This new data poisoning tool lets artists fight back against generative AI
The tool, called Nightshade, messes up training data in ways that could cause serious damage to image-generating AI models.
https://www.technologyreview.com/2023/10/23/1082189/data-poisoning-artists-fight-generative-ai/
The tool, called Nightshade, messes up training data in ways that could cause serious damage to image-generating AI models.
https://www.technologyreview.com/2023/10/23/1082189/data-poisoning-artists-fight-generative-ai/
Comments
https://www.reddit.com/r/aiwars/comments/168d010/glaze_is_kinda_bad/
https://bsky.app/profile/nafnlaus.bsky.social/post/3kcinns6a3y2n
There are entire algorithms which can inherently catch *any* attack of *any* type. Example:
https://arxiv.org/abs/2302.10326
Basically, a quick test to try the image out and see if it has a corrupting impact. If so? Tossed.
But this is one of those ethical questions that need to be addressed. I agree that living artists should be able to opt out, like a central database where living authors subscribe to if they want to be left out, perhaps.
I don't care if AI uses my work, though.
It's like trying to copyright open source.
1) The *very first step* in training is cropping and downscaling images, which throws away pixel-to-pixel changes like Glaze and Nightshade.
2) Creating new corrupt images has no impact on *existing datasets*, which again, *already exist*.
1) First actually work:
https://www.reddit.com/r/aiwars/comments/168d010/glaze_is_kinda_bad/
2) Not be readily removed:
https://github.com/lllyasviel/AdverseCleaner
This is beyond wishful thinking.
https://github.com/lllyasviel/AdverseCleaner
The problem would be however, how to recognise an artwork as being from an optout? Using metatags?
Trainers are generally more than happy...
Secondly, demanding that even public-domain data be opt-in is a complete non-starter.
@theglazeproject.bsky.social
The claimed intention of the Sept 2022 update was to prioritise expert content and accredited news, and across the industry is has done the exact opposite
If we have to go back to the old days of the internet without algorithms, so be it
Learn how to create with AI or die off like the dinosaurs.
Just like fucking AI learns to create art.
You're a whiny dumbass.
Learn or die like the dinosaurs.
and you people wonder why your hated
Dumbass.
You have no idea how hard that doesn't and wouldn't matter even if it I was.
https://bsky.app/profile/did:plc:yxovbwf4jzgq2rezjmvtcsoq/feed/aaaodfhid7sle
& of possible interest to SciArt feed 🐡
"adapt or die", this IS adaptation, that they don't like it is not our problem