• LooseLemons@lemmy.world
    link
    fedilink
    English
    arrow-up
    2
    arrow-down
    8
    ·
    2 hours ago

    I think you should experiment with local models!

    Train them, embed some custom things, and see what happens!

    Compare the results to public services with the same prompts, etc.

    As for the semantics of “knowing”, I won’t engage there 😉

    • brucethemoose@lemmy.world
      link
      fedilink
      English
      arrow-up
      5
      ·
      1 hour ago

      Hot beef is right. That’s a good way to phrase it.

      And before you bring it up, I’ve been experimenting with local models since GPT-J and old GAN finetunes. I have MiMo 2.6 Flash loaded right now… TBH, I’m not sure how that’s relevant.

      • LooseLemons@lemmy.world
        link
        fedilink
        English
        arrow-up
        1
        arrow-down
        2
        ·
        35 minutes ago

        Are all of you seriously trying to argue that local experiments with model training data ARE NOT RELEVANT to our conversation about LLMs and their training data?

        Am I losing my fucking mind here or is the hive mind that strong? I’m all ears buddy.

        Hot beef is literally trying to argue this point. Are you? It is actually ridiculous.

        • brucethemoose@lemmy.world
          link
          fedilink
          English
          arrow-up
          1
          arrow-down
          1
          ·
          21 minutes ago

          I am not with the Lemmy hive mind consensus on LLMs, not even close. But we are discussing something more specific.

          • LooseLemons@lemmy.world
            link
            fedilink
            English
            arrow-up
            1
            arrow-down
            1
            ·
            14 minutes ago

            Yeah. I think people wildly misunderstand how these things work, then want to argue about them. It’s crazy.

            All I’m trying to say is that if I train AI on my mythical creature, the “Fijokopil” , a flying tiger ant, then the LLM will 100% be able to define a “Fijokopil”.

            And, ChatGPT won’t know what a “Fijokopil” is.

            I think I gotta take a break, it’s clear that my implications are going right over peoples heads.

      • LooseLemons@lemmy.world
        link
        fedilink
        English
        arrow-up
        1
        arrow-down
        4
        ·
        56 minutes ago

        You’re not sure how at home training is relevant when we’re discussing the output of LLMs? Or home sepecifc stuff?

        • brucethemoose@lemmy.world
          link
          fedilink
          English
          arrow-up
          1
          ·
          21 minutes ago

          Home inference.

          And that’s a very broad statement. But nothing we can do locally will get an LLM to intuit or reveal its own dataset.

          We can get it to output Claude traces it was trained on. I am not trying to be pedantic, but the semantics/phrasing there is very important.

          • LooseLemons@lemmy.world
            link
            fedilink
            English
            arrow-up
            1
            arrow-down
            1
            ·
            17 minutes ago

            Not byte for byte, but one can prove an llm was trained on data by asking specific questions. It’s why we’re having all of the copyright lawsuits.

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

      Why would I experiment with local models?

      I already know how to do the things I want to do. When I don’t know how to do something I want to do, I just teach myself. It’s not that difficult.

      I too would choose not to engage in semantic arguments if I found myself occupying such an untenable position; Fortunately I don’t often need to run that mental equation, being of sound mind.

      • LooseLemons@lemmy.world
        link
        fedilink
        English
        arrow-up
        1
        arrow-down
        2
        ·
        44 minutes ago

        After going through the process I suggested, you will see that your assumption earlier :

        “You do realize that, regardless of what output an LLM gives you, it doesn’t actually know anything right? ”

        Is really incorrect. You can train it on obscure and weirdly specific ideas that you came up with, and it will know about them.

        How do you think a model that is unplugged from the internet knows what Disney is? At some point, that word was in the training data.

        It’s that simple!

        If you applied even a tiny bit of critical thinking before replying, you would have figured that out yourself. And now that I’ve seen your lack of effort, and how disingenuous your responses are, it’s mute time baby!