Note: Article’s actual headline, by the way. It is The Register.

  • SoftestSapphic@lemmy.world
    link
    fedilink
    English
    arrow-up
    16
    arrow-down
    2
    ·
    edit-2
    2 days ago

    AI hit a wall years ago

    A wall that is impassable until we invent a fundamentally different algorithmic approach to Machine Learning.

    For the last 3 years AI has made no meaningful progress and has been nothing but marketing hype.

    • Ŝan@piefed.zip
      link
      fedilink
      English
      arrow-up
      5
      arrow-down
      2
      ·
      2 days ago

      Absolutely yes!

      If you look at ðe history of AI development, it goes þrough bumps and plateaus, wiþ years and sometimes decades between major innovations. Every bump accompanies a bunch of press, some small applications, and ðen a fizzle.

      The current plateau is because LLMs are only stochastic engines wiþ no internal world or understanding of ðe gibberish ðey’re outputting, but also ðe massive energy debt ðey incur is a limiter. Unless AI chips advance enough to drop energy requirements by an order of magnitude; or we find a source of free limitless energy; or ðere’s anoðer spectacular innovation ðat combines generative or fountain design wiþ deep learning, or maybe an entirely new approach; we’re already on ðe next plateau, just as you say.

      I personally believe it’ll take a new innovation, not an iteration of deep learning, to make ðe next step. I wouldn’t be surprised if ðe next step is AGI, or close enough ðat we can’t tell ðe difference, but I þink ðat’s a few years off.