Ritawho
828
49
12
More info here: https://mastodon.green/@gerrymcgovern/113169638815535363
And here
https://www.tomshardware.com/tech-industry/artificial-intelligence/using-gpt-4-to-generate-100-words-consumes-up-to-3-bottles-of-water-ai-data-centers-also-raise-power-and-water-bills-for-nearby-residents
Read about the cycle of water
https://oecd.ai/en/wonk/how-much-water-does-ai-consume
WoodyGoodman
StrandedonEarth
And how does thy compare to crypto mining?
NonWellKnownImgurian
https://youtu.be/brQLpTnDwyg?si=1NNZ34RqbeElGEgw
moonshadowkati
This article is suspect. It uses a lot of scary language but provides little context. For example, what happens to that water? Is it destroyed or unusable afterwards? Furthermore, is the power use of 164 people a lot in a country with 330 million people? What about other studies that show AI art and text use less power than human-created art and text? https://www.sciencedaily.com/releases/2024/04/240402140354.htm
moonshadowkati
"Results showed that Bloom is 1,400 times less impactful than a U.S. resident writing a page of text and 180 times less impactful than a resident of India.
In terms of illustration, results showed that DALL-E2 emits approximately 2,500 times less CO2e than a human artist and 310 times less than an India-based artist. Figures for Midjourney were 2,900 times less for the former and 370 times less for the latter."
Ritawho
How much energy & water did it take till DALL-E2 was able to draw anything ? Are you suggesting we should replace artists by DALL-E2 stolen work?
moonshadowkati
I was not speaking to ethical or artistic concerns, only energy use. AI artwork represents a tiny fraction of overall energy and resource use.
InfocalypseRising
AI sucks for a lot of reasons but this seems like it's probably exaggerated or cherry-picked somehow.
contextisimportant
Every power generated in an actively cooled power plant evaporates thus uses water and hopes the nature will deal somehow with the additional heat. And if AI needs extra power supply then we are talking about large scale.
taez555
"ChatGPT, in 1000 words, tell me how to save 3 bottles of water."
GenshiV
Listen here you little shit...
oozabooza
To save three bottles of water daily, adopt small changes in your routine. Shorten your showers by a minute or two, turning off the tap while brushing your teeth or shaving. When washing dishes, fill a basin with water instead of letting the faucet run continuously. Collect rainwater for gardening, and use a broom instead of a hose to clean outdoor areas.
oozabooza
These simple adjustments can significantly reduce your daily water consumption, adding up to three bottles or more saved. Small actions contribute to larger water conservation efforts, helping both the environment and your household water bill.
Merdock
I also like the idea of installing a recirculating line so you don't have to wait for the water to heat up, but there is a bit of a problem with it. A recirculating line increases energy usage, so if you aren't getting your energy from a green & renewable source you are likely trading one problem for another
HeresYourSauce
I don't want to come off as supporting AI here, but I do not believe the math in that article, and combined with the headline I think it's just wrong.
1. It's not "consuming" water, it's using it. The processing needed to reuse that water is virtually zero. When they're comparing it to farming or drinking, both require massive processes to reuse.
2. I can not imagine it needing 3 bottles of water per hundred words. You can run GPT on your own computer, which needs 0 bottles of water to run.
Ritawho
Please read about the whole cycle (3rd link), water evaporates and it ends somewhere else
"While the evaporated water still stays within our planet just like any other matter, it may go somewhere else and further contribute to the already uneven distribution of global water resources."
HeresYourSauce
I only see the one link, but going off that quote I still very much stand by my comments here. They're not talking about evaporating three bottles of water per 100 words. That's an absurd amount of energy.
To quote myself "the environmental costs need to be considered, but this article is sentimentalism designed to target people who hate AI already."
Ritawho
Hope it's not the river that supplies your town they're drying
moonshadowkati
The third link states that AI uses less than half of a 1 percent of American water. They estimate it may grow to half a percent by 2027 if America ends up hosting half of all AIs around the world.
HeresYourSauce
Cooling is a thing for data centers and the environmental costs need to be considered, but this article is sentimentalism designed to target people who hate AI already.
Ritawho
https://www.theregister.com/2024/09/06/datacenters_set_to_emit_3x/
contextisimportant
How do you cool hot water? You could use proportional heat pipes and or fans. Or just evaporate it. What is cheaper? The number of power plants having a closed water cycle for all water used for cooling is like zero. _You_ cannot imagine the cost of water for a ChatGPT task? And it can run on your PC? Well then they are so stupid needing a large scale computing center. If you claim your calculations are fine, then I have bad news for you. Btw: the water used to power your PC is above zero.
HeresYourSauce
They're not stupid for using a data center approach. One instance could run on my machine, but asking each consumer to install a program that's hundreds of gigs and needing 10gigs of ram at idle isn't remotely practical.
It's also way more electricity efficient. Data center hardware is almost always more energy efficient than consumer grade stuff.
Are you thinking of the water usage to gather the power, or cool the machines? The article seems to be separating water and electricity use. 1/2
HeresYourSauce
The water data centers use themselves is for cooling.
I don't want to come off as some climate denier or *gasp* a republican. The environmental costs of running a data center are real, and have real effects. My concern here is that sensationalist articles like this one make it easy to dismiss the entire argument.
Click bait and sensationalist articles are bad, even if the point they're trying to make is good.
contextisimportant
Well, my take from this article - in combination of another where the idea is proposed to reactivate Three Mile Island (/gallery/three-mile-island-may-reopen-aA8cG6T) - is, that AI needs huge data centers to perform - aka huge amount of electricity to run.
And then there is the question "how to calculate water and power consumption?"
To get clean water the facility needs electricity.
To get electricity you need - usually - water for evaporate cooling the power plant.
contextisimportant
And while I agree that click bait articles are bad for any topic - someone who wants to find a hair in a soup will find it. While ignoring anything around it.
manhawk
For #2, you are not running chat GPT on your computer. You are accessing chat GPT from your computer but it is being run from a data center somewhere. That's where the water/power/etc is being used.
I don't know if the 3 bottles is accurate, but to say "my computer ran this and my power consumption didn't increase" misses the point of how LLMs and SaaS works. The computation (product) is hidden from the user.
moonshadowkati
No, you can run actual LLMs on your computer (such as FreedomGPT).
HeresYourSauce
That wasn't what I was saying. You are correct that in general you run chatGPT on their server. You don't see any of the costs for that. They're hidden.
You don't *have* to run it on their server though, you can run it locally and if you do you can see its usage requirements yourself. It's obviously not in the 3 bottles per 100 words range.
(You still won't see the cost to train the model, but the article was specific 3 bottles per 100 words.)
manhawk
Hmm, interesting TIL. You download their package and then use it as-is without their model updates?
HeresYourSauce
I think so, looking at the tutorials. OpenAI used to make everything open so there's a ton of stuff floating around. Though I'll be honest I haven't tried it myself.
It also looks like in practice it's much better to use another model.
manhawk
Yeah, I can see that as a good way to set up a custom GPT that doesn't get messed up as Chat GPT changes versions...
Not that I have had any specific chat GPT update give issues to me specifically, but I don't use it for anything important.