AI's Electricity Use Is Spiking So Fast It'll Soon Use as Much Power as an Entire Country

Alex de Vries, a data scientist at Vrije Universiteit Amsterdam, found that AI companies could soon use as much power as a country worldwide.

Futurism
@petergleick on the other hand, the chatbots are also informationally poisonous, so there's that

@Jplonie @petergleick @kcarruthers

🤔 Not only not true, but also a wrong comparison on so many levels.

I guess if you consider a future where all cars need water cooled graphics cards to run offline. 🤷‍♂️

@Jplonie Image description:
One man labeled “BLOCKCHAIN'S POWER USAGE IS GOING TO
CRASH THE PLANET” stands at the front.
Another, feral-looking man labeled “AI” is creeping up to him as if to tackle or attack him.
@GreenSkyOverMe @Jplonie The other feral looking guy is btw Jason Momoa ;)
@hackbyte Thanks, I'm not able to tell because I'm faceblind

@GreenSkyOverMe Ah .... i don't know anyone else either.... But i totally loved his character as Ronon Dex on stargate....

So seeing this image .... there would be only two possible persons wearing this style of hair and beard. Kyle Hill and Jason Momoa.... Rechecking then was easy. ;)

Totally get you with face blindness (it's dysfunctional face memory for the most part on my side ;))

@hackbyte Well faceblindness = dysfunctional face memory :D
@Jplonie @petergleick @kcarruthers Yet way less nerds using this as argument against LLMs than they did against cryptocurrencies.
@Jplonie
Alt text:
Meme of Jason Momoa labeled AI about to tackle an unaware Henry Cavill labeled Blockchain's power usage is going to crash the planet.
@Jplonie This is even more true now that more and more chains move away from using energy consumption for validation đź‘€

@Jplonie @petergleick @kcarruthers That's a myth.

Only first generation blockchains were power hungry.

Newer ones working with "proof of stake" etc. are very efficient.

@petergleick

also, all the fresh water usage

@petergleick I guess even if they get the energy usage per unit down, there will then be even more incentive to expand AI because it will be cheaper!?
@petergleick
Maybe we should slow the AI trend down until we have solutions to limit AI electricity use....

@petergleick I'm believing that given the tremendous cost of each AI query, and the poor quality, that companies who offer these bots will burn through their cash and fold.

However, it really does seem like we are racing to consume and waste as much as possible before it all collapses. In the future, all of history will be be divided into BC (before collapse) and AD (after doom).

@petergleick Very similar to the situation of cryptocurrencies. Maybe less prone to use energy in countries where corruption/mafia/terrorism/war crimes are common. But same planet-burning problems.
@petergleick AI doesn't destroy the planet. Greedy CEOs who want to make a quick buck by abusing technology they don't need or understand destroy the planet.
@petergleick
AI is not destroying the planet, companies and systems incentivizing them to do this are.
@petergleick Once could hope that they'll counter this by heavily investing in renewable energy. And then if/when AI crashes, we'll be left with a massive surplus of renewables. Not optimistic though.
@petergleick We already know that an energy crisis would be imminent. In Matrix trilogy, humans were used as batteries so that the Matrix would continue to run. AI may never be as smart the Matrix, but if the theories were correct, humans alone would be enough to do it for them.
@petergleick such a human thing to over fixate on a negligible risk and ignore a fairly obvious and greater risk.

@petergleick Hmmm. Feels like some nuance is needed here. Microsoft have an explicit goal to be net zero by 2030 and negative beyond that to pay back their total carbon debt. Amazon Web Services, not as explicit, but they are pretty focussed on that goal also. That matters as a bit of context - and they should be challenged for sure.

And cloud server farms are MORE efficient than local run smaller scale data centres. It's one of the reasons to go to the cloud.
(Yes local energy mix big issue!)

@petergleick The paper is good, the article is not.

The paper points out that there are not enough GPUs in the world to satisfy the AI growth requirements that a naive "every Google search uses a large language model" would require. You don't have to be worried about LLMs burning down the world if you can't get there from here.

Further, Google is green energy neutral, while Microsoft and Meta are close behind. Amazon intends on getting there in 2025.

@petergleick You can't burn the planet with green energy either, because there are no emissions.

A disappointing article, I recommend reading the paper.

https://www.cell.com/joule/fulltext/S2542-4351(23)00365-3

@petergleick As predicted by Star Trek nearly 60 years ago with the M5
@petergleick - I've been thinking (and saying) this all along. The real question is why it's not immediately obvious to anyone that using arrays of GPUs for frivolous tasks is going to take a lot of electricity.
@petergleick This is really quite interesting. I have some friends and contacts who work on AI regulation across various govs around the world. Might have to share this with them!

@petergleick

The algorithms are unnecessarily long.

@petergleick I'm waiting for SkyNet (to use nuclear bombs on humanity).
@petergleick Use solar and other green energy.
@petergleick Even without batteries money could still be saved when the sun is shining not including initial set up (solar) costs.
The climate cost of the AI revolution

On the energy cost of Large Language Models, what their widespread adoption could mean for global COâ‚‚ emissions, and what could be done about it.

Wim Vanderbauwhede
@petergleick It's not AI or Blockchain that needs all the electricity, but GPUs. Same hardware for both techs, same senseless "must compute" impulse to generate nonsensical weird outputs.