This is extraordinary: "emissions by data centres needed to train and deliver AI services are estimated at around 3 per cent of the global total, close to those created by the aviation industry." - ThreadSky

louiestowell.bsky.social • 3 days ago

This is extraordinary: "emissions by data centres needed to train and deliver AI services are estimated at around 3 per cent of the global total, close to those created by the aviation industry."

Reposted from Catherine Fletcher

Using big LLMs for basic queries is 'like turning on stadium floodlights to look for your keys' - this is good on the AI environmental questions.

Comments

litterygeniarse.bsky.social•3 days ago

AI == Acquiring information

Our gullible political leaders are incompetent to judge the merits of the Emperor's new tech.

clarewrites.bsky.social•3 days ago

The bugbear here is that we don’t know quite how to solve aviation emissions yet. But absolutely could create enough renewables to make AI, at least environmentally, ok. Like if Labour wants us to become a centre for it, then they bloody well ought to do the renewables maths.

aefrankes.bsky.social•3 days ago

but who’s going to plan my child’s birthday party and my wife’s silly vacation for me?? is my arrogant abdication of human connection for my sole convenience not worth the cost????

phillyhart4.bsky.social•3 days ago

📌

zp39a.bsky.social•3 days ago

📌

louiestowell.bsky.social•3 days ago

Since I've shared a downer, here's an upper: a surprisingly cheerful story about Gen Z and reading, feat Lucas Maxwell (is he on here?) #KidLitUK https://www.ft.com/content/41fc29dd-082f-4a4a-97f6-1d8e67a74eda

tracyshopkins.bsky.social•3 days ago

He is @lucasjmaxwell.bsky.social!

louiestowell.bsky.social•3 days ago

Yay! Lovely to see you pop up in that piece, Lucas!

caseymckinnon.com•3 days ago

I hope they get sabotaged by activists sooner rather than later.

louiestowell.bsky.social•3 days ago

Governments certainly aint doing anything useful.

caseymckinnon.com•3 days ago

sandyofipswich.bsky.social•3 days ago

New versions of basic programs like Windows make it more and more difficult for ordinary users to avoid AI or even to delete unwanted data. We need more rigorous regulation of IT not less.

louiestowell.bsky.social•3 days ago

Oh blimey yes. I spend an irritating amount of time trying to disable AI, whether it's in search or Windows and it's nigh on impossible.

leewardmister.bsky.social•3 days ago

I can’t access the piece: does it distinguish LLM use from AI? (The former being a tiny fraction of the latter)

louiestowell.bsky.social•3 days ago

It does in terms of individual examples but doesn't break down the energy use figures like that - it talks about how hard it is to get that data.

seanedmund.bsky.social•3 days ago

The author is part of this project - currently looking at top 10 ML tasks on open source models and found a 62,000 fold difference in power required. Big issue (which they acknowledge) is that the main players simply aren't releasing their code so they're only as accurate as their dataset.

louiestowell.bsky.social•3 days ago

What he said!

seanedmund.bsky.social•3 days ago

I think it's a great project. I am v. much anti-AI for pseudo-creative work and all the ways it's inserted unbidden into life, but mainly pro-ML for massive data crunching in fields like healthcare and engineering. However this transparency is vital to weigh up the benefits if there are to be any.

ianbetteridge.com•3 days ago

His 3% isn't quite accurate – the IEA estimates that data centres use 1.5-2% of global energy, and that's *all* data centre usage, not just AI – AI accounted for around a quarter of that. BUT... the real issue, as he says, is growth in use...

ianbetteridge.com•3 days ago

20% of all the growth in energy use in advanced economies will be driven by AI – that's a big chunk, although not as huge as, say, heavy industries. The key question is can that growth be supported by renewables?

ianbetteridge.com•3 days ago

And, equally important, is it actually *necessary*? Whereas we need to transition transport from fossil fuels to electric, do we really *need* to ramp up data centre use? Probably not if the prime use is "making silly images to impress my friends".

louiestowell.bsky.social•3 days ago

If the use is anything generative/"creative"/search then NOPE. Thanks for the figure correction! Def worth writing to the FT to say that if you can be arsed.

ianbetteridge.com•3 days ago

Oh, probably doesn't need a correction - the difference between 2% and 3% is a rounding error :) And he's absolutely right that the key problem is growth. The AI industry claim that it will aid to reduce impacts elsewhere (by optimising use) but that's all about machine learning algos, not LLMs.

louiestowell.bsky.social•3 days ago

Oh cool. And my intense frustration is that we need to transition energy for our EXISTING needs (and reduce THOSE needs) not create NEW needs. (I realise I'm preaching to the choir here but grr)

ianbetteridge.com•3 days ago

(Good article with the numbers in here, btw: https://archive.is/WGJdk)

dayv.wtf•3 days ago

It's amazing how much money is being thrown into "AI" for no measurable gain. It makes the cryptocurrency industry look... almost normal

ainemccaul.bsky.social•3 days ago

I often wonder at the environmental impact of all those pictures of people's dinner.

deeplyclassical.bsky.social•3 days ago

📌

lisajallen.bsky.social•3 days ago

It's mind-boggling and really worrying. And yet no one is doing anything about it, it seems!

stevewfolds.bsky.social•3 days ago

I am more concerned about data centers potable water use. It is contaminated by chemicals to increase flow & heat transfer. Dumped, unusable for consumption or agriculture.

Comments

Posting Rules

Reply