Changed my stance on AI art, post as much of it as you can in the places they train the AIs
@stavvers haha, i am so happy this is biting them in the arse.
@whangdoodler cannot wait until they start insisting AI art is tagged so the models can exclude it lol

@stavvers @whangdoodler

Some AI art is watermarked so AI recognises and ignores it.

I learned this in the same moment I learned that someone had created a tool to add the same watermark to any image.

@stavvers @JetlagJen taking watermarks out of AI art should be simple enough.
@stavvers @JetlagJen i mean, smudge something on it with what ever you have available or crop it out.
@whangdoodler @stavvers I think it was some kind of pattern that was overlaid across the whole image, which AI would recognise but the human eye couldn't distinguish. But, yeah, if there's an algorithm to add it, I wouldn't be at all surprised if someone could write an algorithm to at least obscure it enough that the AI couldn't recognise it.
@stavvers @JetlagJen @whangdoodler where might one find this tool?

@TatoGremlin @stavvers @JetlagJen @whangdoodler
Not sure if this is the same thing being talked about here, but check this out, it's called Glaze:

https://glaze.cs.uchicago.edu/

Glaze - Protecting Artists from Generative AI

@stavvers @whangdoodler They can suck a dick. They didn't bother to make it clear in the past, so 
@stavvers @whangdoodler unfortunately the EU regulation also propses that images should be tagged
@stavvers *everyone* who studied AI saw this coming.
@Tattie @stavvers
I'm pretty sure the Venn Diagram of people who study how AI works and people who market scams for profit is two unconnected circles

@stavvers

Omg this is awesome!

An IRL example of feedback bias.

Machine learning AI may output great results, but sometimes has a little mistake.

In that case, if it trains with results it outputed, it's possible that what it trained with has a mistake.

But since it doesn't know that it's a mistake, it trains with that faulty information and integrates it partially.

Then it makes another little mistake alongside the mistake it just learned from, and outputs worse and worse results.

@stavvers
Clearly, AI art needs a hand
@stavvers @paezha I dunno, sounds like the same process humanity goes through to get from Monet to whatever the eff it is that hangs in corporate lobbies.
@stavvers Just made the most Michael Jacksonesque hehe noise as I read this toot.
@stavvers AI art is like MSM on a no news briefing day!

@stavvers just to mention that: the same happens with AI generated text.

Shit in shit out. Thatโ€™s how it works.

@stavvers
LOL do you have any lรถรถps brother...
It's too bad this wasn't actually used as an artist tool, so the individual artist keeps their individual model updated with their own body of work to make references and drafts to expand upon.

People should just use it in mass to generate dank memes like baked beans in things until all images generated contain baked beans.

@stavvers this reminds me of a story on important if true from a few years ago wherein some researchers were making AIs negotiate with each other in English, but they accidentally started negotiating in some sort of incompressible but consistent variant on english syntax, but it takes some leaps to get to that conclusion imo lol
https://youtu.be/_e6OYlIXJ3Y&t=1098 18:18
Important If True 23: To Me, To Me, To Me

YouTube
@stavvers i think this is the way to kill LLMs too. But also kill the internet. Although that's mostly dead already.
Cyber Yuki (@[email protected])

Content warning: AI art degeneration: A prophecy.

hackers.town
@stavvers It may be one time accelerationism is justified.
@stavvers i hope it starts fixating on dogs again
@stx
If it does, it will be indistinguishable from me on a walk to my local park
@stavvers
@stavvers the inbred AI no fucking way๐Ÿ’€๐Ÿ’€๐Ÿ’€
@stavvers @tedivm How could everyone have foreseen this.
@stavvers interesting, a model that feeds on what it produces, a cycle that would be intriguing to witness.
@stavvers the thing about an ouroboros is it ends up consuming and producing its own waste

@stavvers when humans do the same, it's a style

If humans were to replicate typical artifacts of AI art, it would be called a style

But in any case, if this were to become an issue for commercial uses, they'd just roll back to data until 2021, and have some people manually insert new human art from decently reliable sources

As long as data is in their servers, they'll find a way