• h0tbeef@lemmy.zip
    link
    fedilink
    English
    arrow-up
    1
    ·
    3 days ago

    I should clarify - I do think there are use cases for smaller local models performing very specific tasks - the technology I am specifically referring to as “useless” are the massive data center LLMs.

    AI is such a blanket term (intentionally conflated with useful technology by the people poised to profit from said LLMs) that it’s a bit difficult to discuss it as a singular technology. I have no problem with machine learning, or with neural networks in general.

    I am, however, having difficulty reading this as a good-faith article

    “Lukasz Kaiser, one of the people who invented the Transformer, said on “Unsupervised Learning” in June:”

    That man (Lukasz Kaiser) is an employee of OpenAI, which is a massive conflict of interest imo, and disqualifies him from making impartial claims about the technology.

    What I find particularly sketchy is the vague use of the word transformer in this context. The transformer was invented in 1886, there are no living persons who helped to invent it. To present this man as the inventor of “the transformer” without clarification/disambiguation is disingenuous. It smells fishy to me.

    • unpossum@sh.itjust.works
      link
      fedilink
      English
      arrow-up
      1
      ·
      3 days ago

      Kaiser is one of the coauthors of Attention is all you need, the paper that introduced the Transformer architecture, the basis for all major LLMs.

      3b1b has a writeup: https://www.3blue1brown.com/lessons/attention/

      For example, imagine that the text we input was most of an entire mystery novel, all the way up to a point near the end, which reads:

      Therefore the murderer was…

      If the model is going to accurately predict the next word, that final vector in the sequence which began its life simply embedding the word was will have to have been updated by all of the attention blocks to represent much more than the individual word.

      It will have to have somehow encoded all of the information from the full context window that’s relevant to predicting the next word

      Attention is the mechanism that lets an LLM use the (correct parts of the) entire context to predict the next word.

      • h0tbeef@lemmy.zip
        link
        fedilink
        English
        arrow-up
        1
        ·
        3 days ago

        Oh yeah, I was able to figure out why the article said that with a couple search queries, but I still think that the phrasing in the article is misleading (and likely intentionally so).

        Edit: Also, I think part of what’s confusing people is that they keep using terms like “attention” and “inference” to describe computer processes that may or may not have some kind of underlying similarity to the corresponding human capabilities. I also believe this to be deliberate obfuscation.