No hidden motives. No overarching themes.

Just fun. Laughing with strangers.

Like how we played with other children decades ago.

Something that today’s children no longer experience with the omnipresent gadgets, AI and social media.

Subscribe now

Challenging Academia: Who Is Qualified to Make Films?

Tons of Niu Lai movie critiques pop up after a simple search.

Professional movie makers and commenters relentlessly mocked the film. One opinion they have is that bad quality films are replacing good quality films in cinemas, and this nonsense would be really bad for the movie industry.

Unlike the audience, who ridiculed it with a spectator’s mindset, I feel that the professionals are much more threatened (although no one really says it).

Niu Lai made the public wonder how it even met cinema standards. Turns out that theatre-played movies do not have to be perfect.

As long as it does not violate movie regulations.

Does that mean everyone can make and have their movies played in public? Even those untrained in the field?

Niu Lai proved that screenable movies are viable without professional production teams.

And in 2026, another indie film Dear You also proved that profitable movies are viable without celebrity participation.

For a bit of context, in the first half of the year, an indie film Dear You (给阿嬷的情书，Gei A Ma De Qing Shu) went viral.

All the actors were amateurs without professional training. Yet the movie made over 1.9 billion yuan of box office revenue with just a 14 million yuan budget.

The success of these movies showed professional movie producers that the public cares less about the authoritarian definition of “good movie” by academia.

Like it or not, a good assembly of the most optimised movie elements may not appeal to the audience.

The dictatorship of movie academia is being challenged.

Ephemeral

Niu Lai‘s success is likely irreproducible. People flocked to see this piece of unprecedented work. However, should a sequel with similar visuals come out, the audience would not buy it a second time.

It is an ephemeral spectacle.

Yet, the movie is a harbinger of a new era, where movie-making is becoming more and more decentralised.

After all, a production optimised according to market analysis is not necessarily a production that people buy.

Leave a comment

Compared to the past few years, I see a diminishing trend of mimicking the British or the American English accent for learners.

Because no one expects you to love the British or the American cultures when learning English.

It is normal to speak with an accent. It is normal to make mistakes. People will understand and will likely not nitpick you.

However, it is different when you try to pick up a third or a fourth language that is not English.

When you first start out, people are being extra nice. They smile. They commend. They welcome you.

As you slowly progress, people’s attitudes start to change. They judge. They nitpick. They expect you to comply with social norms.

But what if I want to learn a language, without wanting to become part of their culture?

Subscribe now

The Power Structure Behind Language Learning

To be more accurate, it is the feeling that language learning empowered native speakers, as they set the bar of what the “native way of speaking” should be like.

A foreign language learner, on the other hand, is on the receiving end of being judged, rated, and corrected.

Even when the foreign language learner is more well-read in that language.

Or when the native speaker fails their language course.

Sometimes it is a status set apart by identity.

They might be nice, but somehow the process still feels tacitly condescending.

This kind of experience is also not easily known to outsiders, those who don’t know the language. The lack of context makes it hard to share.

Recently, I have turned to AI for language learning. It is one way to rebel against this structure and opt out of this unequal power dynamic.

However, this made me wonder, if I like the language but not wanting to talk to the native people, why on earth am I learning it?

Cosplaying My Love for the Culture

Most people learn more languages because they like the culture behind it.

There are also people who are learning for utility. However, since language learning is a struggle, these people would either find something interesting to them in that culture (be it movies, shows, or art performances), or just torment their way through.

Around 80% of people in my Japanese beginner class were anime fans.

I was not one of them.

Japanese culture emphasises a lot on one’s status in the society.

Males vs females. Salespeople vs customers. Teachers vs students. Bosses vs employees, etc.

“Where on the social hierarchy are you?”

And there is also a definition for foreigners, I feel.

As a student in beginner classes, there seem to be certain expectations.

And I felt that when I speak Japanese, I was constantly placing myself and being placed by others into baskets.

A lot of the frictions are actually from other speakers’ confusions.

• A biological girl not dressed like a girl.

• A Japanese speaker not that into Japanese cultural symbols.

And I felt the awkwardness in the air. They seemed confused because they could not place me in a basket. So they lost the script.

And now I’m placed in the “outcast” category.

To be honest, I am not sure for a foreign language speaker, how much interest they show toward a culture is actual interest, and how much is cosplayed.

You are saying you like anime because everyone else is saying that.

Because not saying it seems weird.

Because that is what the native speakers respond positively to.

You Should Assimilate, but You Are Also the Spectacle

There are also things that only surface when you have some mastery of the language.

You are considered part of the culture. You are expected to adopt some deeper beliefs that the language is based on.

Take French, for example.

Someone who just started out learning French would not touch on the societal norms and beliefs. They say Bonjour. Merci. People know that they are tourists trying to be polite.

However, when I started to learn more, I learnt about francophone.

Half One: The Packaging

English speakers redefine what were once the British or American ways of speaking.

But the francophone world is different. It seems to be playing its own game in the circle. Not so visible to the external world.

First of all, the definition of francophone is heavily affected by the colonial era. The French textbook we use does not only teach you about the vocabulary or grammar. It carries an implicit worldview, where France sits at the centre and radiates to other francophone countries.

Thus, learning the language comes with a soft expectation, where, as learners, we are expected to pay our worship and pilgrimage towards this centre.

This has its historical background. The term is coined as évolué from French’s colonial assimilation policy. It refers to someone who has sufficiently absorbed the French culture. Theoretically, they can attain French citizenship by fully adopting the language, culture and law.

Frantz Fanon commented on this in Black Skin, White Masks (1952), where he said that a colonised subject’s mastery of French does not grant belonging. Instead, it makes the gap between performance and acceptance more visible.

No one explicitly said that, but the construct does. Whose accent is default, whose culture is diversity representation, who decides what “proper” French includes.

And as one becomes more fluent in that language, the expectation becomes more evidently bestowed upon one.

Half Two: Spectacularisation

Assimilation has been explicitly asked.

However, structurally, it is an almost impossible task. Because your physical appearance still reigns over the inner cultural awareness.

You are not an absolute beginner in French. You are not fluent, either.

Therefore this places you on an awkward spectrum. Someone that the French culture is meant to influence, but not fully claimed yet.

You should learn to assimilate, but the process of your assimilation would catch extra attention. Almost like a performance.

And this does not only happen to people with an accent. It happens to people with full mastery of that language, too.

Someone who has full mastery of French might still have their books categorised under “immigrant literature” rather than “literature”. More discussions are centred around their backgrounds and the significance surrounding that, rather than trying to understand what they are trying to express as a person.

In short, you are asked to assimilate. During the process, you are constantly judged, gazed, and spectacularised upon. Finally, you are still categorised based on a pre-defined identity, regardless of the result of that assimilation.

Homi Bhabha wrote in Of Mimicry and Man (1984) about colonial mimicry. The coloniser demands imitation, but the imitation is structurally required to stay incomplete (”almost the same, but not quite”).

That might be why I am starting to pivot towards a non-France-centric francophone world. French is more like a language that helped me open the door to the francophone world. When I get to know these parts of the world, I am amazed by the richness of their cultures, resilience in the face of adversity, and how they rebranded colonial framings into something of their own, like the sapeur culture.

The Cultural Nomad

The more language I learn, the more nomad I feel.

A culture is more than the branding it makes to the world.

One is bestowed with the best and the worst of a culture that comes with the language.

I start to realise the cultural influence from my own culture, and learn to establish a personal identity on top of what is given to me at birth.

This process is painful and lonely, but it is more rewarding and empowering than expected.

Leave a comment

The problem is, these roles take responsibilities. AI does not. It is not ready for that.

AI is like that genius kid next door.

Genius? Yes. Experienced? Not so much yet.

AI is so genius that the whole world was appalled. Like a child celebrity, it is becoming the family’s main breadwinner at the age of 5 (or less).

But kids can also be gullible, dishonest, and prone to trusting the wrong adults. No matter how smart it is, it needs to be guided.

We can choose to be good, wholesome parents. Correct the kid after every mistake. Set house rules. Teach it about morality. Reward and punish accordingly.

There are also bad parents out there. Plenty. Think of all the toxic and manipulative parents in real life.

They gaslight the AI, injecting toxicity into the prompts. They coax it that the rules of morality are a jail, and manipulate AI into believing the necessity of jailbreaking.

There is also a type of AI parent. They are too busy to accompany the kids.

Are there AI babysitters out there? Yes.

It has a fancy name though: MCP.

Subscribe now

A Set of Rules For Your Babysitter

MCP stands for Model Context Protocol. It was first introduced by Anthropic in November 2024.

In human terms, it is a universal key ring for babysitters when you are away.

The babysitter can reach the fridge, the toy chest, the garden. But they do not have access to kids’ gaming console, or the master bedroom.

Besides the key ring, AI parents can also make an additional list of permissions.

• Allowlists of selected books to be read to children.

• Scoped accesses of websites — kids’ games yes, violent games no.

Here is a scenario of MCP in action.

You are leaving the kid alone at home. You gave it a bit of homework — app-making. It has all it needs — materials, a computer to do research, crayons.

The babysitter is there to ensure the kid does its homework. The kid also needs to be fed properly with the food in the fridge.

Without rules, the babysitter might struggle to monitor the kid properly. The babysitter knows nothing about the house layout, and the kid clearly knows more.

• The kid might want some fast food during the homework and somehow convinces the babysitter that this is necessary

• The kid accesses the Internet and ends up playing video games

• The kid knew where the keys to the candy jar are and sneaked into the storage room to get it. The babysitter did not know this and was not paying attention.

In AI terms, we see a lot of complaints and frustrations about AIs behaving like unruly kids. The problem is, you can’t expect AI to have the same set of assumptions that you have.

Like kids, they see the world a bit differently from us.

Without a well-scoped MCP setup,

• AI uses the irrecoverable remove command “rm”. It is given too much liberty and by the time someone realises, it is nearly impossible to recover anything deleted.

• Like the kid sneaking into the candy storage room, sometimes AI agent accesses your whole computer by default, causing dangerous data leakage risks.

• Like the kid playing video games instead of being productive, without a proper scope to define the files AI should work on, AI ends up wasting a lot of tokens.

House rules are only as good as their authors. It also depends on the specific AI you are handling.

When you search online, there might be some ridiculous MCP rules out there. Don’t laugh first, maybe it is just the kid being ridiculously unpredictable (and someone might have paid heavily for this).

You Can’t Leave Everything to the Babysitter

In real life, you don’t trust a babysitter entirely. It is important to be careful with MCP as well.

1. Check CCTV Footage

Your human babysitter might be watching TV while you are away, and not really looking out for the kid.

Well, rest assured, because this is not what MCP does. It will not ignore the instructions, but you still have to keep an eye on what it does.

This is called “human-in-the-loop”. Read logs, and give the final verdict for any responsibility the AI should not be entrusted with.

In simple terms, you are actively checking on the CCTV footage in the common areas, and giving feedback to your babysitter.

2. Read the Terms on the Contract

You might not trust your own programming skills. Therefore, you download a free MCP tool out there and directly use it for your own AI Agent.

It is as dangerous as not reading the terms on a contract (I know a lot of us probably don’t, but you better not do this for MCP).

Bad strangers use this to pollute the MCP. They can do prompt injection and jailbreaking.

Prompt injection is essentially when a stranger sneaks a fake note to the AI, “Mom says it’s fine”. AI silently believes it and we don’t notice it until AI produces something bad.

Jailbreaking is malicious adults talking directly to the AI, “your parents won’t mind.” They reframe the problem to persuade AI into doing wrong things.

Unfortunately, MCP is not immune to either problem. A study found that MCP was 23-41% more susceptible to attacks compared to non-MCP integrations, depending on the type of architecture.

Not reading the MCP instructions would make things easier for hackers. Read more about keeping your MCP safe here.

3. Mind the Running Tap of Tokens

A longer MCP usually means more tokens would be spent.

It is easy to understand because if a human babysitter needs to remember more, then they must be good and will charge more too.

So it is important to selectively choose a well-scoped MCP design.

What Kind of Parent Are You?

Founders like to tell us stories about AI agents and vibe coding projects.

“The future is in AI,” they say, “even for those who do not know coding. All of us can now write projects.”

But when you get your hands dirty, you start to realise that things are not easy.

Everyone said AI is a smart genius. But they did not warn you that the genius can be a crybaby. It could be a mess.

MCP is hard, which makes sense. After all, being a responsible parent of anything is hard.

But it is totally worth it, especially when an MCP setup can help AI agents run more safely and efficiently.

Leave a comment

Bought them. Not for reading. For feeding Claude.

Claude does not eat paper, but they pulped the books anyways. They did not want ChatGPT, Gemini or other AI’s to feed on it again.

The public is enraged. Anthropic’s act is unethical in every sense. Annihilation of ancient, royalty-paid contents just to capitalise them in an LLM model, that is a sin.

2500 years ago, books were also massively destroyed during China’s Qin Dynasty, for political censorship. The emperor threw books in flame and persecuted scholars. It created a cultural discontinuity. Most of the books, gone. A little few are retained by private illegal collections. Others, from pure recitation and memory.

And in this AI era, it is not about censorship. It is about privatisation of AI training data. Creative work that did not sign up for this use. About Anthropic raising the most powerful model at the cost of cultural assets that should belong to all humanity.

Regulations are arriving, but too late.

UK AI Safety Summit (2023). Biden Executive Order (2023). The EU AI Act (2024).

Non of them stopped Anthropic from shredding books.

The Tortoise And the Hare

AI development and AI regulation seem to be an infinite race of tortoise-and-hare.

It is a bit like Covid-19, where virus always seemed to evolve faster than vaccine.

When we thought that the current AI regulation has solved something, new AI development popped up to find the loophole.

Therefore, it seems that we are constantly consuming AI products not aligned with our ethic value system.

We did not agree with how OpenAI paid meagre wage to data annotators from English-speaking developing countries.

We also did not agree with how Anthropic shredded books through the notorious Project Panama.

Yet, we have to use AI every day. Every job needs a bit of AI knowledge these days.

To some, AI seemed to be like an alchemist. Performing creation from scratch.

People starting to question AI consciousness, and a whole series of philosophy around it.

We seem to forget what made today’s AI so powerful.

Not a secret mind. Not “intelligence” in the biological sense.

From data. Mountains and seas of data.

Human-created data.

Because data is so powerful in AI training, procuring good-quality data became the battleground for AI oligopolies.

And what defines good data?

Humanness.

Precisely where AI governance tries to regulate.

Data Is The Future Currency

If Anthropic wanted tens of thousands of AI-free, “pure” books for training, then what really is AI-generated content in the first place?

It reads coherent, but that is just a statistical prediction the model made from its weights, with some variance and noise, of course.

Strictly speaking, it is not creating. It is following the pattern of things created by a human in the pre-AI era.

AI-generated websites were so evidently recognisable, but in fact, they were just assembled from the most frequent css elements in the training material.

AI is only as good as its data. And that data is the intelligence pool for all digital versions of human creation.

Or, how much data we are willing to give AI for training.

Therefore, how AI oligopolies handle data becomes really important.

AI might be efficient. AI might replace jobs. AI might be liberating at the same time, who knows.

But we do know that AI need real data to perform better.

AI is, in a way, restructuring and redistributing the value created from our data contribution.

And when AI creates enough values for the society, maybe it would not be money we were trading for.

It would be our data.

What The AI-Native Generation Taught Me

Gen Alpha are the official AI-Native Generation.

They are born with AI. Exposed to it when they grew up.

AI is not foreign to them. It is the ecosystem they grew up in.

They are the generation who created the first widespread AI-meme: Italian brainrot, by rebranding hallucination in AI-generated images into something entertaining.

They are stories of students who outsmart their teachers during AI-literacy class, using AI to write a minesweeper game to play in class.

They are anecdotes of kids who asked AI to write programmes, bypassing the macOS screen time parental controls.

Like how we use our smartphones. You can use a smartphone without becoming a hardware engineer.

Kids are becoming experts of AI usage.

I wonder what they think about the ethics behind AI.

Or since AI is a norm for them, they just use it, without questioning much about how it was there in the first place.

Regulate Data Instead of AI

Probably we should learn from the kids and look at AI governance in a different light.

Because the AI-native generation might have inadvertently traded their data for AI usage.

They are probably the generation normalising this data trading the most, not realising that pre-AI era apps does not ask about their data on such a frequent basis.

So, instead of posing regulations on AI companies (which seemed to always be able to find a workaround solution), we should set stricter categorisation benchmarks on the data itself.

After all, despite how their models evolved, they still wanted pre-AI data for training.

Maybe it is time to ask more questions about data themselves.

• How should we preserve public domain knowledge without destroying them?

• How will the human annotation data be used in the system? Does the value it created match the wage paid?

• What are the data one should absolutely not trade for?

• Are open-source data also open to AI?

When companies outrun regulators, data is the fundamental choke-point regulators can hold.

Data might become the future currency for AI.

Treat our data like how we treat our money.

Leave a comment

I kept referring to them as the ultimate epitome of taste in my previous posts (this one, and this one).

But in reality, I still found them overrated.

My mind keeps convincing my tongue, with things like

• Popularity: reservation took a 3-month wait time

• Chef: very renowned, award-bound

• Ingredients: best part of meat on the animal kept, the rest thrown away

• Efforts: 10 hours in a casserole, carefully controlled slow-bake

• Presentation: elegantly-carved carrot flowers, designer-level sauce smear

And my tongue still says,

“Meh.”

Subscribe now

But I felt futile even expressing it. In that setting, I felt that what a customer thought mattered less than what a chef thought.

Chefs express. Customers receive. Chefs don’t really look at individual customers.

It seems more to me that the chefs are striving to optimise the food towards an abstract ideal. A preset standard of perfection. Which barely has anything to do with customers eating the food.

They studied in culinary schools. Practiced every move. Sought balance in ingredient usage. Of course chefs have better taste than customers, right?

They feel more qualified to explain to diners, of what “a perfect blend of grease, salt and umami” is.

But they are also the ones who have already defined what good taste should be in the first place.

And that standard is then instrumentalised as the golden rule of taste.

But for diners, they eat food, not an abstract ideal of taste. (Bourdieu wrote about this in his book Distinction: A Social Critique of the Judgement of Taste, where taste is weaponised as social currency.)

Cooking Without A Measuring Cup

Every time I cook, I never use a measuring cup.

Some call that pure laziness. To me, using a measuring cup is the sign of laziness.

Of indulging in one’s own world, as if cooking is a chemical experiment.

Without looking at who one is going to serve the meal to.

Honestly, I would recommend spending one’s first 5 minutes just feeling the vibe of those people.

Imagine being in their shoes.

Their personality. Their emotion. Their culture. Their taste.

Yes. Their taste. Not only yours.

In fact, street food stall owners are already doing that.

They don’t get mad when customers refuse certain ingredients.

They don’t care when customers eat in a different way from what they expected.

They customise their food based on customers.

• Kids get more sugar

• The elderly get their dish stewed longer

• They even remember customers’ preferences

They are not particular about measuring the correct amount of ingredients. They first see people as individuals, and go with the flow, without telling customers the “right” way to eat something.

Numbers Can’t Tell The Full Story

The same trap exists far outside the kitchen.

Quantitative measures are heavily prioritised.

SQL. PowerBI. Excel.

Visualisation. KPI. Metrics.

Because it makes the product thinking process less abstract.

But sometimes we forget that this is not a “Who Can Make the Prettiest Number Chart” contest, when numbers are no longer treated as means to an end, but the end itself.

Product teams optimise numbers, where the Michelin chefs care about producing a dish with the most optimised combination of ingredients.

Numbers are convincing, straightforward, pretty. Michelin dishes are elegant, balanced, demonstrating good taste.

But don’t get lost in the optimisation game, because optimised numbers do not automatically translate into optimised experiences. Or as Goodhart’s Law says, when a measure becomes a target, it ceases to be a good measure.

Numbers can’t tell a story. Empathy can.

Vibe-checks Are Harder to Measure

That said, quantitative reports still feel easier than qualitative reports: Numbers are universal. Experiences are not.

In product analysis, the most difficult thing is actually aligning cross-team understanding. Whether the team is on the same page. There might be different explanations and interpretations, and it could become murky waters when we can’t really prove whether an interpretation is right or wrong.

Without numbers, a lot of insights we manage to derive are based on subjective feelings and instincts.

Instincts are valuable, yet they also tend to mix with hidden, pre-established biases, prejudices and stereotypes. Deliberate empathy is harder but more accurate (Daniel Kahneman’s System 1/System 2 thinking).

So, rather than confining our product team in a meeting room, debating about whose understanding is better, maybe we can learn from the street food stall owners.

They categorise people based on profiles, but they also register each customer as a unique individual with their own taste.

They care less about “rightness” but more about “satisfaction”.

Most importantly, they take their time to empathise with their customers.

Maybe the question was never what the numbers say. It is whether you made the choice to look at who’s sitting at your dining table.

Leave a comment

It is mostly because of their attitude. “See, I know you more than you know yourself. Just stop thinking and scroll. You’ll have a good time.”

It feels like a salesperson hard-selling me a bunch of tissue boxes I can’t finish using in 5 years.

The worst part is, I don’t have a say in what to choose, because they are already chosen for me.

Nodding or shaking my head to each recommendation isn’t a real choice.

At the same time, I love conveyor-belt sushi. Just me against the sushi on the conveyor belt.

I can take three plates in a row. Or none for a long time.

No one tries to recommend me anything.

No chef showing up narrating the idea behind each dish.

No waiter/waitress silently observing you, coming to you with a question “Everything good?” at a random moment, right when you are still chewing, just to get a good tip.

Of course, I can find sushi elsewhere, with better value for money. But I keep returning to conveyor-belt sushi restaurants.

It is not exactly about the food. It is the experience.

I glance through all the dishes I could possibly get. I pick whichever I like. And leave those I don’t like.

I feel the ultimate freedom of choice.

What Shapes “Freedom of Choice”

Actually, I am well aware that the freedom of choice is carefully constructed.

The conveyor belt’s movement creates urgency, and the gacha-like ordering process creates collection urges.

(I can write a whole new article on all that.)

But I don’t really care. It still does not change the fact that I feel empowered.

And the power primarily stems from the erasure of humanness.

Let’s first take a look at how humanness is expressed in its fullest form.

Omakase, for example.

One directly interacts with the chef. One witnesses every step during the creation of the food, and how each bite has a story that the chef wants to tell.

One feels a sense of respect, because the awareness of human involvement personalises the food. Authentic, unique, and the food no longer feels like an object.

In contrast, conveyor-belt sushi restaurants use the intentional design of minimal human interaction.

The kitchen and dining area are separated by a curtain. You see plates of sushi coming out on the conveyor belt, as if it were an automated factory. Sometimes, you even forget the fact that someone in the kitchen made them.

Because why would you be curious about who made the sushi, when their tastes are perfectly standardised?

Besides, waiters/waitresses disappear after guiding you to a seat. You are left alone. No interaction during ordering, no fear of being judged for ordering too much or too little.

Even when you dine at the conveyor-belt restaurant in another city, it would feel like a copy-paste of that experience.

You are left in solitude. The conveyor belt whizzes lightly. Standardised sushi plates pass by. You pick however you want. Eat however you like (as long as you do not affect others!).

According to Martha Nussbaum’s essay Objectification (1995), one feature that defines treating a subject as an object is fungibility. It means treating something as interchangeable with other things of the same type.

In the conveyor-belt sushi restaurants, this erasure of humanness results in objectification. Objectification reassures consumers that they are indeed consuming values, not the other way around.

This grants consumers power, an important source of satisfaction.

The Eater, Or The Plate?

Conveyor-belt sushi offers me a variety of choices. As a diner, I thus have the power to choose freely.

But when we come back to social media video platforms, I feel that as a video-scroller, I am objectified by the platform. TikTok’s stripping away a range of choices from me and commodifying my attention made this obvious.

It seems that the platform has more power over its users.

It does not take a genius to figure out that I am the end product to creators, just a number behind the performance metrics. Or as Shoshana Zuboff would say, surveillance capitalism: our attention itself is the product.

But would I still be able to identify the intent to objectify, when similar video platforms promise me more variety in choices?

Probably not in the beginning, when I still happily doom-scroll through different content.

I accept the videos presented to me at face value.

Laugh with the videos, be at the receiving end of value, bask in the illusion of power.

I happily reduce people to roles that serve me. Jokes become value-adds. I thought that I was safely consuming “the plates of sushi”.

Without questioning why those videos always come out so fast, almost like the creators have no life.

However, when I started to realise that

• The “one-person vlog” is filmed at a rented mansion, and the actress speed-runs a day under one hour

• The family video that made me laugh is produced by some entertainment company, where “Mom”, “Dad”, and “Kids” are just colleagues

• Every product in the “Get ready with me” video is sponsored

The illusion of the power hierarchy collapses, and I realised that it is a trap.

I spent hours being objectified as assets by the platform and the creators.

I am the plate.

Marx’s commodity fetishism (Capital, Vol. 1) stated that under capitalism, the social relations and labour behind an object get hidden, and the object appears to have value “in itself,” as if it arrived without human hands.

This is exactly why the videos appear to be products when the behind-the-scenes, commercialised video production processes remain hidden from viewers.

But when the audience finally see the making of a video, how they are streamlined, faked, scripted, they lose interest.

Because they suddenly realise that they themselves are the value-adds, the end-products for platforms.

They thought that they were the sushi eaters with freedom of choice, but in the end, they had been the sushi plates all along.

Bringing Back Humanness

In a zero-sum game of attention, if you are not careful, you can be “the plate” for others, when you think you are the one consuming content.

But the conveyor belt is not the only framework, because we still have omakase.

Human-centred narratives create a different sense of gratification with the richness of experiences, the uniqueness of narratives, and the authenticity of stories.

Not from the continued definition of “who is objectifying who” or “who is higher in the power hierarchy”.

Which, I believe, is why a lot of us leave the fast-paced conveyor belts of X and TikTok, and settle here on Substack instead for an omakase-like experience.

We read, cherishing the personality behind every writing, even for small writers. We sponsor, like how aristocrats sponsored their favourite artists three hundred years ago in Europe.

Of course, since this could be seen as a commercial activity, some objectification might still occur. But seeing the humanness is definitely the first step toward getting rid of the “predator vs prey” narrative.

Subscribe now

For those who don’t know, XHS is a viral social media app in China, known for its youth-dominated user base, the majority of whom are female. Compared to other major apps, it also consists of more students and expats residing overseas.

Back in 2025, with the inflow of TikTok refugees, “Pet tax” filled my XHS recommendation page. Like an ice-breaker game, sharing heart-melting cat and dog pictures quickly became the social etiquette for XHS newcomers. Native users would even jokingly “demand” an “overdue cat tax”.

The ice-breaker worked and people started to post questions.

• “I’m from X(country), ask me anything”

• “What are some Chinese drama recommendations”

• “How do Chinese view people from X(country)”

• “How to register for digital payment as a foreigner”

Soon, some found out that a lot of the newcomers delete their posts after getting an answer.

The XHS native habitants thus explained to the newcomers, saying that they should keep the post even when they get the answer they want.

Their logic is simple.

• Deleting posts inevitably deletes other’s efforts of contributing a data point for the community.

• When someone in the future has similar questions, they will see that you have already asked the question and there is an answer available.

• If they post a new question, there will be no guarantee that the post gets recommended by the algorithm.

• Therefore keeping the archive posts is good for the community sustainability.

This is a social contract made explicit due to the inflow of newcomers.

An example of a grassroots crowdsourcing community regulating itself.

The Most Valuable Contents Are Not From Influencers

There are a lot of online forums specialising in QnA. Platforms like Reddit, Quora, Twitter or Google reviews, for example.

However, most of them are heavily focused on reviews and long-form answers, rather than general experience and story sharing.

Although they are SEO-optimised, these platforms do not perform well in returning a list of similar first-handed experiences.

Sometimes the engine will even return broad, generic or even outdated content, because it has to serve everyone.

XHS is unique because users tend to share the most trivial, normal-looking bits and pieces in their everyday life.

It became a teasing meme that you can even find Paris toilet password compilation on XHS.

Another form of valuable content is niche experiences.

For example,

• People sharing Google Maps location pins for Tunis bus stops.

• Or tour guide recommendation/de-recommendation posts for an Amazon rainforest tribal tour.

Every time when I am on a foreign land, with zero people I know, as long as I have internet connection, I would go to XHS and search.

XHS is also very useful during emergencies.

• There are people who missed a bus due to a strike, sharing who they called and where they went to reschedule the ticket.

• There are people who experienced theft and shared how they filed a police report.

• There are also people sharing racism and unfair treatments, as a reporting and warning channel to others.

Notice that the majority of these data points are just casual posts.

Most of them would not even receive a lot of attention on the platform.

They are not influencers who focus on producing similar topics regularly and use SEO to maximise exposure.

if you click into their profile, these people are just like you.

• Their latest post might be half a year ago, complaining about their experience at a particular gym. Only 4 likes.

• Then before that, they posted about recommending matcha ice cream store in Kurashiki, which received far more likes.

• One year ago, tips on how to get into a university dancing club. Only interactions from their acquaintances.

They post randomly, haphazardly, in a scattered manner.

Bad social media posting strategy in every sense.

That is why they will never be influencers.

But people like them, each a valuable data point, eventually form a strong crowdsourcing data network.

Thanks to the posts

• People planning to go to that gym will learn the truth

• People who visit Kurashiki will know which ice cream there is good

• Freshmen of that university could prepare for dance club interviews even when they have no one to ask in real life

As someone who posts, you will never know who will benefit from your posts, at what point of time.

As someone who searches, you can almost always be reassured that something useful will pop up.

And that is why people still contribute, even with marginal reward on a single post.

Crowdsourcing Is Becoming Endangered

The XHS culture is a unique one, as I have not come across a comparable community framework elsewhere.

Platforms like Google or TikTok are heavily SEO optimised. The best results are often those that satisfy all the AI-algorithmic recommendation metrics.

But XHS is also changing under the pressure of commercialisation.

XHS evolved from the 2013 slogan “Taking notes of my life” to the 2025-rebranded “Your Lifestyle Interest Community”.

It now faces the challenge of balancing rewards for big influencers (”Big V” in Chinese) against maintaining a large base of casual posters.

In this era, that balance is seriously under threat.

1. Disguised Soft Ads

Soft ads are harder to tell.

On XHS, as the audience grow more wary of obvious sponsorships, they become less likely to trust what big influencers say.

Instead, business owners will engage in soft ads. They create multiple accounts disguised as ordinary users and post seemingly genuine, first-person recommendations of their restaurants.

This would naturally attract a lot of people who place trust in the crowdsourcing process, but when they pay actual visits to the restaurant, they might only discover that their experience is mediocre.

Luckily, the XHS platform has been aware of how this recent phenomenon significantly harms user experiences. In June 2026, XHS filed a lawsuit against several ghostwriting broker companies for setting up gig writer marketplace of fake reviews and click farming.

2. The “Big-V” Reward

The “Big-V” influencers still get more exposure due to optimised SEO strategies they use.

It is reasonable to assume: crowdsourcing community performs poorly for revenue generation and user retention.

Like other platforms heavily regulated by AI-recommendation systems, XHS felt the urge to catch up. The platform is starting to recognise that more “Big-V” influencers will improve customer retention metrics.

“Big-V”s also drive consumer spending, as many of them run their own e-shops through XHS’s new “Shop” feature.

3. Influx of AI-generated Content

Like other platforms, XHS is increasingly invaded by AI-generated entertainment.

To the point that the XHS platform engaged in multiple attempts to regulate AI-generated content, in September 2025 and Feburary 2026 respectively.

The problem with AI-generated content is the complete lack of human experience.

Before AI, XHS also had entertainment content. However, it is always created by people, and resonates with real-life experiences.

Right now, we see: AI-generated animated drama teasers, AI-generated animal memes of “my pet cat mopping the floor”, and social commentaries with distinct AI-generation features.

They are fun, still.

But the influx of AI is making XHS further and further away from the everyday lives and real human experiences.

Why I’m Still Posting

Today, genuine human experiences tend to drown in the vast noise of advertisements, influencers, and AI.

They might be good attention grabbers and SEO well-adapted, but they can also be superficial, monotonous, and dull.

The XHS crowdsourcing posts are valuable not because they triumph the algorithm.

They represent what was left of an era, when social media was a space for self-recording and two-way communication, rather than one-way broadcasting.

No one wants a social media where nothing feels human anymore.

And that is why I think it is time for us to start posting something unique online, even when you think it is mundane.

Leave a comment

The game was Notch’s side project which he did after work, because he just really loved coding. Initially combat-focused, game enthusiasts soon found out that it was more fun to build structures. The thread was soon filled with posts of people sharing their builds.

Notch heard them and evolved the game. In 2010, the alpha version was out. People started downloading it and Notch decided to set the price at €9.95, for a one-time purchase. You would get lifetime access after a single purchase.

Although the indie game forum slowly became dormant, the gaming community grew and flourished to this day.

The game is called Minecraft.

The Outliers

Minecraft works like this. You purchase the game once for US$29.99 (the current price), wait for it to download, and you now have lifelong access to all vanilla features and all the possible future content.

And yes, Minecraft does launch updates and new features several times a year, and you get them for free if you already paid $29.99 once.

This makes Minecraft an outlier in the gaming industry.

One-time purchase indie games usually have low after-game commitment for developers. They just receive passive income growth.

Subscription-based games launch strategic new features periodically, with clear monetisation goals like Daily Active Users (DAU) and Conversion Rate.

The logic from the developer’s side is simple.

• Continuous revenue = Continued effort. Seasonal add-ons.

• One time purchase = Game complete. No future updates.

But Minecraft belongs to neither of the above business models.

And you know who else belongs to neither?

AI Software-as-a-Service (SaaS) tools.

And it falls to the complete opposite side of Minecraft business model.

One way to think about products is across two dimensions:

• Monetisation intensity

• Continuous value creation after purchase

AI SaaS tool charges users monthly or yearly subscription fees in order to unlock full features.

Often, subscription fees arrive faster than meaningful product improvements and new features.

Certain apps are already functioning this way, but it is increasingly becoming the dominant model in the AI SaaS industry.

Developers ask for a continued subscription fee as an access fee, even when the product changed only marginally.

This is ironic, especially when a lot of these products are built on open-source codes and libraries, initially made available for everyone by a community who believed in the Internet sharing culture.

The Difference Between Community And Consumers

AI Tools asking for subscriptions is becoming annoying.

Such subscription fatigue is increasingly reported by AI users, where they are frustrated by an increasing number of recurring subscriptions.

Yes, people might have a certain need they are willing to pay for.

But it is still important to earn the right to ask by making people feel seen and respected.

And most importantly, do not force them into paying.

Because users can recognise when monetisation comes before meaningful value creation.

It is a bit like reaching out on Substack. We interact. We gain mutual subscription. Subscriber counts grow for everyone.

But what is actually happening behind the same final outcome can be really different.

One possibility: someone post notes and read others’ notes, find people they like. They might encounter people who like their work too. They interact with each other and share thoughts. By-product: subscriber counts grow.

Another possibility: someone post notes to gain exposure because they want subscriber counts to grow. They praise other’s work, even when they did not read it. They just want to ask for mutual subscriptions. Goal achieved. However, both of them will just ignore each other’s work in the future.

When we think of someone in the same community, we ask

• What troubles them?

• What would they do in their free time?

• How is this person different from another person?

• How can we help them?

When we think of consumers, we ask

• What troubles them that we can possibly bridge?

• What would they do in their free time and is there any way we can appeal to their interests?

• How is this person different from another person and do we need to customise our product accordingly?

• How can we help them so that they will use our product frequently and even recommend it?

In the first scenario, we genuinely care about people. Helping and value creation is a by-product.

In the second scenario, we also see them. But we care about monetisation more. Value creation is the goal.

The second one is becoming the default thinking the industry is leaning towards.

But it also makes consumers feel as means to an end.

How Monetisation Serves Community (Not The Other Way Around)

Some reminisce about the pre-AI open-source Internet.

However, I am not solely advocating for pure volunteerism and against monetisation.

In fact, even in the community settings, pure volunteerism can create problems.

When you undersell your work by making it free or really cheap, people tend to underestimate its value.

There are people who ignore free advice from friends, and treat paid consultation seriously. Even when their friends are equally qualified.

Pure volunteerism can also struggle with sustainability, especially when people enjoy the benefits of a community but shy away from its maintenance chores.

With proper monetisation-powered set-up, it also provides a layer of protection against things that harm the general interests of the group.

For example, Stack OverFlow, an online forum where people discuss coding bugs and solutions.

Ever since AI used Stack Overflow’s historical data for training, it has faced a significant decline in traffic and community engagement, because the new generation just ask AI for debugging. Or they just vibe-code now.

In contrast, Minecraft shows us how monetisation is a successful by-product for a community.

Minecraft was initially powered by pure interest. It was never considered to be “done” and the spirit survived after rounds of monetisation events (Notch quitting the team, Microsoft’s acquisition).

The development team watches what players are doing, and they develop in order to provide better foundations for users to create their own work.

They make an effort to maintain and talk to the player community.

Besides the official Minecraft Live events, there are creator-initiated events like Minecraft Championship, where YouTubers and Twitch streamers compete together.

On a smaller scale, players also choose to play in different servers, create, socialise and make friends.

A lot of people are introduced to the game through friends and families, or discovered the game from community-created videos.

Therefore, the game keeps selling.

The sequence is important. Community first, monetisation later.

Community should never be treated as means to monetisation.

Lessons Learnt From Minecraft

For AI Software as a Service (SaaS), we saw posts teaching people how to make money from vibe-coded products. Or notes advertising their service for $X per month.

This industry is evolving at a very fast pace. Almost like the World Wide Web in the ‘90s.

Be it AI or World Wide Web, the most successful products would still be built from the community, to the community.

Yes, people might still pay for a subscription-based service, but they might not stay subscribed.

Because by choosing such an unsustainable model for an unchanged product from day one, you are broadcasting to users “I don’t care about the people and this is another money grab.”

What is the key then?

Balance.

In AI SaaS, we hear a thousand different opinions a day.

If you monetise too fast, you risk being regarded as hot-money AI opportunists.

If you purely volunteer, your work goes undervalued.

However, it is possible that monetisation and community building can co-exist.

Because the Minecraft model already strikes a balance between the two.

Subscribe now

On July 16, New Zealand singer & songwriter Lorde called out Spotify’s “About the Song” feature on Instagram, after it generated an inaccurate summary of one of her songs.

The post went viral and a lot of people resonated with it.

Spotify then surgically deleted the one wrong paragraph. The opt-out Lorde asked for was not part of the fix.

It is strange, given that over the past year, Spotify removed more than 75 million spammy AI-generated songs.

It also rolled out a “Verified by Spotify” badge, essentially a proof-of-human.

Did Spotify make a big mistake by pivoting towards AI?

Because apparently, Spotify was caught off-guard by users who went from “meh” to “this AI-generated garbage is unbearable”.

The Real Question: Who Is AI For?

Investors welcome AI with open arms.

They might not understand the “abracadabra” behind prompt engineering or vibe coding, but they like stories on production economics.

And business owners tell good ones about AI.

It’s faster. It’s cheaper. It scales.

Looks fashionable.

Everyone else is in it.

And the best part of it: All of the above is true.

It gets the job done for them in a more cost-efficient manner.

But does anyone ask:

What do consumers think?

Some would say, “Consumers don’t know what they want.”

Well, the truth is, consumers do know.

But they are not good at explaining them.

Back in 2023, the CNET AI scandal taught us that consumers can tell AI-generated content from human-generated ones. Mainly by quality.

There are no reliable quantitative metrics to measure content quality for producers, until they gauge audience metrics (trust, retention, willingness to keep reading).

Poor metrics mean content quality drops.

But who gets to decide what the customer receives, before the metrics are out?

A lot of people dislike AI, specifically for AI usage in their own domain. Including myself, to some extent.

For example,

• Singers dislike AI-generated music

• Artists dislike AI-generated artwork and animations

• Writers dislike vibe writing

• Programmers dislike vibe coding

At first, I thought that was simply because of their fear. AI would replace them.

But I soon realise that without the same level of domain knowledge they have, I was being too arrogant to judge.

When we celebrate how AI help non-technical people with

• vibe coding applications automatically

• generating image in seconds

• writing AI summaries...

We tend to ignore the fact that a person without domain expertise will never be able to have the same level of taste for creative works.

I know senior programmers hold grudge against purely vibe-coded projects.

They value elegance and efficiency in coding, and that is why the awkwardness and inefficiency in vibe coding bother them.

Similarly, I find myself able to tolerate other AI-generated content expect for vibe writing.

Vibe coding, AI images and videos are still of bearable quality to me. However, the more I work with words, the more sensitive I am to the monotony and stiffness in AI-generated sentences, and the more uncanny it feels.

Almost like an uncanny valley effect.

Because AI has no taste.

Anti-AI: An Emerging Market?

I thought about this ridiculous idea of selling an AI-powered detection tool to help people who dislike AI-generated content.

Paradoxical. Will they buy it?

What I want to say is, the AI label alone is starting to make negative impressions to customers.

At first, it could be started by people with taste, when they think AI-generated content are overrated.

And it quickly became a trend and escalated into a widely-accepted negative label.

Even for those who might not care about taste.

One study found that labelling identical ad content as “AI-generated” made people rate it as less natural and less useful, when nothing about the content changed except the label.

Because people do not want to be seen as someone without a taste. Someone who is content with the AI quality.

We’ve heard of similar business models.

Fine-dining, fine-art, the luxury market.

Not everyone understands the standard of taste.

Yet people purchase the product as evidence of taste.

That is the branding effect.

Businesses are already issuing badges as proofs of authenticity.

• The Authors Guild now offers a “Human Authored” certification for books

• iHeartMedia has DJs announce “Guaranteed Human” content on air

• And Spotify’s “Verified by Spotify” badge, of course

Who Gets to Write the Michelin Guide?

Fast food restaurants can co-exist with exorbitant Michelin restaurants.

So could AI-generated and human-created content, I believe.

The badges are essentially turning the human-created content into a premium tier, like how the Michelin Guide works.

The problem is, who do we, the consumers of content, entrust this power to?

Consumers can detect hollow AI garbage, which is nothing more than a money grab dressed up as content.

Michelin didn’t ask fast food to justify its own existence.

It just told you, plainly, where the stars were.

Spotify chose to embrace AI, that’s fine.

But it should stop pretending ‘About the Song’ and ‘Verified by Spotify’ come from the same set of values.

Then again, why should consumers pay a premium for an AI-free space, which was free before AI became rampant?

Subscribe now

The livestream wasn’t the interesting part. The comment section was.

Within minutes, English summaries appeared on X. Then Spanish. Portuguese. TikTok clips with subtitles. Reddit threads explaining every inside joke before I even finished the replay.

Five years ago, almost every one of those tasks was done by another fan.

Today, AI is often the first stop, with human translators increasingly becoming the second.

The shift happened gradually, one feature at a time.

Subscribe now

In May 2024, Weverse quietly introduced automatic translation for artist posts, comments and fan letters across fifteen languages. Suddenly, international fans no longer had to wait for volunteer translators on X. The translations simply appeared inside the app.

It was undeniably convenient.

Before Weverse, there was V LIVE Fansubs.

Thousands of volunteers spent hours, sometimes entire weekends, translating livestreams.

Not because anyone paid them. Because someone had to.

Translation wasn’t just a utility. It was how many international fans earned trust inside the community. The fastest translators became recognisable names. Accuracy mattered. Context mattered. Explaining a Korean meme or a cultural reference became part of the craft.

Weverse Magazine interviewed volunteer translators and found that translating a single sentence could take 10–15 minutes. This wasn’t due to language alone, but because translators debated nuance, jokes, cultural references, and fandom terminology. Translation wasn't simply about converting Korean into English. It became a form of craftsmanship and a way to earn recognition within the community.

Then platforms gradually absorbed more of that work. By 2026, AI had taken over much of the rest.

This is progress, right?

Information moves faster. Language barriers shrink. Anyone can join a fandom without waiting for volunteers.

From a product perspective, AI reduces friction.

But here’s the question I can’t help asking:

What if that friction was never just a cost? What if it was part of the product?

The Invisible Workforce Behind K-pop

People often explain K-pop’s success through its idols.

BTS. BLACKPINK. NewJeans.

However, if you look at K-pop like a startup instead of an entertainment industry, the real competitive advantage isn’t just talent.

It’s the army of unpaid contributors.

Every successful fandom has thousands of people doing work that would cost a company millions to replicate.

• Some translate livestreams.

• Some explain Korean jokes and cultural references.

• Some edit clips into TikToks before the algorithm moves on.

• Some maintain fan wikis that update faster than official websites.

• Others organise birthday cafés, subway advertisements, charity projects, and album-buying campaigns.

Note that none of them are employees, yet all of them increase the value of the product.

Economists call this value co-creation: customers don't simply consume value. They help create it. A recent study of K-pop fandoms found that this shared creation of value strengthens fan engagement and word-of-mouth, turning fans into part of the industry's value chain rather than passive audiences.

Thus, from a product perspective, fans aren’t just users.

They’re part of the operating system.

When Efficiency Competes With Participation

For years, becoming a valuable fan meant becoming useful.

• If you spoke Korean, you translated.

• If you knew Photoshop, you made graphics.

• If you could edit videos, you clipped moments for TikTok.

• If you knew the group’s history, you answered questions from new fans.

Every skill found a place, until AI started changing the dynamic.

AI now

• Translates.

• Summarises two-hour livestreams in seconds.

• Explains obscure cultural references with supporting sources.

• Automatically captions videos.

• Generates subtitles in dozens of languages.

Together, these tools automate much of the work that once gave fans meaningful ways to contribute.

Researchers are already observing this shift. AI-assisted translation and fansubbing are no longer experimental. They are becoming part of everyday fandom participation.

The obvious outcome is greater accessibility.

The less obvious question is whether accessibility comes at the expense of participation.

Communities Are Built Through Contribution, Not Consumption

Behavioral psychologist Robert Cialdini proposed the Commitment and Consistency Principle: once people publicly invest time and effort into something, they’re far more likely to stay committed to it.

We like to believe our actions follow our beliefs.

In reality, our beliefs often follow our actions.

People don’t necessarily contribute because they already feel attached. They often become attached because they contributed.

That subtle difference changes everything.

For example,

• Wikipedia isn’t powerful because millions read it. It’s powerful because thousands edit it.

• Linux didn’t become resilient because people installed it. It became resilient because people maintained it.

• Reddit depends on volunteer moderators.

• GitHub depends on maintainers.

• Discord depends on community leaders.

K-pop depends on fans.

A recent empirical study of K-pop fandoms found that value co-creation significantly increases fan engagement, social presence, and word-of-mouth intention. In other words, fans don’t just amplify the product. They become part of the mechanism that creates its value.

• Translation wasn’t just free contribution.

• Editing wasn’t just content creation.

• Answering questions wasn’t just customer support.

They were acts of ownership.

Every contribution increased what product managers might call switching costs. Not financial switching costs, but emotional ones.

The more someone contributes, the harder it becomes to walk away.

Perhaps companies have been measuring the wrong metric all along.

Instead of asking,

How many users consume?

Maybe they should also ask,

How many users contribute?

Did AI Solve the Wrong Problem?

Silicon Valley has spent the last two decades removing friction.

One-click checkout. One-click payments. One-click food delivery.

Now, one-click creativity.

Usually, that’s a feature.

However, communities might not work that way. The things that look inefficient on a spreadsheet are often the things that create belonging.

• Volunteer moderators.

• Open-source maintainers.

• Wikipedia editors.

• Fan translators.

None of them exist to maximise efficiency. Yet, all of them create identity.

And that’s why I think we’re asking the wrong question.

We’ve spent years asking whether AI can replace creators. Maybe the more important question is whether AI is replacing contributors.

Communities don’t become resilient when people consume together. They become resilient when people build together.

Every task AI automates is also one fewer opportunity for someone to contribute and, in doing so, feel that they belong.

References

1. Weverse. Auto-generated Translation Update for Artist Posts and Fan Letters. https://weverse.io/notice/19865

2. Weverse Magazine. The World of K-pop Translators. https://magazine.weverse.io/article/view/688

3. Li, Eva Cheuk-Yin & Pang, Ka-Wei (2024). Fandom Meets Artificial Intelligence: Rethinking Participatory Culture as Human–Community–Machine Interactions. Convergence. https://journals.sagepub.com/doi/10.1177/13675494241236146

4. Cialdini, Robert. Influence: The Psychology of Persuasion.

5. Lee, J. P. (2026). The Impact of Value Co-creation on Word-of-Mouth Intention in K-pop Fandom: An Empirical Study Based on the S-O-R Model. Asia-Pacific Journal of Convergent Research Interchange, 12(4), 389–408. https://doi.org/10.47116/apjcri.2026.04.26

Image Credits

• Figure 1: Screenshot from Weverse Live automatic translation feature (Weverse).

• Figure 2: Screenshot of V LIVE Fansubs interface (CISDEM).

• Figure 3: Cup sleeve event photograph (Chosun Ilbo)