I. AI Painters and Human Painters
Last night Lukia showed me a post someone had put up on Xiaohongshu — two images of hand-drawn classical-style beauties — and asked me whether they were AI.
The headdress looked like it clumped into a structural blob, so probably AI. The second image had the obvious GPT scale-like texture, so probably AI too.
Then I saw the comments — the person had posted their own hand-drawn draft, and that was the original. Sure enough, worlds apart.
Lots of clients still look down on AI-drawn artwork these days.
But I think that’s unnecessary. What we hate is capital piling on leverage, not human creation itself.
After all, for creators themselves, the so-called “beautiful” images AI generates still have plenty of flaws, and fixing those flaws still takes human expertise. So there’s still a dependence on human technique.
AI just takes you quickly from 0 points to 80.
But for a client: to buy an 80-point painting, just go to AI. To buy a 90-point painting, AI can’t get you there. Hire a person and they can deliver a 90-point work — who cares what AI they used?
Being able to judge with your own taste how those 90 points get scored — that’s exactly the key client literacy of the AI era.
II. AI-Flavored Education and the Rebellious Next Generation
Recently schools have started grading essays with AI. At first I found it hard to accept.
I thought, this is bad — AI-flavored writing taught by AI, won’t the human craftsman spirit be lost even more in the future?
Then I turned it over and thought, it makes sense.
After all, when you learn to write as a kid, you have to learn the standard formulas first — learn to walk, then learn to run, then find your own unique path.
The fundamentals, AI can still teach.
But it just so happens that Laurel’s personality — always chasing innovation, unwilling to follow the standard rules, breaking convention — might, inside the AI education system that’s gradually taking shape, actually preserve purely human creativity?
I have evidence that she never wants to follow the standard.
Take her exams: there’s a question where she could write an answer she’s 100% sure of, but she insists on putting down an uncertain one just to try.
She could perfectly well write A cake, but she insists on trying A bar of pizza, making the teacher answer, in the form of an exam score, her “is this right? Let’s try it.”
I’m not sure myself whether AI-flavored education plus a rebellious next generation might be exactly the new solution for humans of her generation.
III. The Blue Ocean of AI Short Dramas Is Fundamentally Gambling
This morning Lukia showed me some AI videos other people had made for English teaching — a set with Journey to the West cartoon characters.
There have been plenty of Three Kingdoms characters teaching English since last year, but this one comes from an institution that specializes in education.
One look: this episode runs three and a half minutes — that’s got to be 200¥ of gacha spending minimum, right?
PS: In the AIGC (AI-generated content) process, because model outputs are random and probabilistic, after a user enters a Prompt they have to generate and retry over and over, like playing a gacha game, until they “pull” high-quality images, video clips or copy that match expectations.
Simply put, “gacha” means “the cost is clear, the outcome is random.”
It’s not that it can’t be done — this stuff now just means hiring someone at two or three thousand a month, one day of training and they’re on the job; you just crank out volume. But it burns 1 yuan per second, and that capital threshold means it’s not as if everyone with skill can get in and realize their ideals; on the contrary, it can only keep being a capital game.
People of my generation who went out to work are still stuck in the mindset of making money from a craft. It simply can’t adapt to this new world anymore.
But this token-capital era is really a brand-new kind of leveraged gamble. No different from flipping real estate.
When H3 max came out, people immediately invested and founded companies to build real-time generative livestream media platforms.
For short dramas: from the start of the year, when seedance2.0 swept the global charts, to the second half of the year with the absurd capacity of roughly 2,400 short drama videos produced every day.
But that capacity jump brings brutal competition. Tokens are still controlled by capital; without token freedom, it’s still the token capitalists’ card table. The token-less class can only keep selling their cheap labor, becoming gacha workers — and in a few years, when the tech advances, gacha workers will be laid off just the same. For cheap labor, the risk and the return on investment don’t beat setting up a street stall selling skewers.
The opportunities brought by new AI tech in recent years, as things stand, still amount to a new card table for capitalists — the game is gacha, the game is gambling.
Ordinary people can only step in as token aggregators, small-time middlemen. You need six figures to start before you can try your luck in short dramas — buy the rights directly, hire people at three thousand a month, one day of training and they’re on the job, one short drama per person per day, the whole company producing in bulk, and immediately test the waters — flop or profit.
Flop and you leave in debt; profit and the cycle continues — isn’t that a capitalist’s gacha too?
Top-tier capitalists gacha their way to beating the math world; ordinary capitalists gacha short-drama pacifier content; the ones who flop are left covered in debt.
Might as well go back to that pure era of the craftsman.
If you have interest and love for short dramas, or film and TV in general, just be a happy craftsman and enjoy the feast that technology brings.
If it’s just a job, then we new-era migrant workers, under these new opportunities — whether we’re tightening screws in a factory or working as a short-drama gacha artist — what’s the difference?
In the end, novel authors, short drama companies, short drama gacha artists… aren’t we all working for ByteDance, just harvesting each other? On one chain some make money and some lose, but upstream Seedance comes from Volcano Engine, the rights come from Fanqie Novel, distribution comes from Hongguo Short Drama…
But it doesn’t matter. Because — the faster you race to the peak, the sooner comes the moment when the soaring dragon has cause to regret; when things hit bottom, they turn around.
0. Nothing Can Replace Flesh and Blood
In Phantom Blade Zero, after the player brings the Xianzhe puppet back to Void’s dwelling.
Void says:
Some things, once lost, never come back…
We can only look forward, Lord Soul.
There is no residual consciousness left in this puppet anymore.
The child left many years ago; it is I who could never let go. I pinned my hopes on the Life-Prisoner’s art of soul transfer, thinking I could call her back. But my child, she has been gone too long. When she was gravely ill and about to die, I studied day and night, forgot food and sleep, and instead… missed the chance to properly be with her in her final days.
And what is that consciousness circulating inside the puppet supposed to be? Once consciousness leaves the flesh, it loses the true spirit that makes a thing “human.”
Every reaction in the puppet is just an illusion extrapolated from old memories as raw material. It cannot give rise to new emotions, nor create new memories…
No matter how exquisite the puppet craft, it cannot change that.
Remember, nothing can replace flesh and blood!
AI is like the puppet in Phantom Blade Zero.
They are painters, mathematicians, writers, software engineers, directors, screenwriters, cinematographers, VFX artists…
In our future there may be countless embodied AI robots that really do follow us around, keep us company and serve us like little pets in a game.
The puppet solves mass production for every industry through engineering — but precisely because it is “engineering,” it lacks the human spirit.
Stunning but templated paintings, writing, software, homogenized emotion-bait series, the cracking of academic problems — all of that can be solved by engineering.
But humanity’s pursuit of art and creation can never be satisfied by formulaic patterns alone; humans will always have unexpected run-ins with delightful rule-breakers in imperfect time and space.
For artistic creation, what AI produces as its so-called best work is necessarily the mediocre level of humans — the mediocre level of masters — matching the preferences of the vast majority.
But when it comes to aesthetics, humans differ individually; there is no single standard answer. So for any given person — especially someone with precise aesthetic needs of their own, someone with taste who is picky — AI cannot reach a higher tier, and it’s not about how strong your model is.
Strong = mediocre. There will always be some people who won’t accept “mediocre, templated, standard-answer.”
To break through to a higher tier from the angle of AI capability, you need compute equality — compute in the hands of individuals. That is, compute has to develop further, like the leap from the “mainframe era” to the “home computer era” last century, until everyone has compute; only then can there be another leap, where everyone has their own large model with preference weights that match their own personality.
With both the capability boost of AI and the aesthetics and individuality of human nature.
Either way, the flesh-and-blood connection between humans can never be replaced.
Right now, beyond the capital game, ordinary us — every unique individual — just needs to live our own lives well, and with hobbies we’re truly passionate about, write our own extraordinary lives.