Leiter Reports: A Philosophy Blog

News and views about philosophy, the academic profession, academic freedom, intellectual culture, and other topics. The world’s most popular philosophy blog, since 2003.

  1. Brent Yarnell's avatar
  2. Mark's avatar
  3. Mark's avatar
  4. AG Tanyi's avatar
  5. Carlo Burelli's avatar
  6. AG Tanyi's avatar
  7. NK's avatar

Anthropic mathematician (a human) claims to have proved Riemann hypothesis (with AI)

I guess “claims” is the key word. Has he?

(Thanks to several readers for sending this along.)

,

Leave a Reply

Your email address will not be published. Required fields are marked *

8 responses to “Anthropic mathematician (a human) claims to have proved Riemann hypothesis (with AI)”

  1. I’m a mathematician, but I don’t work in this area and I am not qualified to judge the proof. However, the reputation of this guy is enough that I think it’s very likely real.

  2. I’m not an expert on any of this, but, going by the statement he posted on his website, it sounds an awful lot like he’s having a psychotic break. (Compare it with, for example, the representation of psychosis in Wouter Kusters, “A Philosophy of Madness: The Experience of Psychotic Thinking.”)

    For example, he writes: “It was my secret in my student days that my plans were for this problem in decades, which meant then that I tried to become familiar with all tools that I could possibly recognize in the problem when it was time.”

    I guess that doesn’t necessarily mean he *didn’t* find a proof. Still, it gives me pause.

    1. For the record, his entire website is like this, so I am not sure this proves anything. As noted, his reputation precedes him, but of course, whether he did indeed provide a proof is for specialists to decide. There is more here:
      https://meteoraweb.com/en/news/anthropic-claude-riemann-hypothesis-alpoge-proof
      “Did Claude solve the Riemann hypothesis?
      No, according to Anthropic. It raised the proven minimum share of zeros on the line from 41.6% to 67.2%. Alpöge claims a complete proof, but it is not verified.”
      It will take weeks to verify the proof if not longer. The Navier-Stokes Equation is still listed as an ‘active problem’ not as a ‘solved problem’ on the Clay site since Clay’s requirements are demanding (publication and two years of scrutiny).
      In short, we won’t decide the matter here in this thread. Specialists will decide it in the coming weeks or months, or even years.

  3. The proof has been posted in Lean, which is a software that automatically checks its validity. So it should be valid, though the software sometimes bugs out, so it’s not fully confirmed yet. I am also reading the written paper is thousands of pages long, so it will take a lot of time for human verification and may be impossibile to check without AI assistance.

    We still have not fully understood open ai Navier Stokes proof, so at least within mathematics we may already have a kind of “superintelligence” (I know you don’t like the word, and its political appropriation was unfortunate, but here I think it fits): proofs that we know to be true but we cannot understand how or why.

    On one side this can promise an accelerate rate of progress, on the other it will open a whole range of ethical and safety concerns. Imagine AI discovers a cure for a disease that we know works, but we do not understand how or why.

    1. > Imagine AI discovers a cure for a disease that we know works, but we do not understand how or why.

      Although I agree with your broader point, my understanding is that this is actually already pretty common.

  4. For those interested, the checking of the lean code is happening here

    https://leanprover.zulipchat.com/#narrow/channel/113488-general/topic/Discussion.3A.20Alp.C3.B6ge.20Proves.20the.20Riemann.20Hypothesis

    This should not take more than a few hours, as I understand. But again, I am no math person.

  5. There’s now a claim that a human mathematician actually had a proof first, but it was still going through peer review:

    https://math216.wordpress.com/2026/11/20/hurried-announcement-alessandro-masullo-grh/

    The link includes a preprint which is 940 pages long! If true, precedence is going to get messy. Also makes me wonder if any parts of the proof were put through Claude at any point, in which case there’s a possibility of training data contamination…

Designed with WordPress