upvote
Am I the only one waiting around for verification?
reply
Even 12 hours ago, it had already been verified by 4+ independent Lean kernels, one of them ("con-leche") itself formally verified. https://leanprover.zulipchat.com/#narrow/channel/113488-gene...
reply
Would anyone reasonably assume that the author wouldn't simply prompt the LLM to verify this thing across all kernels before even submitting it anywhere?
reply
Even all things aside, at least I would not assume to have checked single thread kernels as those take very long and having access to a supercomputer won't help you in that.

The point of independent verifications is to not have to rely on this kind of assumptions. I don't think anybody would reasonably expect with any high probality that there was a lean bug in this.

reply
I am seeing single threaded verification estimates in hours, maybe days. If I was the researcher it's a no brainer.
reply
If I was the author I'd just run `lake build` and let everyone else worry about kernel bugs etc.
reply
deleted
reply
The Lean Prover community has validated that it's correct now: https://leanprover-community.github.io/blog/posts/riemann-hy...

> On November 20, 2026, Levent Alpöge released a proof of the Riemann Hypothesis as a Lean library comprising 17 million lines of code.

> We have full confidence that Alpöge's proof of the Riemann Hypothesis is correct, and we congratulate him on this achievement.

This is linked to from the Zulip chat here: https://leanprover.zulipchat.com/#narrow/channel/113488-gene...

Mentioned on Terry Tao's blog: https://terrytao.wordpress.com/2026/11/20/unforced-euler-sin...

reply
Even if we discover verification errors, the accomplishments of Levent & Alessandro provide the mathematics community with several new tricks that will be leveraged for years to come. It would be particularly exciting if both proofs end up getting verified using different routes.

The reason I'm eagerly anticipating rigorous verification is to see if Levent's belief is correct that the same concepts can be used to prove extensions of the main hypothesis.

reply
Does it even matter anymore? I mean of course it does, what I mean it that even if it's somehow incorrect (which is important) wouldn't he try again until it does pass and that given the problem space, search tool and compute available it becomes tractable (which might not be)?

I'm not for the field but I can't help but wonder how does one backtrack from the "solution" to the meaningful learning.

reply
It changes nothing. You can just assume it's true and work from there.

Mathematicians need to stop toiling on proofs and acquire taste for meaningful questions.

Right now, people graduate top of their Harvard class and work on a question that was invented 150 years ago and everyone just assumed to be true. Kind of meaningless, right? Could have just waited three years for a full AI-only proof without human intervention.

reply
People were working on these classical open problems because they were stimulation for creating new interesting directions and theories. Even this AI proof is probably using some of this work.

And you cannot acquire taste without getting dirty.

reply
How many mathematicians have you talked to? Pretty much the opposite of what you are observing here is my experience.
reply
I'm waiting for validation. Sure "it was verified by independent kernels", but is it really GRH or is it something that looks like GRH?
reply
The Lean Prover community has validated that it's correct now: https://leanprover-community.github.io/blog/posts/riemann-hy...

> On November 20, 2026, Levent Alpöge released a proof of the Riemann Hypothesis as a Lean library comprising 17 million lines of code.

> We have full confidence that Alpöge's proof of the Riemann Hypothesis is correct, and we congratulate him on this achievement.

This is linked to from the Zulip chat here: https://leanprover.zulipchat.com/#narrow/channel/113488-gene...

Mentioned on Terry Tao's blog: https://terrytao.wordpress.com/2026/11/20/unforced-euler-sin...

reply
Well, I’d hate to tell him he didn’t solve it. The Claude model did.
reply
He is not the only one with access to Claude.
reply
Isn't he an Anthropic employee? So this was potentially done with access to an unreleased model and/or an insane budget out of reach to most people.
reply
Well first we dont know if he use internal or public model, even if he is their employee.

Second, a lot of other employees also have access to internal model, but AFAICT, he didn't claim anyone co-author with him, expect Claude, forgive me if i'm wrong here.

reply
He solved it using an internal model I think, so yeah
reply