• Azzu@leminal.space
    link
    fedilink
    English
    arrow-up
    9
    ·
    7 hours ago

    I mean is it really slop anymore if it is super hard for humans? Translating one language into another representation is literally what LLMs are best at, they’re the right tool for the job in this specific case.

    • AppleTea@lemmy.zip
      link
      fedilink
      English
      arrow-up
      1
      ·
      1 hour ago

      “Brute force” might be a more accurate description of the methodology. Though, I’d say “slop” still applies given that nobody is realistically going to examine everything the program spits out. There could be tones of redundant and/or inelegant code.

    • standarduser@lemmy.dbzer0.com
      link
      fedilink
      English
      arrow-up
      4
      arrow-down
      1
      ·
      7 hours ago

      It’s jamming a bunch of things together in a sloppy way. It’s throwing a bunch of code it searches for that might do the trick. And sure keep it optimizing areas its asked to when its extremely specific. But most of this still is just that jammed stuff together. It’s not clean, it’s sloppy. Organization helps it, but that’s the human driving it, but again still sloppy

      • P03 Locke@lemmy.dbzer0.com
        link
        fedilink
        English
        arrow-up
        1
        ·
        1 hour ago

        If you have a solid set of tests, LLMs can do a really good job of getting accurate results. In this case, the machine code are the tests, since it would be expected to compile to something close to a 1:1 match between the console and PC versions.

        Sure, each processor is different, but they have very similar machine instruction sets, especially nowadays where an XBox is practically a glorified PC. There’s probably a rules file with a translation set.