Fields Medalist Timothy Gowers Asks: What Sort of Maths Are LLMs Good At?

stevenstrogatz · x · 2026-08-12

Fields Medalist Timothy Gowers published an in-depth blog post exploring the boundaries of current large language models (LLMs) in mathematical research and the specific types of problems they excel at. The article was written shortly after OpenAI announced it had solved ten major mathematical and theoretical computer science problems, including the construction of a non-sofic group.

While acknowledging these breakthroughs as extraordinarily impressive, Gowers points out that LLMs are not yet better than humans at all aspects of mathematics. If they were, their sheer speed would have already flooded academia with new results. He attempts to rule out some bad classifications and explores what kinds of mathematical problems LLMs are currently best at, highlighting areas where there is still significant room for improvement. The post serves as both a snapshot of current AI mathematical capabilities and a baseline for evaluating future progress.

Original post →

More from AGI Musings

AGI Musings channel →