• KeenFlame@feddit.nu
    link
    fedilink
    English
    arrow-up
    1
    ·
    3 days ago

    This is the infantile mansplaining-like phenomenon I was eluding to. You don’t know. You think that it is a compression algorithm. Another, that it is a stochastic parrot and yet another that it is a live being. It is rare except for the people that actually read and understand the code that people admit, that the inner workings of the neural net is still a complete mystery. Oh, we know things like that the k-space seem to emphasize phrases that are adjacent to but not necessarily part of output. We know that we can afford certain sidebands for orthogonal functions we also don’t understand, except that they improve this or that tendency. It is extremely not a compression algorithm for several reasons. The first one being the inability to reform the original message with confidence, or with the slightest hint of efficiency. We do not know how it works, and it is easy to spot emotional counterfactual discourse by the confidence with which they proclaim to understand how modern machine learning models work.

    • lime!@feddit.nu
      link
      fedilink
      English
      arrow-up
      1
      ·
      3 days ago

      i feel like there is a fundamental difference between “we don’t know why it chooses the coefficients it does” and “we don’t know how it works”. if anything, that’s what’s infantile. that, and accusing people of it. the fact that gzip can be used as a language model should be a pretty big indicator.

      • KeenFlame@feddit.nu
        link
        fedilink
        English
        arrow-up
        1
        ·
        15 hours ago

        No, there’s a difference between lying about how things work because you watched a youtube video

        • lime!@feddit.nu
          link
          fedilink
          English
          arrow-up
          1
          ·
          13 hours ago

          now who’s being infantile? i studied machine learning at university.

          • KeenFlame@feddit.nu
            link
            fedilink
            English
            arrow-up
            1
            ·
            10 hours ago

            Still, the people I was referring to…? You seem very upset, and hope your studies go well. I you won’t find a reason in any knowledge of any architecture that can explain how the models work. You should know this.

            • lime!@feddit.nu
              link
              fedilink
              English
              arrow-up
              1
              ·
              8 hours ago

              i finished those studies twelve years ago. and we did in fact go through the mechanisms of training and inference that language models are using, albeit in fewer dimensions. i have no idea where you are getting your information because it’s simply not true. it’s amusingly like the idea of bootstrap physics that acollierastro did a video on yesterday; the idea that we can’t boil it down to something fundamental so we shouldn’t even try.