The Aesthetic Rebellion

[davidrevoy@framapiaf.org]

@datarama @maswan Yes, it's a reference to the famous paper of 2021; "On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? 🦜"

And as a lover of animals, I agree, an insult to real parrots

[lord_tacitus@mastodon.social]

@davidrevoy

I'm in love with that last frame, gorgeous

[screwlisp@gamerplus.org]

@davidrevoy
Yes, yes! Next start tagging with #unix_surrealism !
@mjd

[prahou@merveilles.town]

@screwlisp @davidrevoy @mjd uncredited prahou method?!?!

[paculino@tech.lgbt]

@prahou @davidrevoy I'm starting to think that maybe I should wait for images to load before boosting, instead of trusting alt text.

[prahou@merveilles.town]

@paculino @davidrevoy great idea!

[screwlisp@gamerplus.org]

Around 15 years ago a popular deep learning topic was turning photographs into apparent brushwork similar to different artists. However photographs have extremely high entropy to begin with, I suppose. Whereas genai [] I guess benefits from producing low entropy images from a random noise seed. Revoy's art is (until now) very polished, meaning the entropy is quite subtle and easily misconstrued as being low entropy. cf. storing entropy in bold outlines.
@prahou @davidrevoy @mjd

[khm@hj.9fs.net]

why did you just post the first panel of david revoy's comic

[screwlisp@gamerplus.org]

@prahou @paculino @davidrevoy

Quick visual comparison.

[slothrop@chaos.social]

@davidrevoy This inspired me to paint something awful.

Come and get it, tin can:

https://chaos.social/@slothrop/117130100393897091

[woffs@fe.disroot.org]

@prahou @davidrevoy I'm looking dermaßen forward to this aardvarks underwater egg thing

[prahou@merveilles.town]

@woffs @davidrevoy i'm going to have to really draw it now, aren't I?

[odysseywestra@linuxrocks.online]

@davidrevoy hey, it's a good illustration on Goodhart's Law. Remember that being a thing during COVID. Especially when it came to employers tracking productivity using mouse movements. Then those same employees figured out how to fool that software by installing mouse signaling software or vibrators to their mice.

But then again, how long is it going to take until new LLMs begin to copy those kind of techniques we apply to our artwork? It could be a never-ending cat and mouse chase when it comes to genAI and us.

[floppysalmon@dice.camp]

@davidrevoy ngl this looks pretty cool tho.

[screwlisp@gamerplus.org]

@davidrevoy

@algernon did you have direct commentary on the effectiveness of e.g. iocaine ?

@lunareclipse @catsalad @realFedix

[prahou@merveilles.town]

@davidrevoy

[lunareclipse@snug.moe]

@screwlisp @davidrevoy @algernon @catsalad @realFedix iocaine is a completely different approach, feeding AI scrapers different data than people, the effectiveness is limited by your capability to detect bots and by the ability of the user of the data to detect the data is a tarpit

[screwlisp@gamerplus.org]

@lunareclipse
I suppose they are more different than I thought off hand. I recall nightshade's demonstration was modifying a picture of a cow with subtle visual cues that would have a bot scraper erroneously machine-label it as a leather wallet. In contrast I suppose @algernon 's iocaine is a tool for ensnaring and triangulating bot scrapers (or poisoning them, if you will).

Bemusing avian intelligence is a broad topic.

@davidrevoy @catsalad @realFedix

[davidrevoy@framapiaf.org]

@prahou

[davidrevoy@framapiaf.org]

@odysseywestra Hey, oh, I think AI can already copy these techniques, no problem, they are good parrots.
But what protect this type of style is what their userbase want when they prompt, and how the company design the invisible "master prompt" to provide only "high quality rendering, mastery, featured artworks" and all the label like that , that are pre-feeded to exclude low quality works.
So, it require a lot of prompt engineering to get a rendering like that, and that's a path, imo