AGI vs. Games: The Road to the ARC Prize 2024

แชร์
ฝัง
  • เผยแพร่เมื่อ 14 พ.ย. 2024

ความคิดเห็น • 6

  • @geogeo14000
    @geogeo14000 2 หลายเดือนก่อน

    Very interesting you deserve much more views. Exactly like the Arc challenge that should be much more popular !

    • @geogeo14000
      @geogeo14000 2 หลายเดือนก่อน

      Plus, very interesting that you mentioned how we evolved to get ours priors. My guess is that a potential breakthrough will come from "large evolutionary models" where agents will learn cooperate compete and reproduce in environments, solving tasks and learning how to generalize. Btw deepmind has developped algos where agents where learning games in a rich 3d environment with dynamic objective function and they were able to generalize pretty good on unseen games.

    • @FutureIsAmazing569
      @FutureIsAmazing569  2 หลายเดือนก่อน +1

      @@geogeo14000 Thanks a lot. Arc is really under the radar during the current AI hype

    • @FutureIsAmazing569
      @FutureIsAmazing569  2 หลายเดือนก่อน +1

      @@geogeo14000 evolutionary algorithms coupled with neural networks is such an interesting idea. But I feel that architecture itself needs to subjected to evolutionary pressure. E.g. we would modify DNA with mutations and crossovers, and that DNA could be used to grow slightly different architectures. Evolving into a recurrent network with memory would be totally viable. But of course resource wise this is a huge task.

  • @vitalijusbogomolovas2312
    @vitalijusbogomolovas2312 4 หลายเดือนก่อน

    nice video, could be more info on arc puzzles though

  • @ubit123
    @ubit123 4 หลายเดือนก่อน

    Interesting stuff about ARC, didn't know that AI is failing so miserable one these