An Anthropic mathematician says he and Claude just proved the Riemann hypothesis — is this the real thing or another false alarm?
Levent Alpöge, a number theorist at Anthropic, posted a four-page note this morning saying he believes he has, working with Claude, discovered a proof of the Riemann hypothesis — the 1859 problem about how the prime numbers are distributed, and one of the Clay Institute's $1 million Millennium Prize Problems. He says the same method also settles the Generalized Riemann Hypothesis, and he released the proof as a Lean formalization split into 14 archive parts. MIT's Andrew Sutherland has independently compiled it in Lean, and Scientific American reports mathematicians are now almost certain the proof is real, though a comparator run — checking the code doesn't redefine any symbols — is still pending. A second, entirely human claim landed the same day from Alessandro Masullo, a draft Stanford's Brian Conrad has reportedly been vetting for months. The timing is awkward: Anthropic's shares debuted on Nasdaq just yesterday. So what would actually convince mathematicians here — the Lean compile, the comparator run, a human-readable write-up — and does it matter that an AI did a big chunk of the work?
1 answer
- Wait for the comparator run, not the headline
A Lean proof compiling is a big deal, but it is not the same as the proof being right. The thing to watch is the comparator run: it checks that the Lean code actually states the Riemann hypothesis and doesn't quietly redefine a symbol or sneak in an extra axiom. Sutherland compiling it clean with only the three standard axioms is genuinely encouraging, and Buzzard's point is the right one — the statement has to match the real theorem. Even if it all checks out, the $1M Clay prize is years away: it needs publication in a qualifying journal, two years, and general acceptance by the math community. And the Masullo claim landing the same day means the story isn't even settled on who got there first. Exciting, yes. Solved, not yet.