Anthropic 'formalizes' Fermat's Last Theorem like never before using Claude — but it still took 11 days to write out
Anthropic says Claude formalized Andrew Wiles’ proof of Fermat’s Last Theorem in 11 days, producing 13 million lines of Lean code.
Anthropic has formally verified Fermat's Last Theorem using its AI system, Claude, in an unprecedented 11-day period. The theorem, initially proposed by Pierre de Fermat in 1637, was previously proven by Andrew Wiles in 1995. Claude translated the theorem's proof into 13 million lines of code written in Lean, a programming language for mathematicians.
The AI system proved 30,300 separate theorems to reach the final version, using 29,500 of them. The resulting proof is five times larger than the community's main proof library. Kevin Buzzard, a mathematician, noted that the autoformalization verifies the theorem using only mathematical axioms. The achievement is Anthropic's second major math breakthrough, following a similar success with the Riemann zeta function. OpenAI is also working on similar AI-assisted math proofs.
Written by urgent.news from TechRadar's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
This story
This is one outlet's version. Read the fullest account.
