made-in-akira/writings/mandarinlabs-iplex/chatlogs/chat 1 - concept

50 lines
3.7 KiB
Text
Raw Permalink Normal View History

2024-11-14 19:47:20 +00:00
2022.106 - Lounge (Scope: R2)
# [SERVER: 2022.106 16:31] fofo@HQ: connected to #lounge:r2
ado@OD: ahh, it gets my mind racing just thinking about it
ado@OD: say, this actually sounds like it could be a cool project
ado@OD: swing by optics and design some time if you want to do some brainstorming uwu
easter@MR: :o
# [SERVER: 2022.106 16:33] ado@OD: disconnected from #lounge:r2
akira@BA: gosh
akira@BA: is anyone else tired of machine learning projects in the vein of gpt and stable diffusion
nils@MR: the technology, or the energy around it?
akira@BA: the tech. it feels like a dead end to me
akira@BA: it takes so much power to get to a state that's remotely useful, and even then, it sucks
nils@MR: training takes more effort than you'd think
nils@MR: curating results, getting good sample data, probing for its limits...and losing hours of work when you leave it on overnight and the tunguska event happens to your model
akira@BA: yeah, that, precisely
akira@BA: because the machine wouldn't understand what it's doing. at best, you might make it fear god but--
easter@MR: in my imo the bigger problem is how profit incentives affected the technology itself
easter@MR: after dall-e mini, i feel as though theres been a gradual fizzle out into rudely biased "beauty" filters, and even the big image "generators" are founded on training data they didn't get permission to use
easter@MR: neither of which have had much real world application. as do many tchnologies in their infancy, but everything status quo would want out of ai - replacing stock image libraries, displacing customer support workers - is... you know, very useless to anyone else?
akira@BA: isn't that why we have goat room?
easter@MR: goat room isn't omnipotent! you of all people should know that. you were there last week working on the satellite!
easter@MR: we can protest with it, make some big waves even - remember the elon scandal - but it's difficult to change public opinion
easter@MR: especially the consensus held by the kind of people who care for machine learning.
akira@BA: sorry,
easter@MR: it's fine!
akira@BA: i still believe there is one field in machine learning hasn't really touched that would actually be worth a damn for society
akira@BA: ml models cut such such a big corner in the way their output is actually made
akira@BA: photorealistic illustrations aren't considered for how they would occupy a 3d space. drawings and paintings aren't done with layers and brush strokes. text isn't considered for its semantic or practical value
# [SERVER: 2022.106 17:00] floyd@YM: connected to #lounge:r2
akira@BA: it just takes noise and applies weights the humans told it were good to put ~something~ there
akira@BA: but what if it did consider these things
nils@MR: yeah, that's endgame - a real "artificial intelligence"
nils@MR: the problem is that the amount of resources that would take are entirely out of reach, even for the likes of beige alert and macrodata refinement
akira@BA: what stops us from playing god
akira@BA: what stops us from creating a being equal parts mechanical and biological
akira@BA: and giving them awareness? and the ability to comprehend
akira@BA: like, have you seen how much the human brain can do on less power than it takes to change a log by bolb
nils@MR: the law, parental responsibility. i guess all things do come to an end, so to give them a body prone to eventual failure isn't like horrifically rude but
fofo@HQ: Also. The watchful eyes of Joseph
nils@MR: what
akira@BA: huh?? who's that?
fofo@HQ: hahahahah
fofo@HQ: kidding, sorry
fofo@HQ: i'll see myself out. good night everyone
@floyd@YM: Soooo
@floyd@YM: Anyone down to go out for like, sushi or something tomorrow?
akira@BA: oh hell yeah
easter@MR: haven't had it in a while, so sure