Last August, He Tongxue (何同学) put out a video, We Photographed a Landscape 7,500 Light-Years Away. He’d rigged up a mobile observatory and driven it all the way from Hangzhou to Lenghu in Qinghai to shoot nebulae. It pulled a few million views on Bilibili, and the description of the English version even had a purchase link. The telescope in the video is the Seestar S30, a smart telescope from ZWO of Suzhou, $399, roughly two thousand yuan in China.
The astronomy groups I’m in all blew up at the time. People slammed the video as sloppy, slammed the S30 as a toy, slammed the idea that this could even count as astrophotography. I slammed it harder than anyone.
This kind of anger needs a bit of background to make sense. What does traditional deep-sky imaging actually involve? An entry-level rig — an equatorial mount, a cooled camera, a guide scope — starts at three thousand dollars, and a serious setup easily runs into five figures. For a single photograph, I’ll watch a whole week of weather forecasts, pick a weekend around the new moon, drive a few hundred kilometers into a dark-sky reserve, pull an all-nighter to stack dozens of hours of exposure, and then soak for days in PixInsight doing the post-processing. Everyone I could really talk to in those groups had walked exactly that path.
Then a two-thousand-yuan little cylinder, propped up on a rooftop balcony, two taps on a phone, and out comes an image. By my standards the image is a hopeless blurry mess, but its owner posts it to their feed and racks up no fewer likes than I do.
By the third day of trashing it, I scrolled past a Weibo post. A single question in it stopped me cold: what exactly was I trashing?
The ready-made answer is that I was trashing him for being sloppy. The criticism itself isn’t wrong — He really doesn’t know much astronomy: the celestial motion in his animation is wrong, his imaging method is off, and the aesthetics of the final shot are, let’s say, hard to defend. Later I posted this reflection to the group, and my friends reacted the same way: what we’re slamming is clearly the sloppiness, he was in the wrong first, so isn’t the anger justified?
The criticism is right, but the anger doesn’t add up. How many sloppy videos does Bilibili gain every single day? Nonsense in the astronomy section has never stopped, and none of us ever lost our tempers over it. Same error, so why should this one video have an entire group ranting for three days straight? The size of the error can’t explain the selection; only the size of the threat can.
Put bluntly, “sloppy” was just the excuse I grabbed on the way past. What actually cut to the bone was something else: the astronomy knowledge I’d built up over years, my command of complex gear and post-processing workflows — these are my credentials for being distinct from a beginner, my capital for showing off in the group. Now a clueless video, promoting a telescope we regard as a toy, pulls in more attention than every serious hobbyist combined; a crowd of people who’ve never touched an equatorial mount bypasses everything I’m good at and just walks off with the identity of “person who photographs the stars.” My craft is depreciating, and pretty soon I won’t even get to say what counts as a good photo.
Economics has a name for this: rent — the premium you collect off scarcity. When I trashed He Tongxue, what I was really trashing was the future in which my scarcity vanishes entirely.
A few days ago I read an article, The Proletariat of Judgment from a WeChat account called Budongjing (“Doesn’t Get Economics”). It was extremely sharp, and reading it left me feeling exposed.
Its argument runs roughly like this. Naval Ravikant once said that in an age of infinite leverage, judgment is the most crucial skill. Because leverage multiplies everything, the steering wheel decides whether the final direction is right or wrong.
But AI, of all things, is eroding human judgment. An investor talked on X about something he’d witnessed: he had lunch with a 22-year-old Stanford graduate, and mid-conversation the guy would seize up, having to grope for even the most basic words, unable to get them out for the longest time, like a brain buffering — because he’d grown used to letting ChatGPT complete his train of thought for him.
An MIT EEG study also measured that people who wrote with ChatGPT showed weaker connectivity in specific brain regions than those writing purely on their own, and 83% of them couldn’t even remember what they had just written after finishing. Research from Cornell likewise found that AI’s writing suggestions are pulling people from different countries and cultural backgrounds toward a single Western register, where the favorite food becomes pizza and sushi and the favorite holiday becomes Christmas.
So the article reaches its conclusion: knowledge workers are outsourcing their thinking and becoming the proletariat of judgment. Incidentally, most of the empirical material in this article comes from New Yorker columnist Kyle Chayka’s piece A.I. Is Homogenizing Our Thoughts; if you want the English version of the argument, just read that.
In truth, these studies are of uneven quality. The MIT one, for instance, had only 54 subjects and is a non-peer-reviewed preprint, and the first author, Kosmyna, herself objects to the crude reading that “AI makes people dumber.” But even if all these conclusions hold, what I actually want to talk about is unaffected.
The argument even patches itself in advance, precisely to fend off the cyclical view. Someone might bring up the calculator: didn’t people rail against it the same way back then, and how did that turn out? But following the article’s logic, the rebuttal is ready-made: the calculator only outsourced arithmetic, while AI outsources judgment itself, and judgment is exactly the foundation you use to decide what to outsource. Lose it, and you can’t even judge what you’ve lost.
Once I’d calmed down after reading, I realized that the knottiness in it, the disdain for the mediocre, the heartache of worrying to pieces over ordinary people’s futures — it was almost exactly the tone I’d taken when I was trashing He Tongxue in the astronomy group.
It’s really just a far more elevated, and far more sincere, live broadcast of trashing He Tongxue. Who writes and shares it? Knowledge workers who make a living from reading, writing, and analysis — the crowd who play the craft of language most fluently. And what AI is making commonplace happens to be exactly that craft.
Look back, and every version of this play is on record.
Around 370 BC, in the Phaedrus, Plato used the mouth of an Egyptian king to reject the invention of writing: once people get used to relying on the written word, they stop exercising their memory; what you’ve invented is not the elixir of memory but of reminding, and your students will have only the shell of wisdom, not its substance. Swap “writing” for “ChatGPT” and this passage would still rack up a hundred thousand views today. The greatest irony is that this passage bitterly cursing the limitations of writing only survives to us because Plato wrote it down in writing.
In the 1970s the calculator entered the classroom, and American education argued about it for twenty years. In a 1979 survey, 84% of teachers wanted to use calculators in class, but only 3% of schools were willing to provide the devices. By 1986, textbook author John Saxon was still leading a group of teachers to a math education conference in Washington to protest. The slogans sound extremely familiar: it destroys the fundamentals, it ruins the next generation.
Further along, print media slammed Weibo for fragmenting things, television slammed short video as vulgar, illustrators slammed AI art as a mashup, translators slammed machine translation for wrecking texts. The recipe every round is identical: a new tool beats a craft down to cabbage prices, and the people who had been collecting rent on that craft step forward to protest in a moral key. “Bad money drives out good” is the standard boilerplate of this kind of protest. The accusations are often true — short video really is fragmented, machine translation really is crude, He Tongxue really is sloppy. But there’s no shortage of bad money in the world, and the only coins we ever actually move to drive out are the few that directly threaten our own good money.
My trashing of He Tongxue was just the latest re-staging of this play. And “the proletariat of judgment” is the most philosophically colored monologue in this protest.
Of course, just saying “history has gotten it wrong many times” isn’t a real rebuttal. That vaccine-patch from earlier still stands: this time what’s being outsourced is judgment itself, so the nature of it has changed. This challenge deserves a head-on answer.
There’s actually a simple test: swap the “artificial intelligence” in the panic arguments for a shrewd, capable assistant, and see whether the words still sound frightening.
He’s grown used to letting his assistant sort out his thinking. She hands her whole schedule to her secretary to arrange. He defers all his legal decisions to his lawyer, all his health decisions to his doctor.
Who do these sentences describe? Executives, professors, senior officials — the crowd universally recognized as having the strongest, most central judgment. The CEO of a company worth hundreds of billions has outsourced the vast majority of decisions across a lifetime, but no one would call him the proletariat of judgment. Quite the opposite: when Naval talks about how judgment merely ten percent ahead deserves a salary of hundreds of millions a year, this is exactly the skill he means — knowing what to delegate to whom, and knowing how to check the results that come back, without having to do everything yourself.
Delegating decisions and losing judgment are two different things. Humanity has been outsourcing decisions through the social division of labor for thousands of years. The only new thing right now is the price: this network of on-demand advisors, which only a CEO used to be able to afford, now costs twenty dollars a month.
So the claim that “what AI outsources is judgment” actually pulls a switch: it blends the craft of execution together with the judgment of decision-making.
My own astronomy experience happens to pry these two things fully apart. Driving hundreds of kilometers, staring at guiding all night, soaking for days in PixInsight — these are craft, execution. But judging which region of the sky is most beautiful this season, which shot has bloated stars, which stretch in post-processing has been overdone — these are judgment.
What the smart telescope makes obsolete is the former; as for the latter, it hasn’t touched a single hair of it. The beginner holding the smart device still doesn’t know what to shoot or what makes a shot good. These questions of taste and direction have, if anything, become more valuable than before.
But when I was ranting away in the group, I forcibly bound the two things together and mourned them as one. Because only bound together does the “capital” I was losing sound big enough. Mistaking a well-honed craft for core judgment — that misjudgment is itself the product of the seat commanding the brain. Every profession is a spectrum stitched together from mechanical execution and subjective judgment; I’ve written about this principle specifically before. What I didn’t expect was that this time, in the middle of my own fit of fury, I caught myself red-handed doing exactly it.
Writing has the identical structure. Choosing words and shaping sentences, organizing paragraphs, spreading an idea out to fill three thousand words — these are craft; deciding what to write, spotting which argument doesn’t hold, recognizing from ten drafts which version truly has a soul — these are judgment. What ChatGPT knocks down is the cost of the former, and the former happens to be knowledge workers’ long-standing badge of identity.
That said, judgment does dull without practice. Look at AI’s proposal first and then decide, every single day, and the capacity for independent judgment really will degrade into “that looks familiar.” Both of these have empirical support, and I’ve written about them specifically before. But this is fundamentally a matter of personal usage discipline, like how lack of exercise is a modern ailment a gym can solve — no reason to ban the sale of cars over it.
Beyond the anxiety about judgment, those panic arguments carry another half of the mourning: homogenization, mediocre averages everywhere. This scene, in fact, was rehearsed for us by the auto industry a hundred years ago.
Rolls-Royce’s official customization page sells exactly infinite individuality: forty-four thousand ready-made paint colors are just the starting point, and the color team can even mix a finish on the spot to match a client’s lipstick or a pet’s fur. Tesla sits at the other extreme: worldwide, cars leave the factory in only the monotonous black, white, gray, blue, and red; want customization? No chance. But what that buys is an extremely low price and extremely fast delivery.
Cars on the street really are looking more and more alike, but blaming this sameness on the degeneration of mass taste is aiming at the wrong target. Standardization is precisely the underlying mechanism that pushed the automobile down from an aristocrat’s plaything to a mass consumer product. Offering only limited choices is exactly the precondition for ordinary people to drive a car for hundreds of thousands of yuan rather than millions. To enjoy affordability, the price is tolerating sameness.
This also incidentally resolves a timing hole in the analysis report The Age of Average: the converging cafés, the identical influencer faces, and the monotonous car-body colors it denounces all took shape before that report was published in March 2023, when AI had not yet had time to get involved at all. Homogenization is an old chronic ailment of industrialization and globalization; this time, it’s simply language and thought’s turn.
Besides, individuality hasn’t disappeared; it has merely become a premium option that requires extra payment. Find the Tesla too plain? You can wrap it in a color-change film, buy a customization kit from a German luxury brand, and, as long as the budget is there, keep adding all the way up to a Rolls-Royce. What the industrial revolution swapped out is the default setting: from a past where every car was a one-off but the vast majority simply couldn’t afford one, to a present where everyone can afford a standard unit and pursuing individuality means paying extra. For the vast majority of ordinary people, this trade is a flat-out net gain.
Of course, this net gain comes with a precondition: the dividend has to actually land in most people’s hands. When I was researching the AI short-drama industry, I saw a counterexample: technology cut production costs by eighty percent, but not a cent of the money saved reached the creators — instead it all went to platforms buying traffic and boosting feeds. Once old privileges dissolve, new excess profits tend to pool toward traffic and channels. Compared with fretting like Chicken Little that AI will make people dumber, this imbalance in how the gains are distributed actually deserves more vigilance. It’s just a shame that this kind of complex real-world game is very hard to write into an online blockbuster.
The standardized spread of intelligence is this same trade. Among studies of this type, there’s a detail that always gets selectively ignored: in a controlled experiment on creative thinking, researchers had subjects do divergent thinking with ChatGPT versus with creativity cards. The results showed that AI did not kill off individual creative richness; the convergence of creativity appeared only at the level of group statistics.
Translated into plain terms, AI did not make you mediocre; it merely distributed the same kind of smart equally to everyone. That same kind of smart, in the ears of the elite class, is a nightmare of mediocrity. But for most ordinary people, without ChatGPT, the floor of information they could reach was nothing but stale folk remedies past the third page of search results and pseudoscience flying everywhere; high-end professional consultants and private tutors were never for them. Averages everywhere: seen from above, the ceiling has caved in; but seen from below, the floor has risen. Which feeling is more real depends entirely on which layer you were standing on to begin with. That those panic-laden articles are, without exception, written by the elite class is no surprise at all. Because the people who hold the power of the written word and know how to write are, precisely, all standing on the upper layer that originally enjoyed the dividend.
Back to He Tongxue. Having admitted that my earlier anger actually came from my own position, I had to face a fundamental question anew: at its core, is popular science more about the “science” or the “popular”?
I used to think this wasn’t even up for discussion — obviously the science. Correct, rigorous, deep: this is the identity we serious hobbyists take pride in. But if you go do popular science under the command of that standard, what’s the actual effect? The threshold is terrifyingly high, the jargon is packed dense, and the equipment list takes you straight from “getting started” to “getting buried.” This process successfully filters out the vast majority of people who arrived with curiosity, leaving only the tiny minority who “passed the filter” to nod in mutual approval within a tight little circle. This is hardly spreading knowledge; it’s really using knowledge to filter for social kin.
If we try to flip the standard around — replace professional, rigorous, deep with connection, emotion, and resonance; replace the goal of “understand how an equatorial mount works by the end” with “feel a surge to go see the stars once by the end” — then along this dimension, He Tongxue’s video instantly leaps from failing to industry benchmark. Millions of people who rarely look up at the sky spent seven minutes and developed a flicker of longing for that vast deep space overhead. Even if only one percent of them bought a small device and set it up on the balcony, the size of the entire astrophotography hobbyist community would double in an instant. Although I still don’t much rate his post-processing to this day, his insight into the laws of mass communication is genuinely top-tier. When it comes to popularizing, a feel for the medium is the real skill.
This debate, in fact, rages just as fiercely in the English-speaking community. On Cloudy Nights, the well-known international astronomy forum, an entire section is arguing over whether all-in-one smart devices count as real astrophotography. The opposition finds this approach too simple, with no technical threshold whatsoever, and so it doesn’t count. And the answer from the supporters is one I really like: the night sky is itself infinite, so there’s no need to set up a gate around the act of looking at stars. The debate even forced telescope maker ZWO to step in officially with an article, offering point-by-point technical clarification of the traditionalists’ doubts. You see, even the power to define “what counts as real astrophotography” — that core question — is being fought over anew. At this very moment, the predicament of knowledge workers who make a living from words and thought when facing AI is, word for word, the one we face.
I can’t write the kind of watertight, elegant ending, because to this day I still haven’t fully climbed out of this predicament myself.
Staying at the top of the professional pecking order is, of course, extremely comfortable. The terrain up there is open, the company is plentiful, and every day brings more than enough amateur output to hold up as fodder for critique. But the endgame of traditional print media and television’s decline is a warning to us at every moment: that high platform built from professional barriers — the era is quietly, imperceptibly dismantling its foundation.
Yet if I go all the way toward mass popularization, deep down I have a question I’m unwilling to gloss over: those crafts I once studied so painstakingly — reading weather cloud maps, fine-tuning optical gear, laboring in professional post-processing software to rescue a noise-ruined starfield photo. Into these processes I truly poured a great deal of time and heartfelt love. I really don’t want to, and can’t, pretend against my heart that they’ve become worthless in the face of a tool.
In a new world where the most basic popularization work can be handled easily for two thousand yuan, where exactly should these hard-forged crafts be placed to become, once again, a distinctive value that others need? Should I go write advanced hands-on tutorials? Should I keep cultivating hardcore, deep popular-science content? Or help equipment makers refine more precise algorithms? Honestly, I don’t know either; every possible way out still looks, for now, full of fog.
The only thing I can do is a small daily habit against bias: next time, when I feel a moral indignation toward any emerging blockbuster tool or mediocre mass work and can’t hold back the urge to open my mouth and condemn it, before I hit the keyboard I first ask myself quietly, in my heart:
This time, the thing the new tool has beaten down to cabbage prices and devalued — is it, or is it not, my own craft?
If it is, then this irrepressible indignation is most likely just my instinct defending its rent advantage as it disappears; it merely sounds like conscience. These two inner voices really do sound far too alike.
And being able to tell them clearly apart as the wave comes crashing in is, perhaps, what that article truly meant to say: the independent judgment that cannot be outsourced.