#''basically the first layer of math thats why we have test datasets''

[bgezal]

the ai art debate that's been going on in the armpit of this website is so lame. oh yes it's so new and gangbusters to say that artists are useless and have no talents to offer the world and can be economically replaced by random number generators. also fundamentally untrue if you knew anything about commercial art besides, like, "uhh i bet amazon use dallee to generate concept art now" "i can get pink tiger furry from artbreerder which is basically the same as a fursona commission right" "an ai art won a fine arts competition lol get wrecked fine arteests bet you feel stupid spending all that money on oil paints now" "damn i bet ai art is gonna seriously overtake the modern fine art trade" and the more reasonable one of "broke people can use ml art to make backgrounds for their visual novels probably". a legitimate Statement recently made was "it must be so sad for you artists to have been made irrelevant by technology but that's just how progress is. sorry!" i know an art school grad currently designing [redacted doxxable material that requires conceptual consistency and careful thought towards the full sequential experience], sorry for being a fire scary luddite but uhm :/ Somehow I don't think they're out of a job yet.

am i just goalposts shifting and could ml theoretically produce a [redacted doxxable material] if fed enough preexisting work. yes. do i think any serious commercial enterprise is going to pick stochastic gradient descent as their artistic director. no. and if someone does im betting it goes the way of the NFT cartoon. this whole debate imo is a convenient easy cheap free way to take the general cultural temperature on What An Artist Is and literally nothing else.

for context, i am basically jaded to the entire ml field. nobody knows what the fuck is happening*, they tune their math models and cross their fingers and call it research. it is running billions of data points through thousands of parameters many, many times to "train" a model before it even does all that shit again to produce results on user inputs. a similar "computer does many processes very quickly to produce result" thing would be bitcoin mining, which we all know is Bad for emissions, and there don't exist environmental guidelines/oversight in the ML field yet afaik, so who even fucking knows if ML emissions are comparable or not!

in other words:

Omegalul

>"new configurations" while one of these models was halfway to reproducing a particular artist's SIGNATURE from being so closely trained,

and finally:

I Don't Really Care If You Reblog This But I Typed It Like A Half Hour Long Rant And Don't Think It's That Compelling Even If You Already Agree With Me So. If This Gets To Someone Whos Going To Try And Nitpick Me On How Much Of A Machine Learning Model We Currently Understand/Other Reasons A Contrarian Pulled From Their Ass That We Are The Human Machine I Will Kill Everyone Here