Nov 4, 2026 · 14 min
AI can solve harder math, but mathematicians still matter
AI is changing mathematics -- but not how you think
As AI-generated proofs become more capable, the central challenge is no longer solving problems alone but establishing that the solutions are reliable.
- 1Newer AI systems improve mathematical performance through feedback and answer-checking, reaching problems once associated with elite human competition.
- 2OpenAI’s disputed Navier–Stokes result shows how difficult it can be to evaluate long, opaque AI-generated proofs.
- 3Formal verification tools can catch some errors, but human oversight remains essential when small mistakes compound across research.
Don't miss
The discussion of OpenAI’s disputed Navier–Stokes proof shows why a long, sophisticated-looking argument can remain difficult to trust.
The brief
Geoff Brumfiel traces AI’s progress from basic arithmetic mistakes to much stronger mathematical performance, driven partly by systems that check and improve their own answers.
That progress has fueled a mathpocalypse narrative, especially after OpenAI released hundreds of purported results and drew attention to a claimed Navier–Stokes solution.
The Navier–Stokes case exposes the evaluation problem: an extremely long, difficult-to-read proof may be impressive, yet still resist meaningful scrutiny by mathematicians.
Lean can check formal proof code, but it cannot guarantee that the code matches the intended argument; a sign error helped force OpenAI to retract results.
The episode’s conclusion is less apocalyptic than transitional: AI will permanently change mathematical work, while human judgment remains necessary to detect accumulated errors.
Weekly email
State of the Pod, every Monday
The week's most-discussed names and stories from podcasts, with the moments behind them. Free, five minutes, unsubscribe in one click.