• xmunk@sh.itjust.works
    link
    fedilink
    arrow-up
    76
    arrow-down
    2
    ·
    8 months ago

    Using another AI to detect if an AI is misbehaving just sounds like the halting problem but with more steps.

    • Match!!
      link
      fedilink
      English
      arrow-up
      41
      arrow-down
      2
      ·
      8 months ago

      Generative adversarial networks are really effective actually!

      • Natanael@slrpnk.net
        link
        fedilink
        arrow-up
        3
        ·
        8 months ago

        As long as you can correctly model the target behavior in a sufficiently complete way, and capture all necessary context in the inputs!

    • marcos@lemmy.world
      link
      fedilink
      arrow-up
      26
      arrow-down
      3
      ·
      8 months ago

      Lots of things in AI make no sense and really shouldn’t work… except that they do.

      Deep learning is one of those.