Anthropic uses Claude to produce the first complete computer-checked proof of Fermat’s Last Theorem
Anthropic researchers used the AI model Claude to produce the first complete computer-checked proof of Fermat’s Last Theorem. While the theorem was first proposed in 1637, the first correct human proof was published in 1995 by Andrew Wiles.
Show the rest of this summary
The new computer-verified version was created by formalizing Wiles' proof into the Lean programming language, which allows computers to automatically verify mathematical logic. Researchers expected the formalization process to take several years, but the AI completed the task in 11 days. The process involved dozens of agents producing 6 billion tokens of output and proving 29,500 intermediate theorems. The resulting proof consists of 13 million lines of Lean code, making it the largest-ever file of its kind. The breakthrough was achieved by using an open-source tool called Prove2Me, which helped the AI agents determine the optimal next steps in the workflow. Kevin Buzzard, a mathematician who reviewed the proof, noted that the ability to easily formalize work can lighten the burden of evaluating new results.
Sources
-
Formalizing Fermat's Last Theorem
Anthropic
-
Anthropic uses Claude to formalize proof of Fermat’s Last Theorem
SiliconANGLE
Paywall and unreadable sources
-
AI Just Solved a 350-Year-Old Math Problem By Writing the Longest Proof Ever
Decrypt
-
Fermat’s last theorem formalised by AI agents in just 11 days
New Scientist
-
Claude formalises Fermat's Last Theorem in 11 days, task that would take humans yrs | The proof totals over 13 million lines of code | Inshorts
Inshorts