bananaflag 1 day ago

The arguments are correct, but they are just what LLMs say, the author of the text is a known crank. One should not give them credit.

YetAnotherNick 1 day ago

If they are using AI to find counterexample, which it looks like they do, they should mention it. Various folks in X were able to find counterexamples in minutes with astra.

zug_zug 1 day ago

Just for anybody not paying attention, the Masullo proof is the human generated one. The AI generated one is more or less accepted at this point.

  • alanfranz 1 day ago

    The AI proof was released two days ago, it’s one million (or so) lines of Lean, and it’s accepted already??

    Who says so?

    • zug_zug 1 day ago

      So the whole point of lean is machine-checkable proofs. It's now the gold standard.

      So there's a public chat room of some of the worlds experts on lean (and also they happen to be great mathematicians) which I've been perusing and they say so themselves, they just were debating amongst themselves whether to put out an official statement yet. They also have verified it across 5 independent lean kernels. [1]

      I'm not saying I'm happy about it, just relaying the straight dope.

      1. https://math.mit.edu/~drew/grh/

      • alanfranz 22 hours ago

        Yes, I know there's a machine-checkable proof. But I also see that it may take a bit of time to verify that everything is all right, _and that the proof actually is about the GRH_

        Two days, frankly, seems a bit too short to me.

        I'm not concerned about being "happy".

      • bitwize 20 hours ago

        The drawback of Lean proofs is the same as the major drawback of "provably correct code". You can mechanistically prove a theorem, or a program, correct with respect to some formal specification. The tricky bit lies in the detail of: does your formal specification match what you actually want? In this case, does the Lean code correctly embody the Riemann hypothesis as a mathematician understands it? That is still an open question.

        I do not say this to cast doubt on Alpöge's work. It's just that we won't know know until the Lean has been shown, by human reviewers, to correspond with the natural-language understanding of the Riemann hypothesis. Or rather, when enough time goes by and no one working on the problem has found a major flaw or disharmony between the Lean code and the natural-language understanding, we'll probably all accept the proof as valid anyway. And it probably is valid. But there's still that very little tiny question mark which will probably keep a few mathematicians busy for a while—maybe decades. I think it took a few years for human checkers to work through the full implications of the four-color theorem proof, and this is just a much bigger-scale thing.

        • mirashii 4 hours ago

          > That is still an open question

          It’s not so much of an open question though. The GRH was formalized as a statement in mathlib independently and has already undergone that review, and the proof has been shown to prove the statement as formalized in mathlib. It has been the case before that we didn’t know the formalization of the statement was correct because it was also part of the paper providing the proof, but not here.

  • glimshe 1 day ago

    It is not more or less "accepted" . It's just seen as possibly correct given the credentials of the person who announced it.

    • estetlinus 1 day ago

      It is also harder to falsify 250 000 pages of math, rather than the human variant which fits in the context of 1 agent I guess?

DIE_DIE_DIE 1 day ago

Anthropic executives need to be imprisoned for life, at minimum.

  • hardbass 18 hours ago

    For what?

    • dudeinjapan 7 hours ago

      For advancing humanity's understanding of the mysteries of the universe, duh.