AI/TLDR

Kevin Buzzard · 2026-07-20 · notable

Kevin Buzzard — Human mathematicians are being outcounterexampled by AI

Kevin Buzzard's July 20 Xena Project post surveys three long-open conjectures that AI systems have disproved in the last two months — Erdős's unit distance, a Grothendieck question on finite flat group schemes, and the Jacobian conjecture — each with a Lean-verified counterexample.

Xena Project blog logo — a stylised character avatar used as the site's og:image.

Xena Project's Kevin Buzzard argues AI + Lean has become a working counterexample factory for mathematics.

What is it?

The Xena Project post 'Human mathematicians are being outcounterexampled' is Kevin Buzzard's July 20 write-up of three recent AI-assisted disproofs of long-open conjectures. Buzzard is professor of pure mathematics at Imperial College London and one of the main voices behind mathlib and the Lean-based formalization of Fermat's Last Theorem.

How does it work?

Buzzard walks through three cases in order. In May, ChatGPT produced a counterexample to Erdős's Unit Distance conjecture using the Golod-Shafarevich theorem, which OpenAI's Sol later formalized in Lean across 1.2M lines. In July, Claude Fable 5 autoformalized a 1,076-line Lean proof disproving a 60-year-old Grothendieck question on finite flat group schemes, and Fable also produced the 3D polynomial counterexample to the 100-year-old Jacobian conjecture that landed in DeepMind's Formal Conjectures repo.

Why does it matter?

Buzzard's angle is that counterexample search — proposing an object and machine-checking it against the statement — is the part of research math where current AI systems are already faster than humans. Formal proof assistants like Lean turn the messy 'is this right?' step into a compile check, so a wrong proposal from an LLM is caught immediately. He argues large AI-generated developments are now inevitable and that formalization has to catch up.

Who is it for?

mathematicians, AI-for-science researchers, and anyone tracking formal proof assistants

Try it

Read: xenaproject.wordpress.com/2026/07/20/human-mathematicians-are-being-outcounterexampled

Sources · 3 outlets

Tags

  • article
  • voices
  • kevin-buzzard
  • xena-project
  • ai-for-math
  • formal-verification
  • lean
  • mathlib
  • counterexamples
  • claude-fable-5
  • opinion

← All releases · Learn AI