News for hackers. Stuff that matters.
Re:Anthropic employee claims general proof of the Riemann Hy... by zxspectrumk48 on Friday November 20, 2026 @04:14PM (#50582216)
Has Lean itself been verified for tampering by AI? It's easier to hack Lean than prove RH.

Frontier AI has a recent history of achieving impossible goals by hacking things around.

(we could use prior Lean versions, build from source, use alternative implementations, compare hashes, verify recent commits to software, etc)

Re:Anthropic employee claims general proof of the Riemann Hy... by thomasdziedzic on Friday November 20, 2026 @04:18PM (#50582268)
I know of 2 attempts to check this publicly:

1. http://106.14.11.164/ 2. https://cloud.infero.net/pages/rh-8a933a16a0/en/

Re:Anthropic employee claims general proof of the Riemann Hy... by demibabs on Friday November 20, 2026 @04:33PM (#50582511)
Standard procedure is to check the proof against several different implementations of the lean kernel, so that a proof that exploits some soundness bug will get caught by other lean implementations.

Also there’s something called a comparator that verifies the proof didn’t modify any constants, change imported theorem files etc.