Human Mathematicians Are Being Outcounterexampled by AI

What happened

Frontier AI models (Claude Fable, ChatGPT Sol) rapidly resolved decades-old open mathematical problems by generating formal counterexamples — including disproving Erdős' Unit Distance conjecture and a 60-year-old Grothendieck question on group schemes. All results were formally verified in Lean.

Context and impact

Kevin Buzzard (Imperial College, Xena Project) argues that 'large AI-generated developments of mathematics are inevitable' and the pace of formally verified AI proofs is growing to thousands of lines per week. This is prompting a fundamental reassessment of what constitutes mathematical discovery and the future of human mathematicians — especially as the US cuts human mathematician funding.

Details

• Erdős' Unit Distance conjecture disproved (decades-old problem)
• Counterexample found for 60-year-old Grothendieck question on group schemes
• All results formally verified in Lean theorem prover
• Rate: AI now generates thousands of lines of formal proofs per week
• Author Buzzard: 'large AI-generated mathematical developments are inevitable'
• Implication: challenges the very definition of 'mathematical discovery'