OpenAI's math proofs leave top mathematicians grieving, Steven Strogatz says
- OpenAI published hundreds of math findings in the last few months, and many proofs spanning number theory, theoretical computer science, and logic had stumped the world's best mathematicians for decades or centuries.
- Hugo Duminil-Copin, a 2022 Fields Medal recipient, said after OpenAI's October release that 'They have annihilated my field of research' and that he cried so hard it woke his daughter.
- Steven Strogatz of Cornell said he cried while speaking to Wired, describing disorientation, some days of exhilaration, and 'enormous grief.'
- Derek Thompson does not expect hundreds of millions of jobs to vanish within five years, but he expects intelligence-producing AI to collide with the view that human intellect gives people purpose and dignity.
Hacker News opinions
Mathematician here. I can't believe my colleagues didn't see this coming. Even if you bet on a plateau, you should at least admit there might not be one. I'm sad too, but mine was priced in.
I saw it coming after reading Kurzweil in high school, and it's still heavy. I just had a chance to flinch first. It moved really fast, though. A computer disproved the unit distance conjecture in May, and a 2024 MathOverflow comment had called AI useless for research.
The passage about physicists and writers weeping is interesting to me. These models can crack open math problems but still write in a style you can spot from a mile away. That's very odd, IMO.
As a programmer I've been weeping for a few years now. Translators, doc writers, artists, there are so many. Either these mathematicians assumed their jobs were safer or they were living under a rock. Either way I don't feel more for them than for everyone else.
I'm reading Strogatz's Nonlinear Dynamics and Chaos right now. Nothing in the age of AI threatens how beautiful that book is.
I think the point is that nobody in the age of AI will get to build a career and produce work like Strogatz did.
I still think mathematicians will matter. LLMs are only finding what was possible with the knowledge we already have, through brute force. Once they run out, they'll need new human-found knowledge, which is harder to find but much more valuable.
How do you know they'll run out? That assumes humans can be original in ways LLMs won't be, and I doubt that holds.
What AI is doing with math is basically fuzzing for valid inputs. It doesn't understand what's going on, so it won't build the next generation of abstractions that proofs need.
We already survive on abstractions other people built. If a solar flare knocked out the grid, most of us would be dead within a year. AI abstracting knowledge isn't new, and this argument doesn't prove it's dangerous.
Software engineers are getting laid off right now, while mathematicians keep their university jobs. Mathematicians can still feed their families. I see far more blogs about mathematicians than about us.