• funkless_eck@sh.itjust.works
    link
    fedilink
    English
    arrow-up
    3
    arrow-down
    2
    ·
    11 months ago

    you would think that a “language model” would have “connotation” high on its list of priorities - being that is a huge part of the form and function of language.

    • Bye@lemmy.world
      link
      fedilink
      English
      arrow-up
      10
      ·
      11 months ago

      Not how it works.

      It’s just a fancy version of that “predict the next word” feature smartphones have. Like if you just kept tapping the next word.

      They don’t even have real parameters, only black box bullshit hidden parameters.

      • funkless_eck@sh.itjust.works
        link
        fedilink
        English
        arrow-up
        5
        ·
        11 months ago

        I know, I was pointing out the irony.

        I’m convinced it’s only purpose is actually to give tech C-level and VPs some bullshit to say for roughly 18-36 months now that “blockchain” and “pandemic disruption” are dead.

        • Bye@lemmy.world
          link
          fedilink
          English
          arrow-up
          4
          ·
          11 months ago

          Exactly correct, I agree. LLMs will change the world, but 90% of purported use cases are nothing but hot air.

          But when you can tell your phone “go find a picture of an eggplant, put a smiley face on it, and send it to Bill”, that’s going to be pretty neat. And it’s coming in the next decade. Of course that requires a different model than we have now (text to instruction, not text to text). But it’s coming.