• nymnympseudonym@piefed.social
    link
    fedilink
    English
    arrow-up
    1
    ·
    5 hours ago

    LLMs are just fast sorting and probability, they have no way to ever develop novel ideas or comprehension

    And how do you think animal brains develop comprehension…?

    • just_another_person@lemmy.world
      link
      fedilink
      English
      arrow-up
      5
      ·
      5 hours ago

      Animal brains have pliable neuron networks and synapses to build and persist new relationships between things. LLMs do not. This is why they can’t have novel or spontaneous ideation. They don’t “learn” anything, no matter what Sam Altman is pitching you.

      Now…if someone develops this ability, then they might be able to move more towards that…which is the point of this article and why the guy is leaving to start his own project doing this thing.

      So you sort of sarcastically answered your own stupid question 🤌

      • nymnympseudonym@piefed.social
        link
        fedilink
        English
        arrow-up
        1
        ·
        2 hours ago

        Animal brains have pliable neuron networks and synapses to build and persist new relationships between things. LLMs do not. This is why they can’t have novel or spontaneous ideation

        This Nobel prize winner seems to disagree with you.

        Neural nets do indeed learn new relationships. Maybe you are thinking of the fact that most architectures require training to be a separate process from interacting; that is not the case for all architectures.

        • just_another_person@lemmy.world
          link
          fedilink
          English
          arrow-up
          4
          ·
          2 hours ago

          From your own linked paper:

          To design a neural long-term memory module, we need a model that can encode the abstraction of the past history into its parameters. An example of this can be LLMs that are shown to be memorizing their training data [98, 96, 61]. Therefore, a simple idea is to train a neural network and expect it to memorize its training data. Memorization, however, has almost always been known as an undesirable phenomena in neural networks as it limits the model generalization [7], causes privacy concerns [98], and so results in poor performance at test time. Moreover, the memorization of the training data might not be helpful at test time, in which the data might be out-of-distribution. We argue that, we need an online meta-model that learns how to memorize/forget the data at test time. In this setup, the model is learning a function that is capable of memorization, but it is not overfitting to the training data, resulting in a better generalization at test time.

          Literally what I just said. This is specifically addressing the problem I mentioned, and goes on further to exacting specificity on why it does not exist in production tools for the general public (it’ll never make money, and it’s slow, honestly). In fact, there is a minor argument later on that developing a separate supporting system negates even referring to the outcome as an LLM, and the supported referenced papers linked at the bottom dig even deeper into the exact thing I mentioned on the limitations of said models used in this way.