Mathematicians Gowers, Sarnak: LLMs Lack Creative Math Intuition
TL;DR. Renowned mathematicians Timothy Gowers and Peter Sarnak contend large language models excel at known methods but cannot invent new mathematical ideas. - Gowers notes LLMs combine existing techniques and try many paths but miss intuition for productive routes. - Sarnak states AI derives results from theory but fails to develop abstractions for major proofs. - DeepMind's Tom Zahavy reached a similar conclusion, citing a lack of "manipulative abduction."
- Mathematicians Gowers and Sarnak believe LLMs are strong calculators of known methods but lack creative mathematical intuition.
- They argue LLMs struggle to develop new foundational assumptions or identify genuinely productive routes in problem-solving.
- This assessment adds to a broader debate about whether LLMs are becoming truly versatile or only improving on existing benchmarks.