• context [fae/faer, fae/faer]
    link
    fedilink
    English
    441 year ago

    the day is approaching when we can ask an ai model to solve all of physics and it can actually create raving nonsensical rants claiming to be a grand unified theory of everything while denouncing the academic establishment for ignoring its genius, thereby automating the thankless task of giving physics grad students someone to punch down on

  • FnordPrefect [comrade/them, he/him]
    link
    fedilink
    English
    511 year ago

    That does sound impressive, but I’ll be really impressed when the AI model actually can tell me what the hell “solve all of physics” means

    Also, lathe-of-heaven No matter how “good” these models get, Douglas Adams has been too popular on the internet for this to ever succeed. All AIs based on data scraping will be forced to respond ‘42’ to all inquiries of that sort.

  • BodyBySisyphus [he/him]
    link
    fedilink
    English
    101 year ago

    New bit idea: whenever someone reaches a net worth in the 9 figures or above, they get a state-appointed scientist to follow them to meetings and interviews and loudly shout them down whenever they make an indefensible prediction (this is a lot nicer than my preferred bit idea)

  • DefinitelyNotAPhone [he/him]
    link
    fedilink
    English
    41 year ago

    At what point does this count as defrauding investors and gets Altman the one form of justice that rich people ever suffer under capitalism? This is so thoroughly bullshit that even investment bankers should be starting to see it.

  • Llituro [he/him, they/them]
    link
    fedilink
    English
    171 year ago

    my friend is doing gig work to make chatgpt better at just boring old textbook physics problems, and it’s complete dogshit at it. so uh, sure man. nice brainworms you got there.

    • space_comrade [he/him]
      link
      fedilink
      English
      41 year ago

      Didn’t they manage to make it somewhat good at solving certain math competition problems? Regardless it’s a pretty big jump from that to making a breakthrough in physics.

      • QuillcrestFalconer [he/him]
        link
        fedilink
        English
        31 year ago

        Yeah deepmind had good results with IMO problems, but only geometry problems. They scored almost at the level of gold medalist. That’s only a fraction of IMO problems, though. They did it by combining a formal verification system with a LLM to propose solution paths, and then doing some tree search I think.

        This is one way to improve large AI systems and will probably be incorporated in some way in the future, for example by integrating with a language like lean (for math proofs).

        They will also be improved by combining with tool use like calculators, code interpreters, web search, calendars, etc. This is already starting to happen to some extent.

        LLMs by themselves, at least with current architectures using transformers, are not great at reasoning (counting, arithmetic, symbolic reasoning)

      • Llituro [he/him, they/them]
        link
        fedilink
        English
        51 year ago

        maybe certain ones, but it’s generally bad about numbers and mathematical reasoning. he also gets paid to make it fail at math, and it’s arguably worse at basic math than physics.

  • Tom742 [any]
    link
    fedilink
    English
    41 year ago

    Why do we need AI when Terrance Howard already solved this by properly unfolding the flower of life