Discussion
@aud@spectre.asta.lgbt Not particularly? I just remember that back in 2022 I couldn't get ChatGPT to take documentation and spit out proper ffmpeg commands, and to this day Claude still can't do it if I don't tell it to create a sandbox and test what it's doing.
Which is totally silly. It takes it like 30 minutes to build a script I could make in 5.
@butchrobot@shork.nerdmusic.net i am going to nitpick on this:
the water doesn't fucking matter. it is a fuzzy second-order waste effect of the primary waste: turning energy into heat. them getting you to talk about water is them succeeding in making you care about one of the least important impacts.
if you build a datacenter that uses the equivalent of 20 atomic bombs of thermal energy in electricity per day, of course you're going to lose some water from all of it evaporating
@linear@nya.social Sure, I suppose you can treat the water wasting as a metaphor and catchall for all that's wrong with AI.
And we are certainly due for a narrative shift in this regard.
What would you consider the most pernicious aspects of LLM's and data centers?
Install Fedizen
Get the full app experience
For the record, I do not use AI that often. I test it out every so often just to confirm the hype, if only because I'm a programmer.
Is that bad? Of course I'm wasting water... not nearly as much as the vibe coders, but plenty still. And for that, I'm sorry.
I do self host some AI models and have them do the occasional thing for me -- I had Mistral spit out some test cases for a leetcode problem I was doing recently. That was cool.
IMO the damage is already done with self hosted models, since they're already trained? And I use a Tesla P40 that I got from an IT refurbisher, so I essentially saved this GPU and rack from a landfill.