• Sunny' 🌻@slrpnk.net
    link
    fedilink
    arrow-up
    14
    ·
    6 months ago

    It’s crazy how bad d AI gets of you make it list names ending with a certain pattern. I wonder why that is.

    • bisby@lemmy.world
      link
      fedilink
      English
      arrow-up
      11
      ·
      6 months ago

      I’m not an expert, but it has something to do with full words vs partial words. It also can’t play wordle because it doesn’t have a proper concept of individual letters in that way, its trained to only handle full words

      • Swedneck@discuss.tchncs.de
        link
        fedilink
        arrow-up
        3
        ·
        6 months ago

        they don’t even handle full words, it’s just arbitrary groups of characters (including space and other stuff like apostrophe afaik) that is represented to the software as indexes on a list, it literally has no clue what language even is, it’s a glorified calculator that happens to work on words.

          • Swedneck@discuss.tchncs.de
            link
            fedilink
            arrow-up
            1
            ·
            6 months ago

            not really, a basic calculator doesn’t tend to have variables and stuff like that

            i say it’s a glorified calculator because it’s just getting input in the form of numbers (again, it has no clue what a language or word is) and spitting back out some numbers that are then reconstructed into words, which is precisely how we use calculators.

    • blindsight@beehaw.org
      link
      fedilink
      arrow-up
      5
      ·
      edit-2
      6 months ago

      LLMs aren’t really capable of understanding spelling. They’re token prediction machines.

      LLMs have three major components: a massive database of “relatedness” (how closely related the meaning of tokens are), a transformer (figuring out which of the previous words have the most contextual meaning), and statistical modeling (the likelihood of the next word, like what your cell phone does.)

      LLMs don’t have any capability to understand spelling, unless it’s something it’s been specifically trained on, like “color” vs “colour” which is discussed in many training texts.

      "Fruits ending in ‘um’ " or "Australian towns beginning with ‘T’ " aren’t talked about in the training data enough to build a strong enough relatedness database for, so it’s incapable of answering those sorts of questions.

    • Even_Adder@lemmy.dbzer0.com
      link
      fedilink
      English
      arrow-up
      5
      ·
      6 months ago

      It can’t see what tokens it puts out, you would need additional passes on the output for it to get it right. It’s computationally expensive, so I’m pretty sure that didn’t happen here.

        • adderaline@beehaw.org
          link
          fedilink
          English
          arrow-up
          1
          ·
          6 months ago

          it chunks text up into tokens, so it isn’t processing the words as if they were composed from letters.

      • Jesusaurus@lemmy.world
        link
        fedilink
        English
        arrow-up
        1
        ·
        6 months ago

        With the amount of processing it takes to generate the output, a simple pass over the to-be final output would make sense…