• db0@lemmy.dbzer0.com
    link
    fedilink
    English
    arrow-up
    14
    ·
    1 day ago

    Actually if they can use that to make good ai opponents for 4x games, it’d be pretty sweet

    • Naia@lemmy.blahaj.zone
      link
      fedilink
      English
      arrow-up
      3
      ·
      20 hours ago

      Except you don’t need a neural net to do this, nor would you really want one for it because it would be incredibly slow unless you use the GPU to run it which might be a bit busy running the actual game. It’s like a “make game run worse” option.

      In game AI can be pretty sophisticated even without the addition of neural nets which could be OK with determining strategy based on player action, but really isn’t going to be good for the entire control.

      This kind of thing is only useful as tech demo for playing platformers and the like with the same inputs a human player would, basically more like a bot.

      • db0@lemmy.dbzer0.com
        link
        fedilink
        English
        arrow-up
        2
        ·
        20 hours ago

        In game AI can be pretty sophisticated even without the addition of neural nets which could be OK with determining strategy based on player action, but really isn’t going to be good for the entire control.

        Given that it has been the weak point in almost all 4x games in the past 30 years, I’d be happy if they give anything novel a try

        • Naia@lemmy.blahaj.zone
          link
          fedilink
          English
          arrow-up
          1
          ·
          2 hours ago

          I could see a way for it to work, but this tech certainly isn’t it and you’d need to balance the size of the model for memory use and performance.