• Ech@lemmy.ca
    link
    fedilink
    arrow-up
    19
    arrow-down
    3
    ·
    edit-2
    15 hours ago

    Llms can’t even do math. And odds on them figuring that “highly complicated” technology out before the bubble bursts seems low.

    • backalleycoyote@lemmy.today
      link
      fedilink
      arrow-up
      2
      arrow-down
      1
      ·
      9 hours ago

      Llms can’t even do math.

      Neither can a substantial amount of high school graduates in the US. Corporate is just hedging their bets that AI does slightly closer to accurate math and that meat machines are cheaper for physical labor than having to build robots. Besides, math is unimportant to capitalism. A company that loses billions a year is still worth investing trillions in.

    • wonderingwanderer@sopuli.xyz
      link
      fedilink
      arrow-up
      7
      arrow-down
      2
      ·
      15 hours ago

      It’s crazy that no one has integrated them with a graphing calculator, spreadsheets, symbolic logic software, etc., in order to increase their deterministic reasoning capacity.

      Like, human brains aren’t just the language centers, so I don’t know why the people trying to build an analogue haven’t done more than merely trying to make the linguistic capacity more complex…

      • errer@lemmy.world
        cake
        link
        fedilink
        English
        arrow-up
        10
        ·
        14 hours ago

        I mean the poster above you is wrong, they use math tools internally now when you ask math questions. Very obvious in Gemini. Yes the raw LLM trying to autocomplete the answer to a math problem is gonna be wrong but that’s not the way they are used to solve problems like that anymore.

        • sbv@sh.itjust.works
          link
          fedilink
          English
          arrow-up
          7
          ·
          14 hours ago

          The LLM has to choose to use the calculating tools. Gemini tried to do this one solo:

          4 + 2 + 2 + 2 + 1+ 2 + 0 = 15

          Tbf, it did four of these calculations, and 75% were correct.

      • 0ops@piefed.zip
        link
        fedilink
        English
        arrow-up
        6
        ·
        14 hours ago

        Some of them write and execute Python internally for stuff like that

    • ExtremeDullard@piefed.social
      link
      fedilink
      English
      arrow-up
      8
      arrow-down
      5
      ·
      16 hours ago

      It doesn’t matter what LLMs can and can’t do well: all they have to do is do stuff a person does well enough for the price of the tokens they consume to make that person’s occupation vanish if it’s cheaper - and it almost always is. The metric companies uses is cost, not quality. Because that’s how capitalism works.

      • 4am@lemmy.zip
        link
        fedilink
        arrow-up
        15
        ·
        14 hours ago

        The bills are coming in now and much like all other cloud computing, it suddenly becomes much more expensive to do anything useful once you are dependent on it

        It’s so easy to sucker capitalists with labor replacement schemes

        • TuringCompleteSocialist@lemmy.world
          link
          fedilink
          arrow-up
          3
          ·
          13 hours ago

          I also like folks mentality that “once they see the real price of AI everything will be back to normal”. As if the very same companies didn’t get baited into becoming dependent on American cloud service providers like AWS to find themselves later paying out the fucking ass for them once they are dependent on them. Guess what ? THEY ARE STILL PAYING FOR IT, after more than a decade of being swindled. They will do the same with AI.