• dreamkeeper@literature.cafe
      link
      fedilink
      arrow-up
      4
      ·
      5 hours ago

      It is funny to see the AI mindlessly mirror his word salad, but he didn’t exploit anything. The interview will just be rejected by whoever reviews the result.

      • bedwyr@piefed.ca
        link
        fedilink
        English
        arrow-up
        1
        ·
        2 hours ago

        The other thing though, is if we all posted real sentences that did not make sense, regularly, we poison the AI models going forward, and they don’t have the employees to sift through it all. It’s not my idea, there are groups into getting this to spread, it’s an easy, and frankly fun, way we can help fuck them.

        To have impact we need larger numbers, but even without it, who is to say which error stemmed from where? I mean it was already chiefly trained on reddit so we are half way there already!