• GravityAce@lemmy.ca
      link
      fedilink
      arrow-up
      5
      ·
      1 year ago

      That’s some War Games shit…The CPU usage grows through each iteration of AI learning to post better reviews and AI debunking the better reviews it has posted until there’s no more room for anything else and the LLM becomes consumed with this one task of beating itself

      • LinuxSBC@lemm.ee
        link
        fedilink
        arrow-up
        4
        ·
        1 year ago

        True, but that’s also a well-known machine learning technique called adversarial training, often used in Generative Adversarial Networks (GANs) or when teaching a model to play games like chess or Go.

        • GravityAce@lemmy.ca
          link
          fedilink
          arrow-up
          1
          ·
          1 year ago

          With a game with simplish rules like Go, I think this would work. With something more complicated like language with implicit meanings and tones, I see AI driving off a cliff and learning bad things from itself to the point where the model needs to be trashed and redone