• nialv7@lemmy.world
    link
    fedilink
    English
    arrow-up
    9
    ·
    3 days ago

    A lot of the suggestions there are confusing and/or unhelpful

    Like

    hallucination → undesirable output

    We all know what AI hallucinations are, calling it “undesired output” just removes all the specificity and makes communication more difficult.

    • Echo Dot@feddit.uk
      link
      fedilink
      English
      arrow-up
      1
      ·
      1 day ago

      Actually I remember back when people start to calling it hallucinations there was quite a bit of pushback on it because it’s a complete mischaracterization of what’s going on.

      If you are performing a task and you make a mistake, that wasn’t hallucination, it was a mistake.

      When AI is hallucinate it’s not because they “think” the hallucination is real, it’s just that their weird math brain spewed out the wrong answer, for whatever reason, and now they are stuck with it and need to try and make some kind of coherent sentence out of it.

      • nialv7@lemmy.world
        link
        fedilink
        English
        arrow-up
        1
        ·
        1 day ago

        I remember people were trying to call it confabulation which is still anthropomorphic and it didn’t catch on

    • Imgonnatrythis@sh.itjust.works
      link
      fedilink
      English
      arrow-up
      2
      arrow-down
      1
      ·
      2 days ago

      Agreed. However, the concept of decreasing this anthropomorphization as much a spossiblr is really helpful. All these crappy articles about AI acting like some sort of nefarious human misdirects the real concerns and is just media fodder for click bait news outlets (which is pretty much all news outlets now). I genuinely think it’s a more news is good news for the AI companies.