• RunawayFixer@lemmy.world
    link
    fedilink
    English
    arrow-up
    0
    ·
    9 months ago

    A large language model shouldn’t even attempt to do math imo. They made an expensive hammer that is semi good at one thing (parroting humans) and now they’re treating every query like it’s a nail.

    Why isn’t OpenAi working more modular whereby the LLM will call up specialized algorithms once it has identified the nature of the question? Or is it already modular and they just suck at anything that cannot be calibrated purely with brute force computing?

    • Paragone@lemmy.world
      link
      fedilink
      English
      arrow-up
      0
      ·
      9 months ago

      There are math-specific LLM’s, & coding-specific ones

      ( Yi Coder is one, which I’ve used to translate bits of code into some language I can sorta understand… Julia… I’ve been trying to learn programming for decades, & brain-injury can go eat rocks. : ), too.

      LM Studio has a search-function, so search for “math” in its models-search, & see what it comes up with.

      I’ve used such things to give me a derivative of some horrible equation NASA published decades ago, & then go finding an online derivatives-finder to check it with…


      The thing that kills me is that IT SHOULD BE CHECKED, dammit!

      ie: IF the LLM did some bullshit “arithmetic” on a column-of-numbers, THEN the regular code of the spreadsheet should

      1. display the function that the AI used, if any, &
      2. suggest the SUM() function, AND SHOW THAT-FUNCTION’S RESULT.

      This whole “LLM: take the wheel” idiocy … incomprehensible.

      DuckDuckGo’s AI is hit-or-miss, & sometimes it is stubbornly wrong: no correction gets through to it.

      _ /\ _

      • RunawayFixer@lemmy.world
        link
        fedilink
        English
        arrow-up
        1
        ·
        9 months ago

        One of the other replies said that: “1”+(2+3) is “15” in JavaScript.". So my last theory as to what was going on, was that the creator of the meme had as cell contents =“1”, 2 and 3. And then copilot used python code to sum those, not sum() which would have answered 5.

        But since the answer is a black box, who really knows. This blind trust that open ai+ms expect, makes it unusable for anything that needs to be correct and verifiable. Indeed incomprehensible that they think this is a good idea. I’ll have to try finding something better on lm studio the next time that I have a math problem, thanks for that tip.