Tim Gowers: What sort of maths are LLMs good at?
Summary
Tim Gowers analyzes LLM capabilities in mathematics, arguing they are strong at producing and testing examples and counterexamples and can handle universal statements but may miss deep, insight-driven proofs. He uses Vinogradov's theorem and Gluskin's work to illustrate differences in logical forms, and discusses future directions for training AI systems to assist mathematical discovery.