• just_ducky_in_NH@lemmy.world
    link
    fedilink
    English
    arrow-up
    6
    arrow-down
    1
    ·
    4 days ago

    Didn’t read: paywall, but it seems to me that math would be the one thing we should EXPECT AI to excel at.

    • Treczoks@lemmy.world
      link
      fedilink
      English
      arrow-up
      9
      ·
      4 days ago

      Have you ever tried to do math with an AI? And I don’t mean summing up three numbers or something similarly trivial.

      Actually, math problems will probably show best that the current LLMs are just parrots with a large dictionary.

      • kalkulat@lemmy.world
        link
        fedilink
        English
        arrow-up
        4
        arrow-down
        1
        ·
        4 days ago

        I came to a conclusion about GPT (which is very good with English on most topics) when I asked it how many prime numbers, when divided by 35, leave a remainder of 6. It quickly and confidently said there were none. It hadn’t even tried. The correct answer (there’s a proof) is: an infinite number.

        Two months later it answered that there were 3. Closer … but no cigar

        • General_Effort@lemmy.worldOP
          link
          fedilink
          English
          arrow-up
          1
          ·
          3 days ago

          I’m too lazy to check it now, but here’s GPT-4o mini (via DDG). Maybe I’ll give Gemini a chance later, too.

          Elaborate the proof.

          Here is a high‐level outline of the proof that there are infinitely many primes in the arithmetic progression
           p ≡ a (mod m)
          whenever gcd(a, m) = 1. In your case a = 6 and m = 35, so gcd(6, 35) = 1, and the theorem then guarantees infinitely many primes ≡ 6 mod 35.

          1. Dirichlet characters
            • A Dirichlet character modulo m is a completely multiplicative arithmetic function
            χ : ℤ → ℂ such that
            – χ(n+m) = χ(n) for all n,
            – χ(n) = 0 whenever gcd(n, m) > 1,
            – χ(n) ≠ 0 (and in fact |χ(n)| = 1) whenever gcd(n, m)=1.
            • There are exactly φ(m) such characters (including the “principal character” χ₀, which is 1 on all n coprime to m, and 0 otherwise).

          2. Dirichlet L-series
            For each character χ mod m one forms the Dirichlet L-series
            L(s, χ) = ∑_{n=1}^∞ χ(n) n^(−s) , Re(s) > 1.
            This generalizes the Riemann zeta function (which is the special case χ = χ₀ with m=1).

          3. Euler product factorization
            Because χ is completely multiplicative, one shows for Re(s)>1:
            L(s, χ) = ∏_{p prime} (1 − χ(p) p(−s))(−1).
            This factorization links L(s, χ) to the distribution of primes in each residue class mod m.

          4. Nonvanishing at s = 1
            A key analytic fact (Dirichlet’s theorem) is that for every nonprincipal character χ (i.e. χ≠χ₀), the L-series L(s, χ) extends to a holomorphic, nonzero function at s=1. Meanwhile L(s, χ₀) has a simple pole at s=1, just like the Riemann zeta function ζ(s).

          5. Summing over characters
            Use the orthogonality relations of characters: for any integer n with gcd(n,m)=1,
            (1/φ(m)) ∑{χ mod m} χ(n) over all φ(m) characters = 1,
            and it’s 0 if n is not coprime to m.
            This lets you isolate those primes in a particular residue class a mod m by forming
            {χ mod m} χ(a)̄ · L(s,χ)
            (where χ(a)̄ is the complex conjugate of χ(a)). In the Euler product this picks out exactly the primes p with p≡a mod m.

          6. Logarithmic differentiation and Tauberian theorems
            One takes logarithms of the Euler products and looks at
            –(L′/L)(s,χ) = ∑{n=1}^∞ Λ(n) χ(n) n^(−s),
            where Λ is the von Mangoldt function. Summing these expressions over characters and using orthogonality isolates
            {p ≡ a (mod m)} p^(−s) ,
            plus lower‐order terms. One then applies complex‐analytic Tauberian theorems (or elementary partial summation arguments) to deduce that as x→∞
            #{ p ≤ x : p ≡ a (mod m) } ∼ (1/φ(m)) x / log x ,
            in particular infinitely many such primes exist.

          Conclusion
          Dirichlet’s theorem on arithmetic progressions (proven by exactly the steps sketched above) implies that for any m and any a with gcd(a,m) = 1, there are infinitely many primes p ≡ a mod m. Setting m = 35 and a = 6 shows there are infinitely many primes that leave remainder 6 upon division by 35.

          • kalkulat@lemmy.world
            link
            fedilink
            English
            arrow-up
            1
            ·
            2 days ago

            Very nice. Makes my generic LLM look like a lumpkin. I’d heard that mathers were impressed by some of them.

          • kalkulat@lemmy.world
            link
            fedilink
            English
            arrow-up
            1
            ·
            edit-2
            2 days ago

            One less formal way of looking at it is this: there are an infinite number of multiples of 35. starting with 35, 70, 105. Add 6 to each of the odd multiples. 35+6=41 (prime); 105+6 = 111 (prime). With an infinite # of candidates, you’ve gotta get to an infinite number of solutions (for some value of infinity!)

            As for that Dirichlet stuff, it’s way beyond any of the useful stuff I learned too.

            • General_Effort@lemmy.worldOP
              link
              fedilink
              English
              arrow-up
              1
              ·
              1 day ago

              Yes, but it obviously wouldn’t work for +5 or +7. I don’t think you can just assume that the number of primes wouldn’t converge to some finite number just because the number of candidates goes to infinity.

              Dirichlet’s theorem proves that. I have since looked it up, and that’s correct. I didn’t realize at the time that I was asking it to elaborate the proof for Dirichlet’s theorem. Whether the elaboration is correct is something I will pass.

              • kalkulat@lemmy.world
                link
                fedilink
                English
                arrow-up
                1
                ·
                22 hours ago

                No, I wouldn’t assume non-convergence either … NOR would I assume that that AI didn’t just grab that ‘high-level’ ‘Elaboration’ from some site … without a citation.

                (Very human … Lots of people use quotes to sound smart, hoping they’ll get away with it. LAWYERS! Ministers! Presidents, even! )

                • General_Effort@lemmy.worldOP
                  link
                  fedilink
                  English
                  arrow-up
                  1
                  ·
                  21 hours ago

                  The OP article doesn’t say it explicitly, but those mathematicians are getting paid.

                  The chatbots at duckduckgo don’t have search.

    • General_Effort@lemmy.worldOP
      link
      fedilink
      English
      arrow-up
      3
      ·
      4 days ago

      Not really. They are powered by language models, after all.

      Still, much of math is about manipulating symbols. So it is not mind-blowing that a certain understanding of math is there and can be improved on. But math also involves an understanding of space, or geometry. There I wouldn’t expect much.

      Anyway. My 2 top takeaways from the article:

      1. Mathematicians are hired to improve the major AI services.

      2. They are required to use E2EE (Signal) for communication so that their work doesn’t get picked up and used for LLM-training before the time is ready.