• The_Decryptor@aussie.zone
      link
      fedilink
      English
      arrow-up
      3
      arrow-down
      1
      ·
      42 minutes ago

      I mean, we’re still at 1 here, there’s a reason it’s outputting Lean code rather than actually doing any computation by itself (Because it can’t do that, it’s a text generator)

      • hirihit640@sh.itjust.works
        link
        fedilink
        English
        arrow-up
        1
        ·
        9 minutes ago

        If you ask a local LLM to multiply numbers together without using any external tools, it will still get more right than the average person (who isn’t using any external tools, like pen and paper)

  • melfie@lemmy.zip
    link
    fedilink
    English
    arrow-up
    9
    arrow-down
    1
    ·
    4 hours ago

    Is it even possible to distill this down to the type of proof a human would produce that is orders of magnitude more concise and has coherent meaning, or is it just largely slop that would be easier to start from scratch with the monumental task of a human-made proof than it would be to painstakingly unravel it?

    • bss03@infosec.pub
      link
      fedilink
      English
      arrow-up
      2
      arrow-down
      1
      ·
      57 minutes ago

      Probably some of both. Though, given the short history on these proofs, it will likely be more “slop” (or theft) than anything else.

      While this is orders of magnitude larger, the first 4-color theorem theorem had a lot of cases that were quick for a computer to exhaustively check. But, it turns out a lot of those cases are better explained without exhaustive checking, so the currently taught / best proof of the 4CT is shorter and has less long, mind-numbing checks.

      It’s entirely possible that there’s actually no more “insight” from the LLMs, and they were just better able to generate the massive type signatures and terms that Lean needs to begin mechanically checking, because they are able to generate and discard good syntax much faster. But, there might also be some interesting bits that people (or LLMs) will be able to extract/abstract/generalize/shrink into something humans will find insightful. I think the Navier-Stokes related proof did have some stuff in it that humans found interesting beyond the final result.

  • fluxx@lemmy.world
    link
    fedilink
    English
    arrow-up
    29
    arrow-down
    4
    ·
    6 hours ago

    As a non-mathematics person, this still depresses me so much. It’s like like a kid finding out there is no santa. Or a religious person realizing there is no god. I can’t really quite describe what about this exactly depresses me, but I feel like ground has been taken beneath my feet. Not just this, the whole llm thing. Not because I’m afraid of losing my job or status. Just that the magic is gone makes me feel sad.

    • hirihit640@sh.itjust.works
      link
      fedilink
      English
      arrow-up
      1
      ·
      3 minutes ago

      I guess I never thought human intellect was very special, and when I learned about neural networks, this sort of AI progress felt inevitable. It’s all still pretty mindblowing, but not completely inevitable

    • bss03@infosec.pub
      link
      fedilink
      English
      arrow-up
      2
      ·
      41 minutes ago

      Honestly, I think CS research has hinted we were going in this direction anyway. Dependently-typed languages like Idris and Agda have as term-inference for a while, it was based on exhaustive type-directed heuristically-prioritized search but it could still auto-complete/fill some non-trivial code, and Roq tactics could also make things that were disastrously wordy read much more simply and they are effectively a form of compile-time function (Elaborator Reflection is probably a better approach, but Roq’s two-language model is much more well-established).

      Even without quite so interesting types, there were academic exercises like “Djinn” that could exhaustively enumerate lambda-calculus function bodies given the parametrically polymorphic type signature and a library of standard functions, and that’s been available for 14+ years? In an entirely different direction we’ve had SMT solvers can could be plugged in to a few languages to verify a LOT of properties that can be expressed in first-order logic.

      Still, I think there’s plenty of magic, it’s just not in the places we can lay down (enough) axioms around symbolic structure / grammars.

    • zaphod@lemmy.ca
      link
      fedilink
      English
      arrow-up
      16
      ·
      5 hours ago

      Probably because human thought and intellect are the things that we, as a species, see as differentiating ourselves from any other animals on this planet. And so these intellectual and creative domains feel (and, in the past, were) exclusively the domain of humanity.

      Automation is now taking that away.

      What do we do when it seems like we, as a species, are on track to render ourselves useless?

      I don’t know.

      I wonder if this is what old world explorers felt like when the globe was fully mapped and there was no true frontiers anymore. What do you do when the thing that gave you a unique sense of purpose disappears in the face of technology and automation?

      What do we as a society do when everyone experiences that at the same time?

      • redballooon@lemmy.world
        link
        fedilink
        English
        arrow-up
        13
        ·
        4 hours ago

        I, personally, would be totally fine to spend my days playing unproductive games. Just give me UBI.

        • zaphod@lemmy.ca
          link
          fedilink
          English
          arrow-up
          2
          arrow-down
          1
          ·
          35 minutes ago

          Honestly, me too. But we both know that ain’t how it’s gonna go down…

  • Bustedknuckles@lemmy.world
    link
    fedilink
    English
    arrow-up
    79
    arrow-down
    1
    ·
    9 hours ago

    Isn’t this “Independent mathematicians who also used Lean software to check the proof that the author wrote/checked with Lean”?

    It could still be valid of course, but I thought the proof was like 250000 pages and has only been looked at by the same automated framework. This is out of my wheelhouse and I don’t have a dog in the fight, but I’d still like some expert human eyes on it

    • bss03@infosec.pub
      link
      fedilink
      English
      arrow-up
      9
      ·
      2 hours ago

      They used 2 checkers that are maintained and developed independently of the Lean kernel. But they did use Lean formats/syntax. They’ve also done some manual verification, though there might be more to do. The proof did use the statement if RH directly from the well established Mathlib.

      I’ve also seen claims that the proof has some weird things in it that might be hints that it is finding holes in the theory Lean models.

      Honestly, even if it ends up not being a proof of RH, it will tell us something about computable / constructive maths.

    • wewbull@feddit.uk
      link
      fedilink
      English
      arrow-up
      24
      arrow-down
      5
      ·
      6 hours ago

      It’s 250,000 pages of explanation for 17 million (I think) lines of lean code.

      It’s incomprehensible, and therefore probably of little value.

      • TapedeckNinja@lemmy.worldOP
        link
        fedilink
        English
        arrow-up
        1
        arrow-down
        1
        ·
        1 hour ago

        It’s incomprehensible as a whole but not in pieces.

        It will be simplified, refactored, maybe rebuilt from the ground up. It won’t be incomprehensible for long.

        The last I looked the size had already been reduced by 25% based on community efforts.

        Kind of fun watching the Lean Prover Zulip chat play with it.

    • TapedeckNinja@lemmy.worldOP
      link
      fedilink
      English
      arrow-up
      60
      arrow-down
      4
      ·
      9 hours ago

      There are a bunch of ways that a proof written in Lean can be invalid.

      The most obvious is that the problem statement is invalid or incomplete. This was not really a concern here as the proof used Mathlib’s built-in definitions rather than restating the problem, but nonetheless it has been verified.

      Beyond that Lean proofs can contain custom axioms/assumptions, shortcuts, tricks, exploits. And there’s the possibility that the Lean compiler used in the proof had a bug.

      Anyway all that has been tested and verified. The problem statement is valid. The proof doesn’t contain any tricks or assumptions. It compiles on every major Lean kernel.

      Basically: it is almost certainly mathematically true. The next step is humans actually understanding it.

  • BillyClark@piefed.social
    link
    fedilink
    English
    arrow-up
    29
    arrow-down
    1
    ·
    8 hours ago

    On Friday Anthropic mathematician Levent Alpöge claimed to have used the company’s Claude AI model to solve the Riemann problem, perhaps the most important open quandary in mathematics.

    “We have full confidence that Alpöge’s proof of the Riemann Hypothesis is correct, and we congratulate him on this achievement,” mathematicians wrote in a joint statement by the Lean and Mathlib communities

    Are we calling it “Alpöge’s proof” despite the fact that he didn’t come up with it?

    • richmondez@lemdro.id
      link
      fedilink
      English
      arrow-up
      17
      arrow-down
      2
      ·
      8 hours ago

      I think it’s unclear how the LLM was involved, I doubt the LLM came up with it itself. It’s a tool and assisted him but he ultimately is responsible for it. If it were garbage he’s the one who will take the reputations hit.

      • BillyClark@piefed.social
        link
        fedilink
        English
        arrow-up
        13
        ·
        7 hours ago

        I doubt the LLM came up with it itself

        In math, these things are attributed to those who can actually complete the proof. Maybe this could be called Alpöge’s hypothesis of the Riemann problem’s proof or something if he did come up with the idea.

        I don’t think ownership is determined by whose reputation would take a hit. I mean, at this point, the Lean and Mathlib communities would also take a reputation hit if it was proved wrong, but it doesn’t make sense to name it after them.

      • potustheplant@feddit.nl
        link
        fedilink
        English
        arrow-up
        16
        arrow-down
        1
        ·
        5 hours ago

        Maybe all mathematicians that ever published anything should be credited? After all, the plagiarism machine can only predict characters based on what it’s been trained on.

      • BillyClark@piefed.social
        link
        fedilink
        English
        arrow-up
        6
        arrow-down
        1
        ·
        7 hours ago

        Claude isn’t a person, so no. I think if Claude did the important work, then it shouldn’t be named after anybody.

          • bss03@infosec.pub
            link
            fedilink
            English
            arrow-up
            2
            ·
            37 minutes ago

            You don’t HAVE to name mathematical objects after a person. We could just call it Riemann’s Theorem (a proved hypothesis is a theorem) and include the date and by-line if we need to reference a specific one in (e.g.) a bibliography.

            Decolonize maths!

          • Beej Jorgensen@lemmy.sdf.org
            link
            fedilink
            English
            arrow-up
            9
            ·
            6 hours ago

            Nobody made it?

            Is that so weird, though? I’m drinking soda from a can that nobody made. My desk is covered with things that nobody made. Bots make tons of stuff and we don’t name those things after the bots or their operators.

          • BillyClark@piefed.social
            link
            fedilink
            English
            arrow-up
            10
            arrow-down
            1
            ·
            7 hours ago

            There’s no reason to think that a “body” has to make something just because it exists and is complicated. That’s the same mistake that creationists make instead of understanding evolution.

            But that’s all off-topic. The question isn’t where it came from, but who it is named after.

            • FaceDeer@fedia.io
              link
              fedilink
              arrow-up
              3
              arrow-down
              5
              ·
              7 hours ago

              That’s already long established. It’s the Riemann hypothesis, it’s right in the title.

              I’m not sure what you’re suggesting at this point. When it’s published in a journal are you suggesting the byline should be left entirely blank?

              • BillyClark@piefed.social
                link
                fedilink
                English
                arrow-up
                9
                ·
                7 hours ago

                You have completely forgotten the context for these comments, so rather than asking me to explain things, I think it would be more straightforward for you to just go back and read the comment chain.

                • FaceDeer@fedia.io
                  link
                  fedilink
                  arrow-up
                  2
                  arrow-down
                  4
                  ·
                  7 hours ago

                  Alright, so should it have been:

                  “We have full confidence that nobody’s proof of the Riemann Hypothesis is correct, and we congratulate nobody on this achievement,” mathematicians wrote in a joint statement by the Lean and Mathlib communities

                  Because that seems quite unhelpful to me.

    • missingno@fedia.io
      link
      fedilink
      arrow-up
      18
      ·
      5 hours ago

      We’ve confirmed something that was long suspected to be true, but learned nothing in the process. Hard to say if humans can even understand the proof. It’s true… and that’s it.

      Usually the importance of a problem like this is that the journey to solve it leads to major breakthroughs that can touch other fields of mathematics. So it’s a bit surreal to see the most famous unsolved problem finally felled, with the caveat that we learned nothing in the process.

      • A_A@lemmy.world
        link
        fedilink
        English
        arrow-up
        5
        ·
        3 hours ago

        Ha ! So this is a good starting point for a science fiction story where an artificialy intelligent system creating this proof learns something that gives it an edge over the human specie.

    • Ledivin@lemmy.world
      link
      fedilink
      English
      arrow-up
      51
      arrow-down
      1
      ·
      9 hours ago

      The short answer is “not much.”

      Us assuming that the Riemann hypothesis is true is the basis of much of our mathematics. If it was proven false, it could affect a ton of the math we’ve built on top of it for centuries, but it being proven true is just confirmation of what everyone was sure of but couldn’t quite prove.

      • ExcessShiv@lemmy.dbzer0.com
        link
        fedilink
        English
        arrow-up
        11
        arrow-down
        1
        ·
        8 hours ago

        math we’ve built on top of it for centuries

        Well, almost centuries at least. The Riemann hypothesis is less than 200 years old (167)

    • theherk@lemmy.world
      link
      fedilink
      English
      arrow-up
      19
      ·
      9 hours ago

      It’s counter proof would have been monumental. This is big. Perhaps the biggest of ours and a few generations surrounding, but not practically so.

      It changes little. It’s the result we’ve come to expect. Some other conjectures will use it, maybe Goldbach.

      But the disproof would have hurt a lot of number theory, maybe engineering, and probably crypto.

        • badcommandorfilename@lemmy.world
          link
          fedilink
          English
          arrow-up
          13
          ·
          8 hours ago

          No, not directly at least. Encryption relies on the difficult nature of factorisation, not knowing if a specific number can’t be factorised (is prime).

          • Possibly some of the internal components of this could be extended to help factorisation problems (but also possibly not)
          • Most modern encryption algorithms moved away from primes because of other weaknesses like quantum factorisation anyway
    • AppleTea@lemmy.zip
      link
      fedilink
      English
      arrow-up
      13
      arrow-down
      2
      ·
      9 hours ago

      So there are prime numbers, that are only divisible by 1 and themselves. Here’s a set of all the prime numbers bellow 100:

      [2, 3, 5, 7, 11, 13, 17, 19, 23, 29, 31, 37, 41, 43, 47, 53, 59, 61, 67, 71, 73, 79, 83, 89, 97]

      As you start getting larger and larger numbers, primes show up less and less. The Riemann hypothesis is that there is some pattern, some formula to the distribution of prime numbers.

      So if there is a pattern and you had the formula, you could predict which arbitrarily large number is going to be prime. Otherwise, you just have to find a big number and check if its divisible.

      • gibmiser@lemmy.world
        link
        fedilink
        English
        arrow-up
        8
        ·
        9 hours ago

        The other poster said a ton of math is built upon that idea? Like what and why?

        Sorry I know asking this is alot, im just curious.

        • loppy@fedia.io
          link
          fedilink
          arrow-up
          12
          ·
          8 hours ago

          The Wikipedia page lists various consequences. Mostly things that will not mean anything at all to a layman. The other thing is that claims that “all of modern mathematics” hinges on the Riemann hypothesis are extremely overblown; it’s mostly consequential to number theory, which is something many modern mathematicians work on/intersect, but there’s also many who never touch it and for whom the Riemann hypothesis being true or false changes nothing.

        • PetteriPano@lemmy.world
          link
          fedilink
          English
          arrow-up
          5
          ·
          7 hours ago

          We love using big primes for public key cryptography.

          That’s some black magic math that lets you take a pair of big primes. Keep one for yourself, give the other to the world. You can then encrypt or sign things with your number, and everyone with the public prime can decrypt or verify that you signed it. And vice versa, people can encrypt things that only you can decode.

          As I recall, it all relies on the fact that it’s a lot of work trying to find suitable primes for this. Knowing the pattern to how often primes occur can make it easier to break. This proof only predicts how many primes we’ll find, so it’s the basis for it.

        • AppleTea@lemmy.zip
          link
          fedilink
          English
          arrow-up
          3
          ·
          8 hours ago

          I wish I knew for sure, but I don’t.

          Best guess, I can say that a lot of mathematical profs work by looking for the simplest description, then building up from there. Primes are as mathematically simple as it gets, so its kinda important to keep in mind there’s “simple” numbers hiding out there in the unthinkably large digits.

    • palordrolap@fedia.io
      link
      fedilink
      arrow-up
      2
      ·
      3 hours ago

      Largely unrelated. Generating primes has always been relatively easy. All you need is a random number generator, a primality checker and the ability to keep adding one until the primality checker says it’s prime.

      Now if you want to count the number of primes below a certain number, you’re moving closer to Riemann Hypothesis territory, but I don’t think this result will help much with that either.

    • socsa@piefed.social
      link
      fedilink
      English
      arrow-up
      15
      ·
      8 hours ago

      Riemann isn’t a prime generator, it’s a prediction of how many primes exist in any contiguous sequence of integers.

      • credo@lemmy.world
        link
        fedilink
        English
        arrow-up
        1
        ·
        36 minutes ago

        What if you binary search on that sequence? How much computation is required to predict the number of primes?

    • addie@feddit.uk
      link
      fedilink
      English
      arrow-up
      8
      ·
      8 hours ago

      No more so than you could do before - we’ve been assuming that Reimann is true since forever, a lot of modern mathematics depends on it. And we already have ways to conjure up extremely large numbers that are almost certainly prime, such as Fermat’s primality test. RSA encryption depends on us being able to do that, in fact.

  • Starblender@normals.me
    link
    fedilink
    arrow-up
    3
    arrow-down
    37
    ·
    edit-2
    9 hours ago

    The nerds have had their run. Time to give the jocks another turn and see how things go.

    People who don’t need to put their dick everywhere because they already did… in high school.

    Or maybe the artists. Let’s have them run things.

    • Wildmimic@anarchist.nexus
      link
      fedilink
      English
      arrow-up
      2
      ·
      1 hour ago

      I really do not know what kind of person would comment such an ugly piece of text, but you should realize that ALL of this is the work of nerds.

      And not all of them have dicks.

      And some might not want to fuck everything that moves.

      And most of them have a completely normal social life the moment you remove the jocks.

      • tourist@lemmy.world
        link
        fedilink
        English
        arrow-up
        1
        ·
        7 minutes ago

        It’s an extremely bizarre comment

        If someone figures out how to set up a fediverse account, I hate to break the news, but this is 110% nerd activity. We’re all nerds here.

        What the fuck qualifies as a jock these days anyway? Does anyone still use that term unironically outside of a Hollywood writer’s office?

        You’re good at sportball? You learned all those sportball rules? You keep a catalogue of all the best sportball players? Good job, nerd.

        I seriously can’t even tell what point they’re trying to make