Post

@sundogplanets there's been some awesome reporting from @HighCountryNew on this as well, c.f. https://www.hcn.org/issues/57-8/the-wests-data-centers-suck-water-and-power/

@sundogplanets

Thank you for posting the link to this article.

It is thebest examination of the energy "complexity matrix" around public / commercial A.I. I have read.

I suspect that if the energy use per "cycle" is / 1000, the system will (almost) inevitably expand to consume the available resources in any case.
Its the normal race for incumbency and market dominance.
To expect otherwise is to walk past the SuperHighways of Tulips ... Its natural system behaviour, riddled with excess , Black Swans and cold correction.

Hardware e.g.
https://qant.com/press-releases/q-ant-and-ims-chips-launch-production-of-high-performance-ai-chips-establish-blueprint-for-strengthening-chip-sovereignty/

There are many others making reasonably sane claims around replacing the GPU model with other even less linear neural models .

RiscV tilts the transparency toward the greater good AND enables true LOCAL (and "edge") computes ... Which may also help...

https://www.sifive.com/press/new-x100-series-second-gen-intelligence-family

Perhaps some quantum computing breakthroughs will / maybe / not help with the energy usage ?

This is all way beyond my grokage!

@Tutanota @brave @protonprivacy

@sundogplanets I made this if you want to use it for any purpose at all

@sundogplanets Great article, thanks.
Apart from companies just allowing M$ to bolt on copilot to office 365 without thinking about it, I don't know anyone who is actually paying for any AI services. I'm with Ed Zitron, I don't see how the financials work at all.

@sundogplanets This is definitely the least-awful popular-audience article on generative AI power consumption I've seen.

I'm sure the fossil fuel industry is just *thrilled* by this trend of people thinking AI is to blame for climate change.

@sundogplanets

If you can dodge a wrench, you can dodge a ball.

Here's my wrench Sam. And, I get what everyone is saying. But you can't be in an echo chamber. I don't like the energy and resource uses myself.

Humans aren't tools. What is the energy cost of a human doing that same scenario? And, you need to start from the beginning as they did. 25 years and a lot more than 2.9kwh.

Horses aren't tools. They're horses. Used for hundreds of years, until technology changed their roles that humans used them for. They are still around, the world hasn't ended, but it's, you know, endless junk yards and wars. Not the best scenario.

The problem I see it is that science fiction is becoming a religion. The tech bros are picking out the parts of Ian M. Banks Culture series and ignoring all the inconvenient parts. Like 25,000 years into the future. And it's fiction.

And going back to what I said earlier, I don't care for the data centers myself. I believe that the money should be going towards Material Sciences for fusion containment vessels that last more than 10 seconds. It's then that we can go hog wild with this shit.

Anyways that's my brief liberal arts education take on it in this small space that I can barely articulate. Don't kill the messenger.

@sundogplanets

The statement that AI is unavoidable seems a bit optimistic. If you’ve run across any of the journalism of Ed Zitron, he argues that by simply reading the financial statements of these guys that none of the financials make any sense. People who use these systems are not exposed to the actual cost because they are being subsidized to the tune of 9 to 12 times I think.

Beyond this, even if you could have all the energy you need last I looked you can’t summon water from thin air.

@sundogplanets Thank you so much for giving this talk! ❤️ Do you plan to share any of it (e.g., a handout, slides, etc.) or at least the list of references??

@sundogplanets AI -
Being put under surveillance and control by the robots of our billionaire overlords - and all it costs is our air and our water and the survival of our planet.

They run through an example to show typical slop-AI use and make energy-use estimates:

You ask an AI model 15 questions about the best way to fundraise.

You then make 10 attempts at an image for a flyer, and 3 attempts at a 5-second video.

This is about 2.9 kilowatt-hours, enough to drive ~15 km in a typical EV. (They have lots of disclaimers here that this estimate depends on the LLM used and no companies would give them enough info. Sounds familiar! *cough cough* SpaceX *cough*)