• Artwork
    link
    fedilink
    English
    arrow-up
    2
    ·
    edit-2
    6 days ago

    Thank you, but I do not consider LLM equal to a calculator.

    The latter doesn’t normally have any feedback, and has a constant solid system, where you may always expect the result won’t change in time all of a sudden, and predict it. There’s a circuit and read-only memory of its flashed program looped.

    None of this is true in the context of the former - LLM. Here, an output may change each iterration due to the nature of LLM algorithms as “self-training”. The constant fear of the algorithm “plausible” mistakes, and it confidence in proving those are correct… is… unbearable…

    • Lvxferre [he/him]@mander.xyz
      link
      fedilink
      English
      arrow-up
      2
      ·
      6 days ago

      We’re comparing both tools in different aspects. What you’re talking about is predictability (and accuracy); in these I agree a calculator is way better.

      My point was about our reliance on tools, at the expense of deteriorating our own skills. Use calculators exclusively for long enough, and you’ll have a hard time with simple maths; rely too much on a kitchen scale, and you’ll lose the grasp on the right amounts of ingredients (or how to measure wheat flour without one); get used to an electric screwdriver and you’ll never know if you screwed it too little or too much. It may or may not matter, depending on someone’s profession, but I think some skills are worth kept alive and “un-rusty”, and I feel like the way I was using large models was rusting my skills.