• deranger@sh.itjust.works
      link
      fedilink
      English
      arrow-up
      9
      ·
      5 months ago

      legislation in the works that mandates that companies that spend more than $100 million on training a “frontier model” in AI — like the in-progress GPT-5 — do safety testing. Otherwise, they would be liable if their AI system leads to a “mass casualty event” or more than $500 million in damages in a single incident or set of closely linked incidents.

      Are those models made by companies that would be affected based on the conditions above?

      • Grimy@lemmy.world
        link
        fedilink
        English
        arrow-up
        6
        ·
        edit-2
        5 months ago

        All models are very costly regardless of open source or closed source, but I’m not sure any current model reaches that high. The 100$ million seems to only applies to the cost of computing and not of buying the actual cards.

        The legislation is essentially asking that it can’t make nukes or do massive hacking attacking and only asking it of people that definitely have the money to make sure.

        It’s actually very level headed compared to what most are pushing for. I can’t even see it affect current gen AI, which are mostly harmless anyways.

    • Mnemnosyne@sh.itjust.works
      link
      fedilink
      English
      arrow-up
      3
      ·
      5 months ago

      Yup, exactly. The only regulation I’d be in favor of for AI is this: if it was trained on data which can be accessed by or was posted by the public, it must be freely available, such that if anything in the training data was posted online in a way anyone can see, then then I have free access to tge AI too.

      Basically any other regulation, even if the companies whine publicly, is actually one that benefits them by raising the barrier of entry and making it more expensive for small actors to create AI tools.