• Neato@ttrpg.network
    link
    fedilink
    English
    arrow-up
    14
    arrow-down
    2
    ·
    edit-2
    6 months ago

    It’s well known that generative AI cannot fucking spell. It doesn’t know how to because it wasn’t programmed with actual logic. It’s trained to find patterns and fill results with things that match the pattern. So it somehow deduced “um” and “fruit” both needed to be in the results and searched for that, but then it doesn’t know what actual fruits are and it doesn’t actually know how to spell so when a result turned up “coconut” it couldn’t tell it didn’t end in “um”.

    Edit: I went and tested it with copilot. oddly enough it could do the spelling with fruits but NOT with prime numbers. It wasn’t just a list of all primes, it just couldn’t spell numbers.

    • Barbarian@sh.itjust.works
      link
      fedilink
      arrow-up
      9
      ·
      6 months ago

      I went and tested it with copilot.

      The vast majority of memes like this get patched as soon as they become viral. Some developer somewhere very likely manually entered a list of fruits ending in “um” and got the LLM to refer to it when asked.

      • Neato@ttrpg.network
        link
        fedilink
        English
        arrow-up
        3
        ·
        6 months ago

        Hmm. I wonder where Lemmy is on the social media virality train. I think the flow used to be (roughly, many years ago): twitter, reddit, instagram, facebook…myspace.

      • Prox@lemmy.world
        link
        fedilink
        arrow-up
        3
        ·
        6 months ago

        Ehhh copilot is legitimately not shit. It’s actually downright usable in a lot of cases.

  • shininghero@pawb.social
    link
    fedilink
    arrow-up
    4
    ·
    6 months ago

    I said the same thing on my old kbin account, but I still believe strawberrum should be real, and should be 20% ABV minimum.

    • Wxnzxn
      link
      fedilink
      arrow-up
      3
      ·
      6 months ago

      All AI experts saying generative language models aren’t really AI are in fact AI themselves, trying to cover for AI preparing for full takeover!

  • xia@lemmy.sdf.org
    link
    fedilink
    English
    arrow-up
    2
    ·
    6 months ago

    You know, not many years ago this would have been amazing. Still not appropriate to adulterate search results with… but amazing nonetheless. Is google, like, a decade behind the curve?