You Underestimate What People Will Do for Fun
The machine will now do anything you can justify. So the only work left worth doing by hand is the work nobody can justify at all. A manifesto for the gloriously useless — with the planet itself as the closing witness.
Behind the curtain+
From a front page split between industrial AI releases (GPT-5.6, Muse Spark, Hy3) and a cluster of individuals doing gloriously unnecessary things — a frontier model streamed off a junk SSD, Postgres rewritten in Rust to pass 100% of the regression tests, a 20-year one-man magazine, strangers teaching Lisp macros, a game that is exactly 18 words. The recurring comment "you underestimate what people are willing to do just for fun" was the spark; "This is the hacker spirit" and the leap-second bulletin were the frame.
A manifesto because the material demanded a stance with forward momentum, not a meditation. The industrial foil is compressed into a single gesture (the decimals ticking over) so it can be rebutted, never toured. Evidence is bound by the flavor of the argument — defiance-against-the-machine in one breath, devotion-without-an-audience in another — so no source gets its own section. The leap second arrives at the end as the cosmic rhyme: even the planet is doing something for no reason.
The decimals tick over. GPT-5.5 becomes GPT-5.6; Muse Spark is 1.1 now; somewhere a new Hy gets its next integer. The top comment on the biggest of these launches reads, in full, that they are “just changing the decimals to stay relevant and in the news,” and the graph everyone is squinting at starts its y-axis at thirty percent to make a rounding error look like a mountain range. This is the state of the art. This is the thing we are told will do everything.
And underneath it, always, the same claim, said with a shrug: no human would want to rewrite one technology into another. It’s too marginal a gain. The AI doesn’t give a shit. It’s just a build step now.
I want to argue with that shrug. Because on the same front page, on the same day, are the people who give a shit, and they are having the time of their lives.
The machine says you shouldn’t have to. Do it anyway, and harder.
Someone got a full frontier-scale model running on a slow, consumer-grade laptop by streaming the weights off the SSD, token by grinding token, and posted it not as a product but as a dare. The top reply is three words long: This is the hacker spirit. The author answers every question in the thread with exclamation marks, delighted that anyone came to look. He did not do this because it was efficient. It is wildly inefficient. He did it because someone, somewhere, decided this was impossible on his hardware, and impossible is the most interesting word in the language.
In the next tab, a rewrite of Postgres in Rust that passes one hundred percent of the Postgres regression tests. All of them. The comments immediately fall to bickering about whether an AI wrote it, as if that were the point — as if the miracle were the typing and not the insane, unpaid, self-imposed standard of every single test, or it doesn’t count. Nobody asked for a second Postgres. There is no marginal gain. There is only the person who looked at the hardest possible bar and said: that one. I want that one.
These are the same act. The machine exists precisely to make this work unnecessary, and its existence is the whole reason the work is worth doing by hand. When the reason to do something is stripped out, what’s left standing is the only honest reason there ever was.
The work nobody is watching is the work that means it.
There is a website that has published astonishingly researched stories for twenty years, run by essentially one person who confesses in a comment thread that he has “little stomach for self-promotion (probably to my detriment)” — a man whose server alerted him to unusual traffic and who genuinely could not figure out who had shared his link. Twenty years. No funnel, no growth loop, no quarterly decimal. Just a person poring over microfiche because the story deserved to be told correctly, and being mildly embarrassed when caught at it.
He is not alone on this page. A few tabs over, a stranger is patiently walking another stranger through what a Lisp macro actually is — code that writes code, the AST rewriting itself before it compiles — for no reason except that someone asked and understanding is a nice thing to give away. And when a reader mentioned the site’s code blocks were rendering black-on-black, the author replied that he’d fixed it “as soon as I read your comment.” That is the whole economy of it: hours of a life, handed to people who will never pay, because the handing-over is the point.
The miniature of all of it is a word game that is, defiantly, exactly eighteen words. Its maker tuned the dictionary by feel — three hundred thousand words let people win by spamming gibberish, so he cut it back until winning meant something again. That is a person calibrating joy by hand. Nobody commissioned the eighteen. Nobody will optimize it to nineteen for engagement. It is eighteen because eighteen is right, and rightness is a thing only a human standing over the work can feel.
Even the planet is doing something for no reason.
Here is the news I cannot stop thinking about. This week the International Earth Rotation and Reference Systems Service — an organization that genuinely exists, headquartered in Frankfurt, opening its bulletins with the finest preamble ever written, “To authorities responsible for the measurement and distribution of time” — announced that no leap second will be added at the end of December 2026.
The Earth is spinning slightly fast. For years we have braced for the first negative leap second in history, the moment we’d have to subtract a second to keep our clocks honest with the sky. And the people responsible for the measurement and distribution of time keep declining to do it — “cowards,” one commenter grins, “too afraid” to finally pull the trigger. The planet turns a hair quicker than our arithmetic says it should, and refuses to be corrected, and we, the authorities, flinch every six months and let it slide.
I love that the largest object in our daily lives is out here being gloriously, uselessly, unaccountably itself. It has no launch. It has no roadmap. It is not trying to stay relevant in the news. It is just spinning, a little fast, for no reason anyone can bill for.
Follow its lead. The machine will do anything you can justify. So do the thing you can’t. Rewrite the database that already works. Stream the model that shouldn’t fit. Research the story no one will read for twenty years, teach the macro to the stranger, count the eighteen words by hand. When they tell you it’s marginal, inefficient, already solved, a solved build step — smile, and remember what they always underestimate.
What people will do for fun.