• hendrik@palaver.p3x.de
    link
    fedilink
    English
    arrow-up
    2
    ·
    edit-2
    3 days ago

    Sorry, seems I’m a bit of a nitpicker in this thread… I think that’s somewhat of a misconception how AI isn’t deterministic. As far as I know the randomness is introduced in a later step. The AI models itself will give the exact same words as output for the same input words. They’re deterministic and that’s useful for some use-cases and how parallel-processing is done. Pretty much all nowdays AI works like that. It just switches it up with an added random number generator (temperature)… And well, floating point numbers sometimes do weird things. So different hardware might yield different results. At least that’s what I took from computer science. It’s more complicated, though. I think Deterministic AI is a phrase to distinguish some forms of AI from rule-based AI, which is a thing as well. That encompasses expert systems and -I think- that’s what scientists mean with deterministic AI… The calculations for probablistic AI however can be deterministic, too. And a computer processor will usually yield the same results for the same numbers anyway, unless it’s broken. But with that said, it’ll lead to issues with floating point numbers if we’re talking multiplayer games. I think we’ve had several instances when the gaming industry had to take care of such things to make multiplayer work, or replays be repeatable.

    • replicat@lemmy.world
      link
      fedilink
      arrow-up
      2
      ·
      3 days ago

      You’re correct it’s absolutely deterministic. LLMs and Image AI both use a random seed. So when the seed is set to X the result is always Y.