• aurellence@lemmy.ml
    link
    fedilink
    English
    arrow-up
    109
    ·
    25 days ago

    “If you want AI that tells you when something is fake and where it came from, AI can do that,” he says. “Humans can’t, we’re not that smart.

    dead-eyed stare in librarian

    • Fizz@lemmy.nz
      link
      fedilink
      English
      arrow-up
      30
      arrow-down
      1
      ·
      25 days ago

      How could that even be true since it was humans thay trained ai. What a dumb statement ahh

      • shuffle3765@sh.itjust.works
        link
        fedilink
        English
        arrow-up
        1
        ·
        10 days ago

        When I was in college I built a chess playing program that I couldn’t beat. It wasn’t very good, but it was much better than me. The trick is machine learning: teach it the rules, get it to play millions of games against itself and adjust its strategy after each game.

        The AI can become better at a learnable skill than those who trained it because the AI can spend zillions of hours practicing

        • Fizz@lemmy.nz
          link
          fedilink
          English
          arrow-up
          1
          ·
          10 days ago

          Yes but thats not whats happening here. The base is created by using human data.

    • brsrklf@jlai.lu
      link
      fedilink
      English
      arrow-up
      12
      ·
      24 days ago

      AI detection by AI is not reliable. Lots of false negatives, and comparatively less but still too much reporting false positives to be reliable (and those are especially biased).

      It’s unlikely to get better too. If AI was able to reliably detect AI output, why not use that to correct itself to be less AI-like? And where is the seemingly infinite money thrown at in this struggle?

      • Passerby6497@lemmy.world
        link
        fedilink
        English
        arrow-up
        3
        ·
        24 days ago

        If AI was able to reliably detect AI output, why not use that to correct itself to be less AI-like?

        You just described adversarial training, something I’ve been hearing about for at least a few years now.