An observation I've made about the usual apologists.

The first statement is not even wholly true. While training does take more, executing the model (called "inference") takes much, much more power than non-AI search algorithms, or really any traditional computational algorithm besides bogosort.

Big Tech weren't doing the best they possibly could transitioning to green energy, but they were making substantial progress before LLMs exploded on the scene because the value proposition was there: traditional algorithms were efficient enough that the PR gain from doing the green energy transition offset the cost.

Now Big Tech have for some reason decided that LLMs represent the biggest game of gambling ever. The first to find the breakthrough to AGI will win it all and completely take over all IT markets, so they need to consume as much as they can get away with to maximize the probability that that breakthrough happens by their engineers.

Yeah, I ran some image generators on my RTX2070, and it took a solid minute at full power to do it. Sure, it's not a crazy amount, but it's not like it's running on your iPhone.
- you can get a very small generator running on a modern phone if you want a grainy 400x400 piece of anime trash
- Nothing more than playing a game for that minute

You have more brians than any AI shill will ever be able to emulate.

brians
- You rang?
- I don't use autocorrect, because misspellings and misstypes are usually more legible than what ever 'other' word autocorrect picks.
  ... though I'm leaving this one in. A gaggle of Brians is better than any AI shit.
- Ah, Muphry's law at work

Tax pollution and pay the revenue as cash to poor people, it'll figure itself out

AI training is less energy than AAA game development of which dozens come out each year.

It's that something you have a source on, or do you just have a feeling about it?
- Gpt 4 had a training cost of $78 million. Gta5 cost $300 million. 4000 developers each with the latest GPU burning hundreds of watts per employee to create the assets. A rough estimate of 750watt pc, 4,000 developers, 8 hour a day, 300 days a year, 5 years = 36 giga watt-hours.
  https://en.m.wikipedia.org/wiki/GPT-4
  https://gamerant.com/rockstar-games-with-longest-development-cycles-ranked/ https://www.tweaktown.com/news/88292/over-6-000-people-worked-on-grand-theft-auto/index.html https://www.sportskeeda.com/gta/how-long-take-make-gta-5-a-brief-history-game-s-developmental-cycle
  Off topic: while trying to find my post where I did the math I discovered that Blue Morpho is the name of an AI company. God damn it. AI ruined my Venture Bros reference.
whataboutism
- Seems more like "putting things in scale" than "whataboutism." I'm not sure I agree with the premise, but I don't think it's whataboutism at all. Whataboutism would be "it's fine, because something else is worse," whereas I think the commenter is trying to say "it's not much, since it's less than something else that isn't much either."
- Is it wrong to point out hypocrisy? Is also it wrong to mock SUV owners who complain that buses burn gasoline/diesel?
Good point, we should probably stop doing that. Reject modernity, return to ASCII
- 我该如何用ASCII写这个句子？
sure, while we're splitting hairs let's ban all computers worldwide because they are the cause of the huge increase of energy needs.
Something that sucks is slightly less bad for the environment than something that also sucks. Wow very good argument
- People do bad things for the environment all the time. It's hypocritical to drive around in a big SUV (4080 gaming GPU) and mock people driving a minivan because cars are bad for the environment.
At least human beings get paid to do that and presumably at least some of the people playing the games enjoy the experience.
The concept of a video game also isn’t fundamentally built on consumer deception, either.
Hmmm, is this true? The fact that AI corpos are talking about standing up brand new power plants just to keep their GPU farms running makes me a little dubious of this claim. I don't remember Ubisoft ever being that desperate for wattage
- If you do the math, which I've posted here, you can see it's true. This is a situation where hundreds of SUV drivers are sitting in traffic, see a large bus bellow out a large puff of diesel smoke, and think, "Wow buses are bad for the environment."
  People are bad at math. They don't see their individual contributions add up to really bad things.

How much energy does AI really use? (zdnet) Seems like queries aren't that expensive, so I guess the enormous energy cost of AI must be mostly from training. I reckon this is why apologists try to minimize it.

That was a shit "article" IMO. "What is a data center" lol, and no figures or explanations. Or did I miss something when I speed read the SEO crap?
- Here's google's environmental impact report, then.

I know how machine learning "AI" works, and there, the training is the costly part, using it is close to free.

Could someone knowledgeable explain what's the difference with those more recent large language models used by ghatgpt etc. ?

Heard something about 3 mile island... Relevent if true.