• Leviathan
    link
    fedilink
    arrow-up
    16
    arrow-down
    1
    ·
    2 months ago

    I don’t compare millions of pieces of data and choose the one that is statistically correct. I analyse a question and answer based off of my understanding of the actual subject.

    LLMs don’t understand shit, they text match and statistically poop out “correct” answers.

      • naeap@sopuli.xyz
        link
        fedilink
        arrow-up
        3
        arrow-down
        1
        ·
        2 months ago

        Wasn’t that the point?
        To show current LLM models aren’t the same system as our brain

        • pressanykeynow
          link
          fedilink
          arrow-up
          2
          arrow-down
          2
          ·
          2 months ago

          No, I meant

          I analyse a question and answer based off of my understanding of the actual subject.

          What does “understanding” mean here? Isn’t it analysing millions pieces of data and giving answer that is statistically correct?

          • naeap@sopuli.xyz
            link
            fedilink
            arrow-up
            4
            ·
            2 months ago

            Nah, and it isn’t what LLMs are doing now

            LLMs just act on tokens and those are their limit in resolution.
            That’s why they’re so bad when it comes to rhyming or counting letters.

            They’re not reasoning. And there is a difference between reasoning and giving it the statistical most probable next token - which doesn’t have a meaning whatsoever