• ArmokGoB@lemmy.dbzer0.comM
    link
    fedilink
    arrow-up
    2
    ·
    edit-2
    5 months ago

    Here’s some research on how much energy various machine learning models use.

    In 2021, Google’s total electricity consumption was 18.3 TWh, with AI accounting for 10%–15% of this total.

    Let’s call it 10% to make it seem as energy-efficient as possible. That’s 1.83 TWh a year, or about 5 GWh a day. An average US home uses 10.5 MWh a year. You could power 476 US homes for a year, and still have some energy left over, with the amount of energy Google uses on their AI-powered search in a single day.

    • Yprum@lemmy.world
      link
      fedilink
      arrow-up
      1
      ·
      5 months ago

      But then the problem is how google uses AI, not AI itself. I can have an LLM running locally not consuming crazy amounts of energy for my own purposes.

      So blaming AI is absurd, we should blame OpenAI, Google, Amazon… This whole hatred for AI is absurd when it’s not the real source of the problem. We should concentrate on blaming and ideally punishing companies for this kind of use (abuse more like) of energy. Energy usage also is not an issue in itself, as long as we use adequate energy sources. If companies start deploying huge solar panel fields on top of their buildings and parkings and whatnot to cover part of the energy use we could all end up better than before even.

      • ArmokGoB@lemmy.dbzer0.comM
        link
        fedilink
        arrow-up
        3
        ·
        edit-2
        5 months ago

        I agree that we shouldn’t blame the tools. I also believe the idea that generative AI can be used for good, in the right hands. However, denying the negative impact these tools have is just as disingenuous as saying that the tools are only going to be used by fat cats and grifters looking to maximize profit.

        Also, did you know that you can just mod random people? It doesn’t even ask you. You just wake up one day as a moderator.

        • Yprum@lemmy.world
          link
          fedilink
          arrow-up
          1
          ·
          5 months ago

          But is it the tool that has the negative impact or is it the corporations that use the tool with a negative impact? I think it is an important distinction, even more so when this kind of blaming the AI stuff sounds a lot to distraction techniques, “no don’t look at what has caused global warming for the last century, look at this tech that exploded over the last year and is consuming crazy amounts of energy”. And saying that, I want to make sure its clear, that doesn’t mean it shouldn’t be handled, discussed or criticised (the use of AI I mean), as long as we don’t fall into irrational blaming of a tool that has no such issue.

          I didn’t know about the mod stuff, but also not sure why you mention it, am I going to find myself mod of some weird shit now? X)

          • ArmokGoB@lemmy.dbzer0.comM
            link
            fedilink
            arrow-up
            1
            ·
            5 months ago

            But is it the tool that has the negative impact or is it the corporations that use the tool with a negative impact?

            Running machine learning models is extremely computationally-intensive. To my knowledge, it doesn’t scale particularly well when you have a bunch of users making arbitrary requests. The energy problem is mostly to do with the number of users, rather than the fact that it’s corporations doing it. This isn’t to say that big tech doesn’t create a bunch of other problems by controlling closed-source models.