Email or username:

Password:

Forgot your password?
Gerry McGovern

"ChatGPT consumes a lot of energy in the process, up to 25 times more than a Google search. Additionally, a lot of water is also used in cooling for the servers that run all that software. Per conversation of about 20 to 50 queries, half a litre of water evaporates – a small bottle, in other words."

AI is predicted to consume twice as much energy as the whole of France by 2030

Training GPT3, took 1,287 MWh (Megawatt hours) of electricity.

brusselstimes.com/world-all-ne

84 comments
Jo Jitsu

@gerrymcgovern it needs banning. And everyone using it for kicks, shits & giggles, can fuck off.

Jo Jitsu

@gerrymcgovern everyone using it to make their jobs ‘easier’ should lose their jobs.
Every employer implementing it should go financially bankrupt along with their already existing moral bankruptcy.

JoeBecomeTheSun

@JoBlakely @gerrymcgovern Ban this ban that. AI can be genuinely useful, it just needs to be uncensored and local without the large AI companies lobotomizing the llms from the beginning to prevent the general public from having a tool that is useful. AI shouldn't be banned, and since when is using electricity and water a bad thing? We should make electricity and clean drinking water cheaper and more accessible to lift people out of poverty.

Gerry McGovern

@JoeBecomeTheSun
I don't like banning either. All energy has a cost of the environment, though. No such thing as green, clean or renewable. Creating electricity will nearly always pollute drinking water, so you're solving one problem while creating another. For example, electrical infra requires huge amount of mining, and mining does incredible long term damage to water table.
@JoBlakely

JoeBecomeTheSun

@gerrymcgovern @JoBlakely And yet because of electricity we have the tools to mitigate the damage while enjoying the benefits. We have power tools and construction trucks because we have fossil fuels and electricity, we can use these power tools and construction trucks to build infrastructure that mitigates the damage, we can build desalinization plants to produce clean drinking water from the ocean, we can build nuclear power plants in the gigawatt range that have little to no carbon footprint and yet can recycle their own waste until there is no useful fuel left. All of this technology can be done today, and yet government wont let us and the environmentalists don't want access to cheap clean power or drinking water. They don't like the use of artificial fertilizers and GMOs that let us grow crops in suboptimal and bad soil that are resistant to pests and do not require poisonous pesticides and are more nutritious because they have been engineered to have higher nutritional value. Edit: there might not be a things as perfectly free clean energy, but perfect is the en enemy of the good. There is most definitely a thing as cleaner energy, cheaper energy and efficient energy. If america embraced nuclear power like france does we wouldn't need to import oil and gas prices would be cheaper.

@gerrymcgovern @JoBlakely And yet because of electricity we have the tools to mitigate the damage while enjoying the benefits. We have power tools and construction trucks because we have fossil fuels and electricity, we can use these power tools and construction trucks to build infrastructure that mitigates the damage, we can build desalinization plants to produce clean drinking water from the ocean, we can build nuclear power plants in the gigawatt range that have little to no carbon footprint and...

zrb

@gerrymcgovern and, to be clear, just because the water is evaporated into the atmosphere does NOT mean it is instantly reusable.

If you draw water from an aquifer faster than that aquifer's recharge rate, the sediment surrounding the reservoir will compact and subside, leading to a permanent capacity reduction. This is already happening to aquifers around the world:
nature.com/articles/s41467-023

zrb

@gerrymcgovern I've had tech bros in my replies saying data center water usage is "a nonsense stat" because the evaporated water "goes right back into the water cycle" qoto.org/@LouisIngenthron/1122

Gerry McGovern

@zrb
So much wrong with that sort of thinking. Water used by a DC is not available locally. Evaporated water will not stay local. DC water is wastewater with toxins and must be treated. Evaporation system leave major crud waste behind them.

It's hard to have to educate tech bros, who think they know everything, on really basic things.

⠠⠵ avuko

@gerrymcgovern @zrb so, eh, fuck tech bros? Let’s not post to educate those unwilling or unable to learn. I’m just happy (in a sad way) somebody brought this to our attention, and just another reason to refuse anything from the chatty Great Pile of Trash and its fraggly friends.

John Francis

@avuko @gerrymcgovern @zrb lol, the trash pile was pretty wise in FR, IIRC. It's a very generous comparison for LLMs

⠠⠵ avuko

@johnefrancis @gerrymcgovern @zrb 😂

You are right, I also remember it as a wise… creature(?).
But I could not resist.

🌹 Rose Bailey 🌹 🏳️‍⚧️

@gerrymcgovern Please don't take that as a challenge, whole of France.

(Or any other country.)

Nantucket E-Books

@gerrymcgovern I'll take this as a learning moment: the water evaporates... isn't it just getting returned to the water cycle?

Gerry McGovern

@saxicola
A human is using ChatGPT and that human needs to drink water whether they do the thinking or ChatGPT does it for them. So, we need to add the water AI uses and the water humans use while using AI. It's additive. And with each new tech innovation the water and material demand grows exponentially. @nantucketebooks

Saxicola ✅ 🏴󠁧󠁢󠁷󠁬󠁳󠁿🇪🇺

@gerrymcgovern @nantucketebooks
Ah but, I tend to sweat more in stressful situations as opposed to letting someone (or something) else do the work for me. :)
Ok, I appreciate the differential is minimal though.

Saxicola ✅ 🏴󠁧󠁢󠁷󠁬󠁳󠁿🇪🇺

@gerrymcgovern @nantucketebooks
This assumes the human is still alive after losing their job to AI and starving to death in an alleyway somewhere.

Acin on .art

@saxicola @nantucketebooks @gerrymcgovern

By my rough calculation, a human goes through about 4% of the water ChatGPT uses if the person or bot is answering a page of questions.

That's the same as what the Google search uses—1/25th of the evaporation for a chatbot.

Saxicola ✅ 🏴󠁧󠁢󠁷󠁬󠁳󠁿🇪🇺

@shadowfals @nantucketebooks @gerrymcgovern
Excellent! I love the people here, when you get proper (science based) answers to speculative questions.

Gerry McGovern

@shadowfals
There is nothing--so far--more efficient and economical than the human brain and body to do stuff. I did some calculations a while back and found that remembering something was so less polluting and energy intense than searching a search engine for it.

If you have to burn energy, burn your own energy.
Think first.
Remember first.

@saxicola @nantucketebooks

@shadowfals
There is nothing--so far--more efficient and economical than the human brain and body to do stuff. I did some calculations a while back and found that remembering something was so less polluting and energy intense than searching a search engine for it.

If you have to burn energy, burn your own energy.
Think first.
Remember first.

argv minus one

@gerrymcgovern

I don't know about your brain, but mine tends to lose stored information over time. As it ages, its storage capacity shrinks ever further.

I imagine this is why written language was invented. Words on paper don't fade so quickly.

@shadowfals @saxicola @nantucketebooks

argv minus one

@gerrymcgovern

As for search engines, the point of them is to find information I need when I don't know where to find it. How else would I do that? By walking up to random people on the street and asking them questions? I'd be lucky if none of them violently attacked me, let alone gave me useful answers.

@shadowfals @saxicola @nantucketebooks

Acin on .art

@chaosmonkey @saxicola

I started with how much time a human subject matter expert might take to write out quick answers like if chatting with a curious friend or a story writer. This is from my personal experience with authors, tech researchers, avid forums helpers, and educators—compared to what I know of ChatGPT. So, you know, it's a guesstimate.

For the comparable queries with at least chatbot-level accuracy: 15 minutes, or a quarter of an hour.

(...continued in next post)

@chaosmonkey @saxicola

I started with how much time a human subject matter expert might take to write out quick answers like if chatting with a curious friend or a story writer. This is from my personal experience with authors, tech researchers, avid forums helpers, and educators—compared to what I know of ChatGPT. So, you know, it's a guesstimate.

Acin on .art

@chaosmonkey @saxicola

I needed something quick for how much water the model person would use. That was "an average person needs eight cups of water a day".

8 cups / 24 hours = 0.3333 cups/hour

A quarter of that is 0.0833 cups / 15 minutes.

1.000 litre = 4.227 cups

0.0833 / 4.227 = 0.0197 litres

But I'd needed half a litre, not a litre in the conversion, so the number is doubled for 0.0394.

As a percentage, that's rounded to 4%.

Martijn

@saxicola @shadowfals I always welcome and appreciate open, honest, transparent and civil discourse / reactions based on facts and logic. 😀

The calculation has definitely given me food for thought.

Martijn

@shadowfals @saxicola I appreciate the calculation and by extension your transparency. ❤️

Personally I would probably have started with the premise of "the average person requires at least two litres of water per day" but that's what I'm used to.

I think I'll try out some similar calculations. Maybe create a site where people can compare energy expenditure between various things like "the average google search" or "three minutes of youtube"

Gerry McGovern

@nantucketebooks Data center water evaporation is a toxic process. And the water that evaporates is lost to the locality. Many data centers are in water stressed areas.There are many other water systems used by data centers, where the water is treated with a whole range of chemicals, becoming wastewater, which must be treated.

Chee_Koala

@gerrymcgovern So they do, what of it? Did they steal the power? Did they force anyone to use it? What's the message here: new technologies not as efficient as older technologies?

How would you specifically target AI to enforce lower power use?

Am I gonna have to explain to my electric company why I game on a cool PC or sometimes use an electric space heater?

Joel Falcou, Ph. D

@Chee_Koala @gerrymcgovern the ratio of energy used vs net result is horrendous.

AI is a pile of statistics bullshit that techbros don't understand and yet hype as the solution to inexistant problem.

Banning AI is a net win for everyone.

2xfo

@joel_falcou @Chee_Koala @gerrymcgovern

I tried to do some math to put this into terms people might actually grok

Bringing 10 grams of water from room temperature to vapor takes about 70 watt hours. So, every question you ask chatGPT is equivalent to leaving an old lamp on for an hour, or using a hair drier for about 3 minutes.

Gerry McGovern

@Chee_Koala
We are in the beginning of a global environmental collapse driven by the massive overconsumption of energy and materials causing the degradation of water, soil, air, biodiversity, and along comes AI with massive new energy and material demands. That's the so what. We're devouring our environment. To have any hope of avoiding catastrophic environmental collapse, we must massively reduce our energy and materials consumption. We must prioritize things that are truly useful.

Jan Ulrich Hasecke

@gerrymcgovern And we consume the earth to build the next global monopolies. AI will never act in the interest of the people but always to enrich the superrich.

Intel-Graphy \ - v - /

@gerrymcgovern If those servers are using too much energy to the point of overheating, then maybe these people could at least try to design better systems to withstand or even prevent overheating and excessive energy usage. But both tech companies and users just don't give a fuck cuz they want that sweet clout and bank.

Nonya Bidniss 🥥🌴

@gerrymcgovern Sickening. While it produces lies and garbage it's accelerating environmental damage exponentially. It really should be banned.

Dan Hoover

@Nonya_Bidniss @gerrymcgovern Just read an article about how AI will be the answer to greatly improving energy efficiency. Say what?? Maybe that article itself was a LLM creation?

P Stewart

@ddmgmgh @Nonya_Bidniss @gerrymcgovern it's probably a safe bet to assume "LLMs will solve insert-problem-here" takes are LLM output at this point, since that's mostly coming from either true believers or hyper-credulous tech 'journalists.'

Laikulo

@gerrymcgovern I'm quite surprised that swamp coolers are cheaper to operate than the more typical closed-loop cryo units. Without real numbers I have to assume it has to do with lack of externality cost, or some other strange economics. I know rejecting heat out of an exchanger in arid environments is less efficient than normal, but it's not that much...

MTW/HTW systems also get a little bit of advantage, because they're able to reject to ambient without quite as much effort.

All my experiences with closed-loop DC cooling systems, as my local area is not reliably dry enough to use a swamp cooler, therefore I can't really speak intelligently about systems that evaporate to the atmosphere.

My gut says that the cost of water should increase in these areas, eventually to a point where it's just too expensive to use lossy systems.

@gerrymcgovern I'm quite surprised that swamp coolers are cheaper to operate than the more typical closed-loop cryo units. Without real numbers I have to assume it has to do with lack of externality cost, or some other strange economics. I know rejecting heat out of an exchanger in arid environments is less efficient than normal, but it's not that much...

Chiclet

@laikulo @gerrymcgovern

That is what is missing from Gerry's post. Evidence that all ChatGPT servers are in DC's that use evaporative cooling. The servers used are all over the world, so "it depends" is going to be the most accurate answer.

Christine Schuhmann

@gerrymcgovern

District heating could reuse at least part of the the energy and condensate the water for reuse...

Miroslav Kravec

@gerrymcgovern some data centers use generated heat to heat up swimming pools.

I saw some news about such DC being built in Czech Republic.

But, yes, the power could be used for something more meaningful...

However, it's the same as the majority of the technology, the first iterations are very inefficient.

Take It EV Podcast 🎙️

@gerrymcgovern @lisamelton same can be said about using web browsers and web services to achieve things that your computer can already - like - most webapps have a desktop equivalent .

Remença

@gerrymcgovern chatGPT does more than google search. The comparison is off.

vetehinen

@remenca@mastodont.cat very true, a single google result page by itself does not get you much in many cases

@gerrymcgovern@mastodon.green

Abimelech B. 🐧🇩🇪 | SUSECON ™️

@gerrymcgovern Or summarized:
#challenge - write a program that calculates the area of a rectangular.
Junior dev takes 1 coffee and 3 minutes with a result of 12.
#machineleaning takes 5 TByte of data, 60 hours training with a result of 11.997459...

yopp

@gerrymcgovern on top of that I wonder how much energy each weekly update of 200mb+ Facebook app to 1 billion of people is consuming globally.

Carnildo

@gerrymcgovern Everyone's busy learning the wrong lesson from the logarithmic scaling of LLM performance. The correct lesson isn't "we need massive datacenters to gain a minor edge", the correct lesson is "a tiny little LLM that can run on a cellphone doesn't sacrifice much quality compared to a datacenter-spanning behemoth".

🇺🇦🇪🇺 cweickhmann

@carnildo
True. But it’s not about quality for the end user. It’s about control and access to input data for the vendor.
@gerrymcgovern

mos_8502 :verified:

@carnildo @gerrymcgovern I mean, if your LLM runs on a single solar panel and is trained only on public domain and properly licensed content, then I got no issue with it.

JoeBecomeTheSun

@gerrymcgovern "In the space of one hundred and seventy-six years the Lower Mississippi has shortened itself two hundred and forty-two miles. That is an average of a trifle over one mile and a third per year. Therefore, any calm person, who is not blind or idiotic, can see that in the Old Oolitic Silurian Period,\’ just a million years ago next November, the Lower Mississippi River was upwards of one million three hundred thousand miles long, and stuck out over the Gulf of Mexico like a fishing-rod. And by the same token any person can see that seven hundred and forty-two years from now the Lower Mississippi will be only a mile and three-quarters long, and Cairo and New Orleans will have joined their streets together, and be plodding comfortably along under a single mayor and a mutual board of aldermen. There is something fascinating about science. One gets such wholesale returns of conjecture out of such a trifling investment of fact." - Mark Twain

@gerrymcgovern "In the space of one hundred and seventy-six years the Lower Mississippi has shortened itself two hundred and forty-two miles. That is an average of a trifle over one mile and a third per year. Therefore, any calm person, who is not blind or idiotic, can see that in the Old Oolitic Silurian Period,\’ just a million years ago next November, the Lower Mississippi River was upwards of one million three hundred thousand miles long, and stuck out over the Gulf of Mexico like a fishing-rod....

JennyFluff :heart_trans:

@gerrymcgovern
if bitcoin taught as anything...

... ludicrous power consumption is not a reason that people with either stop using it or it being banned internationally

Woefdram

@gerrymcgovern Reminds me of something Morpheus said:

"We don't know who struck first, us or them. But we do know it was us that scorched the sky. At the time, they were dependent on solar power. It was believed they would be unable to survive without an energy source as abundant as the sun."

D2

@gerrymcgovern my house is far less useful to 7b souls than gpt3, uses on the order of 1-2 MWh each month, and that costs me a couple Benjamins. Is my math wrong or is 1287MWh under $150k of electricity? Hardly seems an existential risk.

Gerry McGovern

@cascheranno
"Global electricity demand from data centers, cryptocurrencies and artificial intelligence could more than double over the next three years, adding the equivalent of Germany’s entire power needs, the International Energy Agency forecasts in its latest report.“
bloomberg.com/news/articles/20

In 2024, China Water Risk estimated that China's data centers would use about 1.3 trillion liters of water annually, which is sufficient for 26 million people. 2030: 3 trillion liters

@cascheranno
"Global electricity demand from data centers, cryptocurrencies and artificial intelligence could more than double over the next three years, adding the equivalent of Germany’s entire power needs, the International Energy Agency forecasts in its latest report.“
bloomberg.com/news/articles/20

D2

@gerrymcgovern did you answer my question? I missed that. As a pro-nuc engineer in an area that gets a lot of protests, I’ve gotten used to trying to base claims when I don’t know source of data. So, rather than another ‘this article says these opponents say X’, I like validating data a bit directly. Is my math wrong?

Gerry McGovern

@cascheranno Your maths may well be right. Training is just part of the cost, though.

The IEA quote I gave you says that data centers may add a Germany's worth of electricity demand in the next 3 years. In 3 years. That's an incredible increase.

Other reference estimates China's AI water demand alone will be 3 trillion liters a year by 2030.

And it's only just beginning. Out of nowhere, AI is surging. Big Tech is super surging.

D2

@gerrymcgovern its a concern, but the estimates (and your new ones) aren’t what I’d call ‘well-documented’. I want iron-clad negatives when I rail against greed, wanton climate-damaging acts, etc.

VessOnSecurity

@gerrymcgovern The article confuses two very different things - the process of training an LLM, which is computationally intensive and uses a lot of energy but is done once, and the process of using the trained LLM, which isn't.

thestrangelet :fedora:

@bontchev @gerrymcgovern No, it doesn't, it calls out both, but doesn't confuse the two.

Scholablade

@gerrymcgovern

Utterly insane, I suggest people stop using these models, it's a lose-lose situation.

Gerry McGovern

@Scholablade Could we have an AI strike? That would be a wonderful thing.

Save the planet. Kill AI.

It's either AI or Nature.

Don't feed AI.

Scholablade

@gerrymcgovern

That would be indeed wonderful, unfortunately unless the value of labor gets heavily eroded, a strike doesn't seem likely, people lack solidarity with a "it's me vs everyone" even though we live in the same planet.

I think a good antidote to this mentality is workplace democracy, since everybody has a voice therefore everybody is responsible.

Martijn

@gerrymcgovern While I appreciate the concerns with regards to energy use & I personally feel that #AI is a serious #existentialThreat to #humanity..

I do wonder why some people feel the need to #insult others like "fuck tech bros" etc. Seems #unnecessary.

In the end most, if not all, #energy slurping services keep existing because of users. Including "social media" like X, Face**, etc and yes even #Mastodon & the #Fediverse with its glorious cat pics..

So I applaud efforts to reduce usage. ❤️

Gerry McGovern

@chaosmonkey Good points. When I talk about "tech bros", I do not mean all those involved in the tech industry, by any means. It is a particular subset best represented by the likes of musk, zuckerberg, altman, et al.

Yes, the users are the ultimate "end users", and there is definite responsibility there. However, the very best minds in Silicon Valley have for many years had a relentless focus: how best to manipulate and addict people using technology. It's hard as a user to resist.

DELETED

@chaosmonkey @gerrymcgovern

Techbros: “Look at what you made us do!! We were only just trying to help you poor losers post cat pics. Sure, we got rich from raping the environment, but we only did it for you!!”

Fuck techbros.

Martijn

@eldubuu @gerrymcgovern Wow... that's exactly the kind of post I meant.

I wasn't aware that in certain circles "tech bros" specifically refers to people like Musk, Zuckerberg etc.

Though I have no love for them, I do believe in being civil.

Also, a term like "tech bros" feels (to me and possibly others) like you're addressing a lot more than just Elon M. and Mark Z., including just you average IT worker.

FoolishOwl

@chaosmonkey @eldubuu @gerrymcgovern I've had to say before that tech bros differ from tech workers like bros differ from people in general or even just men.

Tech bros don't necessarily even have any technical skills.

Cybarbie

@gerrymcgovern "AI is predicted to consume twice as much energy as the whole of France by 2030"

Assuming nothing changes between now an 2030 which of course given the rate of progress in the past six months is absolute nonsense.

Frank Ring

@gerrymcgovern Instead of using ChatGPT, you can run your own AI locally on your computer. There's Chat with RTX and also gpt4all

Go Up