Levent Alpöge, a mathematician at Anthropic, has successfully disproved an 87-year-old conjecture regarding polynomial maps first posed by a German mathematician in 1939, demonstrating that it is not always possible to recover a single input from given outputs. This groundbreaking proof was verified using Lean, a formal verification tool that ensures machine-checked validation for complex proofs. This event highlights a growing trend where mathematicians combine domain expertise with advanced AI models to solve long-standing mathematical challenges.

Anthropic: Anthropic is an AI research company that builds advanced language models, with Claude as its flagship system designed for safe and reliable reasoning. Levent Alpöge, a mathematician at the company, applied one of its models to discover a counterexample that resolves a long-open mathematical question. This work underscores Anthropic’s role in supporting cutting-edge applications of AI beyond traditional domains.
Levent Alpöge: Levent Alpöge is a mathematician employed at Anthropic who specializes in algebraic geometry and related fields. He identified the counterexample to the 1939 conjecture using AI assistance and had the resulting proof verified in the Lean formal system. His contribution demonstrates how domain experts at AI labs are accelerating progress on longstanding theoretical problems.

AI-Assisted Proofs: Mathematicians are increasingly pairing domain expertise with advanced AI models to tackle unresolved questions in pure mathematics.
Formal Verification Tools: Theorem provers such as Lean provide independent, machine-checked validation for complex proofs developed with AI support.