• absentbird@lemmy.world
    link
    fedilink
    arrow-up
    3
    arrow-down
    2
    ·
    4 days ago

    It takes less power to run my local model than it does to play Baldur’s Gate 3, but I haven’t seen anybody shaming people for playing games. Not every LLM is a giant wasteful cloud provider, many are open source and self hosted.

    It’s kind of like saying all vehicles are gas guzzling enormous pickup trucks and therefore nobody should travel anywhere. Self hosting on a PC you already own is more like riding a bike in this metaphor.

    • MrSmith@lemmy.world
      link
      fedilink
      arrow-up
      5
      arrow-down
      1
      ·
      edit-2
      4 days ago

      Show me the percentage of AI prompts done on local models, and if it’s more than a rounding error, I’ll eat my hat.

      Also, was your local model trained locally?

      • absentbird@lemmy.world
        link
        fedilink
        arrow-up
        4
        arrow-down
        2
        ·
        4 days ago

        What percentage of trips are done on a bicycle?

        I’m working on software to help more people do it, but I fear that anti-ai sentiment has lost focus on the problem. Local models are super useful for assistance with code, writing, and all sorts of general tasks. I’ve been working on a tool that allows you to tell the computer what you want and it generates a command line prompt with an explanation of how it works.

        • MrSmith@lemmy.world
          link
          fedilink
          arrow-up
          4
          arrow-down
          1
          ·
          4 days ago

          Going back to my analogy, it doesn’t matter what you do with your special case of LLM tech. What matters is that most are using it in a destructive way. People do not care if you use your massive pick-up truck to cure cancer. People see a dumb pick-up truck, they assume you voted for Trump. So I’m pretty sure my analogy stands.

          And the answer for the training question is a “No”, I presume.

          • absentbird@lemmy.world
            link
            fedilink
            arrow-up
            3
            arrow-down
            2
            ·
            4 days ago

            Yes, I agree, but I think it does matter where we go from here. We could say all vehicles are bad, or we could focus on the source of the problem. Corporate AI is what’s using all the electricity and water, it’s what’s creating the worst issues.

            Most open source models don’t need additional training, they’re already plenty good for most plain language tasks and the weights are all free to use. Why would I waste power doing my own training when the public options are perfectly adequate?

              • absentbird@lemmy.world
                link
                fedilink
                arrow-up
                2
                arrow-down
                1
                ·
                edit-2
                4 days ago

                No, though I’m not really sure how it makes a difference. If I used a different model that was made using less resources what would be improved? Both already exist, using one over the other would not save any energy.

                Edit: To go back to the BG3 comparison, I don’t know how many resources were used to make the game, or what sources the developers pulled from. I play it because it’s fun and another person enjoying it doesn’t cost the world anything (except a little electricity)