Fields Medalist Tests ChatGPT: Solves PhD-Level Math Proofs in Under 2 Hours
新智元 · wechat · 2026-07-23
As LLM capabilities surge, AI is upending mathematical research. Fields Medalist Timothy Gowers found that ChatGPT solved cutting-edge math problems in under two hours, generating proofs deemed entirely original by the authors of the papers being improved.
AI Cracks Research-Level Problems
- Gowers tested ChatGPT on a combinatorics problem. The model compressed an exponential bound to a polynomial level in minutes and generated a formal preprint.
- Isaac Rajagopal, the author of the original paper, confirmed that the AI's approach was not only correct but a highly original insight.
- An internal OpenAI reasoning model also recently disproved Erdős's 1946 unit distance conjecture, shifting a decades-long consensus.
The Academic System Faces Shockwaves
- Top scholars, including xAI co-founder Christian Szegedy, suggest that as AI contributions dominate, the Fields Medal may lose its original meaning, predicting this boundary will last only until 2030.
- Gowers noted that if AI generates the core ideas and technical proofs for major theorems, the era of attaching a human name to a theorem is ending.
- Mathematicians like Terence Tao are already changing their workflows, unable to keep up with AI-generated proofs, and suggesting separate publication venues for AI versus human mathematics.
More from AGI Musings
- Cognitive mercantilism: AI panic eerily mirrors early capitalism's anxieties — adityaag · 2026-10-27
- US chases AGI dominance while China, EU and ASEAN treat AI as public infrastructure — BubblyExtent1693 · 2026-10-27
- Lab monkeys at $30-80K each: a 4-year breeding cycle is the real preclinical drug bottleneck — zebird0 · 2026-10-27
- CJR: journalists should stop hyping AI's supposed humanity — just say 'rogue' — ShakeelHashim · 2026-10-27
- PhD applicant emails up 4x YoY due to AI — writing without AI now makes you stand out — AI4Code · 2026-10-27
- Weil's collapse shows BigLaw dies from eroding belief, not a broken business model — jkubicki · 2026-10-27