Anthropic-linked mathematician Levent Alpöge claims general proof of the Riemann Hypothesis, with Lean code
Summarized from the Hacker News discussion because the article couldn't be read
- Commenters report that Levent Alpöge claims a proof of the Riemann Hypothesis and that the Lean formalization is split across a 14-part tar.gz archive linked in footnote 1.
- Commenters say the argument is too long to read, and the author chose not to release a PDF that would have run over 250,000 pages.
- A commenter reports that the author credits Claude with much of the final step, saying Claude far outpaced him near the finish line.
- Commenters note Alpöge was one of two mathematicians on the Navier-Stokes work OpenAI published, and that he has several verified results this year.
Hacker News opinions
Insane if true, but plenty of prolific mathematicians have claimed proofs that turned out to be wrong. Court of public opinion isn't the same as checking it.
The Lean formalization is linked in footnote 1 as a 14-part tar.gz. Didn't expect the author to ship the formal proof instead of a 250,000 page PDF.
I remember Wiles's Fermat proof took about two years for the community to accept it. This one might be worse. Short elegant proofs are probably over.
Saying he has Riemann because he's a top person is an appeal to authority. The proof should stand or fall on its own, and the mathematicians need to check it. Mochizuki is the example of smart people getting it wrong.
Honestly it's Claude doing most of the heavy lifting by the author's own account. The anthropomorphizing in that write-up is cringey.
My worry is nobody will ever read a proof this long. We'll hand the understanding to AI and it turns into belief instead of reproducible science.
I'd split it into answer and understanding. A machine gives you the answer and humans get the understanding later. That's still a big help.
How do you even run the Lean proof? Does anyone have 10TB of RAM lying around?