Math Got Solved. Mathematicians Want a Refund.

Listen to Free Audio Version Here

On October 6, OpenAI walked into the house of mathematics, dropped more than 700 preprints on the kitchen table, and left them on GitHub for everybody to find. All of them generated by an unreleased internal model. Scientific American counted results on 372 open problems in math and computer science. Not one problem. Not one clever paper. A dump truck.

And the mathematicians? They did what any proud guild does when the machine shows up at the door. They threw a fit.

One mathematician called it a “slopocalypse.” Alvaro Lozano-Robledo, a number theorist at the University of Connecticut, called the sheer number of papers “staggering” and asked why OpenAI would release hundreds of solutions all at once. Ursula Martin at Oxford said OpenAI was “tossing the community a messy first draft and expecting us to clean it up for them.” Michael Harris at Columbia went full monk: “I can’t bring myself to spend my time studying the outcome of this process, which I consider illegitimate. So I will not be reading these papers.”

He will not be reading them. Picture a 1910 stable owner announcing he will not be looking at the automobile. The automobile is fine with that.

My favorite came from a group calling itself the Association for Human Mathematics. Quote: “Releasing over 700 files at once is not a demonstration of scholarship, but a demonstration of power.” Well, yes. That is exactly what it is. That is the whole point. Nobody ever demonstrated power by politely asking if they could be on the program next fall.

Now let me be fair, because the facts matter more than my mood. This did not come out of nowhere. Back in May, an OpenAI model took on an Erdős conjecture that had stood for about 80 years, the unit distance problem, and showed Erdős was wrong. Cambridge’s Tim Gowers, a Fields medalist, said that if a human had submitted it to the Annals of Mathematics, he would have recommended acceptance. Will Sawin at Princeton said his first reaction was disbelief, and then he became convinced it was the most significant achievement by AI in mathematics so far. Then in September, according to Nature and Scientific American, the same internal model produced a Navier-Stokes preprint, one of the famous million-dollar Millennium Prize problems. And now this pile.

So the warning shots were fired. Months ago. Out loud.

And it is not all slop. Scientific American reports that about 42 percent of the results are coded in Lean, a formal proof language where the computer checks every step. That means their logic is almost certainly sound, even when the English write-up reads like a tax form translated through three languages. Harvard’s Michael Douglas said, “Yesterday will be the day when a new era of math and mathematical physics began.” Levent Alpöge, a mathematician who works at rival company Anthropic, admitted there will be sad stories for researchers who got scooped, then said, “It’s obviously the most significant moment in mathematical history.”

Are there problems? Of course. Three of the papers have already been withdrawn after experts found a significant error. MIT’s Andrew Sutherland expects most of the proofs to hold up, but says some will contain mistakes, and he wants OpenAI to release the thousands of problems it failed on, because that is how open science works. Fair point. Show the misses with the hits. I’ll give him that one for free.

But listen to the actual complaint underneath all of it. It isn’t really “the math is wrong.” Most of it looks right, or fixable. The complaint is that it’s ugly. NYU’s Roland Bauerschmidt said some of the papers are so badly written that if he had received them by email from a nobody, he probably would have deleted them. Back in May, Science News summed up the Erdős result as perseverance rather than creative insight. It just ground through the trial and error that no human would have the patience for.

Google Ads copy by Twin Rivers Communications

Translation: it was right, but it wasn’t beautiful. It got the answer, but it didn’t suffer properly.

Here is where I do some high-end pondering. For three hundred years, mathematicians told the rest of us that their work was the purest thing in the human mind. Elegance. Beauty. Truth. And it turns out a big chunk of the job was patience and grinding, and a machine has more patience than any tenured professor who ever lived. It doesn’t get tired. It doesn’t need a sabbatical. It doesn’t fight in the faculty lounge over credit.

So the old guard has been cooked, as we say. Not the math. The math is doing great. What got cooked is the idea that the only valid route to a true answer runs through a human being with chalk on his sleeves. The proof is the proof. If the answer is correct, nobody in fifty years will care whether it was hand-carved out of walnut or stamped out by a machine at three in the morning. Nobody asks whether your calculator showed its feelings.

And mathematics is just the tip of the iceberg. It goes first for a simple reason: in math you can check the work. A proof either holds or it doesn’t, and Lean can tell you which. That makes it the cleanest test lab on earth for machine intelligence. Biology is next in line. Drug discovery, protein design, materials, chemistry, medicine. Every field that runs on long hours of trial and error and a priesthood guarding the doors is about to get the same 700-file surprise on the kitchen table. And they’ll throw the same fit, with the same open letters. Back in June, more than fifteen hundred people signed a declaration calling for tight guardrails on AI in math research. You can sign as many declarations as you like. The ocean doesn’t read them.

People still don’t understand what’s coming. They think the intelligence explosion is a movie with red-eyed robots. It isn’t. It looks like this: a quiet Tuesday, a GitHub folder, and a few hundred problems that stumped humans for decades that are suddenly not problems anymore. Then the next field. Then the next one.

One last piece of high-end pondering, the gallows kind. Somewhere a grad student spent four years on a problem that got solved in that folder. That’s a real loss, and I’m not laughing at him. But the answer to that pain isn’t to stop reading the papers. It’s to learn to ride the thing. The mathematicians who pick up these tools are going to do more math in the next five years than their whole field did in the last fifty. The ones who boycott will be very principled. And very retired.

The machines don’t need your blessing. They just need a problem. Fortunately, we have plenty of those.

Sources:

My Living AI News Now. Your Place for AI News. Commentary, not financial advice.

C. Rich

Your Toaster Is Not Plotting Against You

Nice or Nasty: Which One Makes AI Smarter?

Somebody Just Pulled $30 Billion Off the Table