On October 6, OpenAI released over seven hundred mathematical manuscripts at once, written by a model that isn't even out yet. Claimed solutions to open problems, in a pile, overnight. Three were withdrawn within a day, and outside review has barely begun. But that's not the event. The event is that so much came, so fast.
The mathematicians have taken it hard. Proofs and Prompts collected over a hundred responses from researchers after the release, and they range from curiosity to grief over a way of working that is on its way out. Hugo Duminil-Copin, a Fields medalist, wrote that the shock was immense and that he was paralyzed — but also that tomorrow we will find a way forward. Scientific American had already asked, before the release: if AI can do math, what's the point of mathematicians?
I understand them. This isn't a tool that has taken over one step of the job. It's a life's work compressed into a few hours of computation. Someone has spent twenty years on a problem and now reads that it may be settled in one file among seven hundred, and that no one has yet had time to look at it.
And why would it stop at mathematics? Mathematics came first because it's the easiest to check, not because it's the hardest. I think physics is next in line, then biology. Everywhere a result can be tested without anyone having to build anything, the same thing will happen.
What I think we'll have to get used to is that human brainpower is no longer what's scarce. Faster, better, cheaper is the whole argument, and it doesn't have to apply to everything to apply to enough. I've built a career on being the one who thinks things through for others. I know what I'm talking about.
Scott Aaronson is a complexity theorist and worked for a while on safety at OpenAI. He called October 6 one of the biggest days in mathematical history. In the same breath he said that just about none of the proofs has yet been understood by any human. In the pile was a claimed proof of the Unique Games Conjecture, which his wife — also a complexity theorist — has worked toward proving for the entire time he's known her. He still thinks it's untenable to want to keep the models away from open problems, and he's back at the university in Austin.
For my part, I'm not going to stop thinking because someone else thinks faster. I'll keep up the woodworking too, and nobody does woodworking because it's the cheapest way to get a shelf. The forest doesn't care what was released on GitHub. Time with the family doesn't get shorter because a model computes faster.
I have done none of the things I like best because I was the most efficient at them.