• ylai@lemmy.mlOP
      link
      fedilink
      arrow-up
      5
      ·
      edit-2
      4 months ago

      In the case of Google/DeepMind’s SIMA it is an instruction-following agent for simpler, but menial tasks in a game. It is particularly not autonomous, and has no notion of reward. And what is being used is a modified behavior cloning with a text-conditioned policy network.

  • ReakDuck@lemmy.ml
    link
    fedilink
    arrow-up
    2
    ·
    edit-2
    4 months ago

    So what. It can play “games it had never seen before” but at the same time “by watching human players”?