Comment on Black Mirror AI

• lipilee@feddit.nl ⁨1⁩ ⁨month⁩ ago

  water != energy, but i’m actually here for the science if you happen to find it.

  source

  • EldritchFeminity@lemmy.blahaj.zone ⁨1⁩ ⁨month⁩ ago

    It can in the sense that many forms of generating power are just some form of water or steam turbine, but that’s neither here nor there.

    IMO, the graph is misleading anyway because the criticism of AI from that perspective was the data centers and companies using water for cooling and energy, not individuals using water on an individual prompt. I mean, Microsoft has entered a deal with a power company to restart one of the nuclear reactors on Three Mile Island in order to compensate for the expected cost in energy of their AI. Using their service is bad because it incentivizes their use of so much energy/resources.

    It’s like how during COVID the world massively reduced the individual usage of cars for a year and emissions barely budged. Because a single one of the largest freight ships puts out more emissions than every personal car combined annually.

    source
  • vivendi@programming.dev ⁨1⁩ ⁨month⁩ ago

    This particular graph is because a lot of people freaked out over “AI draining oceans” that’s why the original paper (I’ll look for it when I have time, I have a exam tomorrow. Fucking higher ed man) made this graph

    source
• xthexder@l.sw0.com ⁨1⁩ ⁨month⁩ ago

  Asking ChatGPT a question doesn’t take 1 hour like most of these… this is a very misleading graph

  source

  • vivendi@programming.dev ⁨1⁩ ⁨month⁩ ago

    This is actually misleading in the other direction: ChatGPT is a particularly intensive model. You can run a GPT-4o class model on a consumer mid to high end GPU which would then use something in the ballpark of gaming in terms od environmental impact.

    You can also run a cluster of 3090s or 4090s to train the model, which is what people do actually, in which case it’s still in the same range as gaming. (And more productive than 8 hours of WoW grind while chugging a warmed up Nutella glass as a drink).

    Models like Google’s Gemma (NOT Gemini these are two completely different things) are insanely power efficient.

    source

    • xthexder@l.sw0.com ⁨1⁩ ⁨month⁩ ago

      I didn’t even say which direction it was misleading, it’s just not really a valid comparison to compare a single invocation of an LLM with a continuous task.

      You’re comparing Volume of Water with Flow Rate. Or if this was power, you’d be comparing Energy (Joules or kWh) with Power (Watts)

      source

      • vivendi@programming.dev ⁨1⁩ ⁨month⁩ ago

        I can’t really provide any further insight without finding the damn paper again (academia is cooked) but Inference is famously low-cost, this is basically “average user damage to the environment” comparison, so for example if a user chats with ChatGPT they gobble less water comparatively than downloading 4K porn (at least according to this particular paper)

        As with any science, statistics are varied and to actually analyze this with rigor we’d need to sit down and really go down deep and hard on the data. Which is more than I intended when I made a passing comment lol

        source
• Sorse@discuss.tchncs.de ⁨1⁩ ⁨month⁩ ago

  What about training an AI?

  source

  • vivendi@programming.dev ⁨1⁩ ⁨month⁩ ago

    According to arxiv.org/abs/2405.21015

    The absolute most monstrous, energy guzzling model tested needed 10 MW of power to train.

    Most models need less than that, and non-frontier models can even be trained on gaming hardware with comparatively little energy consumption.

    That paper by the way says there is a 2.4x increase YoY for model training compute, BUT that paper doesn’t mention DeepSeek, which rocked the western AI world with comparatively little training cost (2.7 M GPU Hours in total)

    source