 "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. 84 comments
 @gerrymcgovern everyone using it to make their jobs ‘easier’ should lose their jobs. @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. @JoeBecomeTheSun @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: @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" https://qoto.org/@LouisIngenthron/112231723658590505 @zrb It's hard to have to educate tech bros, who think they know everything, on really basic things.  @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.  @avuko @gerrymcgovern @zrb lol, the trash pile was pretty wise in FR, IIRC. It's a very generous comparison for LLMs @johnefrancis @gerrymcgovern @zrb 😂 You are right, I also remember it as a wise… creature(?). @gerrymcgovern Please don't take that as a challenge, whole of France. (Or any other country.)  @gerrymcgovern I'll take this as a learning moment: the water evaporates... isn't it just getting returned to the water cycle?  @nantucketebooks @gerrymcgovern @saxicola @gerrymcgovern @nantucketebooks @gerrymcgovern @nantucketebooks @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. @shadowfals @nantucketebooks @gerrymcgovern @gerrymcgovern @shadowfals @nantucketebooks  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. @argv_minus_one @gerrymcgovern @shadowfals @nantucketebooks 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. @chaosmonkey @shadowfals @nantucketebooks @gerrymcgovern 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%. @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. @chaosmonkey @shadowfals @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" @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. @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? @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.  @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. @Chee_Koala @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. @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.   @gerrymcgovern Sickening. While it produces lies and garbage it's accelerating environmental damage exponentially. It really should be banned. @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? @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.' 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. District heating could reuse at least part of the the energy and condensate the water for reuse...  @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.  @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 .  @remenca@mastodont.cat very true, a single google result page by itself does not get you much in many cases @gerrymcgovern Or summarized:  @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. @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".  @carnildo  @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. @gerrymcgovern ... ludicrous power consumption is not a reason that people with either stop using it or it being banned internationally @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." @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. @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? @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. @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.  @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. @bontchev @gerrymcgovern No, it doesn't, it calls out both, but doesn't confuse the two. @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. 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. @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. ❤️ @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.  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. @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.  @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.  @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.  @gerrymcgovern Instead of using ChatGPT, you can run your own AI locally on your computer. There's Chat with RTX and also gpt4all |
Gregory's Server
@gerrymcgovern it needs banning. And everyone using it for kicks, shits & giggles, can fuck off.