Except that the stats are wrong/out of date. It's probably closer to 5ml than 500ml for the suggested usage. This stat is only ever going to get wronger -when people's phone's comfortably generate such responses it will be clear that it clearly doesn't get that hot.
the phone is the interface not the platform, although encouraging people who use AI off their phone to put it into a glass of water is actually good praxis
I meant local-running models. It won't be that long before a battery-powered model can do a decent 100 word email. Although you might consider the model training as embodied water-use?
I would definitely consider model training to be embodied water use and also very seriously who other than spambots needs something to write a 100 word email. just write the fucking email yourself!
Hasn't this always been an issue with data centres that people have ignored? The rapid acceleration of water usage due to AI has drawn many people's attention to a problem that has been building for years 🤔
It's almost as if people are fine with data centres using water for their Amazon shopping, or Facebook scrolling, or TikTok posting, or Netflix streaming...
Very funny, but the message I'm getting from it is that a query uses a few ml of water, as opposed to the litres that were used to produce the sheet of paper the poster was printed on.
Comments
https://bsky.app/profile/jksteinberger.bsky.social/post/3litxrmfsxk2k