I'm Frankie, I play TTRPGs too much. I will be reblogging and/or posting a lot of furry arte and some of that's going to be kink stuff so heads up
AKA Nerts but that's going back a while


atomicthumbs
@atomicthumbs

the only ethical use of generative AI tools is elaborate shitposting. it won't put anyone out of a job, actually allows some innovation in the form (unlike most other uses of the tools), and actively costs the companies money for little or no benefit on their part, if not an active detriment


atomicthumbs
@atomicthumbs

generative AI imagery's specific powerful niche is "find the most remote corners of the network and edge cases that slide right between powerful concept clusters" to make something really fucked up that's an artifact not of the network's intent (epic sexy elon chad, standing on mars, ready player one, badass, corvette darkwave), but of its training and origins.

without generative AI we wouldn't have portrait photograph of homer simpson holding his dog or portrait photograph of peter griffin embracing homer simpson. and that's not just because a traditional artist wouldn't paint something like this, it's because I created the images with the intent of highlighting something about the network and tool itself. the pictures aren't about homer simpson, they're about the features the training algorithms extracted from the absolutely massive training dataset and interpreted as "homer simpson," and the effect of structuring a prompt in exactly the right way to cause some kind of horrifying photograph/homer metastability. it's about the way the network rings when you strike it hard enough.

it's built to optimize its generated image to maximally match it to two mutually exclusive ideas and makes something horrifying. and to me that's something truly fascinating. all it takes to cause model collapse is eight carefully selected words.

and that's also one reason generative AI sucks so much in general: the hucksters who are trying to sell it as a solution to anything are trying as hard as they can to get rid of the weirdest, most interesting parts of its output, because it makes them look bad


the-doomed-posts-of-muteKi
@the-doomed-posts-of-muteKi

I also would strongly encourage the use of generative ai in ways that attempt to highlight racial training biases, but that's less fun than photorealistic homer



You must log in to comment.

in reply to @atomicthumbs's post:

in reply to @atomicthumbs's post:

My favorite thing to do with ai is take a concept that mostly doesn't exist that could be easily conceptualized by a human person and see how far off it is. So far my favorite one is "crunchy hotdog". You'd think it'd be like burnt or something but the ai thinks "crunchy? frying things makes them crunchy ig" and just adds bubbles to it. It ends up just looking like a soapy hotdog