- cross-posted to:
- hackernews@lemmy.smeargle.fans
- cross-posted to:
- hackernews@lemmy.smeargle.fans
Remember how we were told that genAI learns “just like humans”, and how the law can’t say about fair use, and I guess now all art is owned by big tech companies?
Well, of course it’s not true. Exploiting a few of the ways in which genAI --is not-- like human learners, artists can filter their digital art in such a way that if a genAI tool consumes it, it actively reduces the quality of the model, undoing generalization and bleading into neighboring concepts.
Can an AI tool be used to undo this obfuscation? Yes. At scale, however, doing so requires increasing compute costs more and more. This also looks like an improvable method, not a dead end – adversarial input design is a growing field of machine learning with more and more techniques becoming highly available. Imagine this as sort of “cryptography for semantics” in the sense that it presents asymetrical work on AI consumers (while leaving the human eye much less effected).
Now we just need labor laws to catch up.
Wouldn’t it be funny if not only does generative AI not lead to a boring dystopia, but the proliferation and expansion of this and similar techniques to protect human meaning eventually put a lot of grifters out of business?
We must have faith in the dark times. Share this with your artist friends far and wide!
I can’t endorse Glaze or Nightshade, sorry. If literally nothing else, it’s not Free Software and it’s offered with a nasty license:
So I’m not allowed to have the discussion I’m currently having, nor to include it in any Linux distro. To me, that’s useless at best and malicious at worst. Ironic, considering that their work directly builds upon Stable Diffusion.
Also, Nightshade will be ineffective as an offensive tool. Quoting from their paper:
This is not only an admission of failure but a roadmap for anybody who wants to work around Nightshade. Identify poisoned images by using an “alignment model,” which correlates images with sets of labels, to test whether an image is poorly labeled; if the image appears well-labeled to a human but not to an alignment model, then it may be poisoned and will need repair/corroboration from alternate sources.
I also ranted about this on Mastodon.
I wasn’t aware of the licensing terms - that immediately drops my opinion of it too. I was thinking about looking into it (soon.gif, due spoons) to see if I could make a run-your-own-nightshade deployable for people, but I guess that’s off the table now
as to effectiveness: as I’ve said elsewhere, the genie is out of the bottle. unless this shit gets (toothfully) regulated out of existence (and that might be impossible for a variety of reasons), I fear that it’s going to become a similar arms race as spam, and there will continue to be a tug-of-war for a while.
gutfeel, it strikes me that the (current) biggest hope is that the models themselves not only fail at providing but don’t really have a path to achieving that either, so it’s mostly a case of how long VCs can fund the hype. from previous cycles it looks like that spans 2~5y windows. at the point the hype funding runs out, this stuff will significantly lose traction even though it won’t disappear entirely just yet.
damn large amount of damage that’ll happen in
Ha! Nope, not buying it.
Funny you mention licenses, since stable diffusion and leading AI models were built on labor exploitation. When this issue is finally settled by law, history will not look back well on you.
Doesn’t seem to prevent you from doing it anyways. Does any license slow you down? Nope.
Not sure that’s true, but also unnecessary. Artists don’t care about this or need it to be. I think it’s a disengenous argument, made in the astronaut suit you wear on the high horse drawn from work you stole from other people.
Sounds like an admission of success given that you have to step out of the shadows to tell artists on mastodon not to use it because, ahem, license issues???
No. Listen. The point is to alter the economics, to make training on image from the internet actively dangerous. It doesn’t even take much. A small amount of internet data actively poisoned requires future models to use alignment to bypass it, increasing the marginal (thin) costs of training and cheating people out of their work.
Shame on you dude.
Good luck on competing in the arms race to use other people’s stuff.
@self@awful.systems can we ban the grifter?
I am buying it. I don’t think @corbin@awful.systems is pro AI art, just that countermeasures like Glaze and Nightshade are not great either, and I agree.
I care about it. Some artists use Debian. Please don’t shit on people who care about software freedom, even if you don’t.
Making artwork unusable by exploitative machine learning models is cool and based, but using a proprietary tool that’s itself made by from the same pool of exploited artists’ work is less so.
yeah, seconding. nothing I’ve seen of @corbin’s posting, here or otherwise, leads me to think that they’re in favour of exploitation or the numerous other issues involved in this shit
corbin’s track record both on and off awful.systems indicates they aren’t any kind of grifter
in fact, myself and @dgerard@awful.systems have previously had a conversation about this type of technology (Nightshade and Glaze) where I was initially quite excited about it, but David and others brought up a lot of the same points corbin did here. there were some very solid social points made around the tech too, beyond the licensing and technical points we’ve seen here — should we really be establishing the expectation that artists need to defend their work using this specific proprietary technology? that feels way too close to the bullshit the NFT grifters pulled, where artists could opt out of their work being stolen and sold as an NFT only by following a specific set of steps for each and every NFT market, which doesn’t work. this kind of tech also opens the door for rent-seeking; techniques like Glaze and Nightshade can be broken by changes to generative models, which would keep artists on a treadmill continually paying for the latest versions of these proprietary tools, or else. it feels rather like a protection racket run by whoever has the most access to the models — and that’ll always be the same assholes who run the generative AI.
so I ended up with the strong impression that this technology won’t make things better for artists and other folks who are being exploited by the AI industrial complex; that this might not be something with a purely technical solution. and I think I understand your strong reaction to some of the posts here, because that fucking sucks. there isn’t a clean engineering solution to this problem that my increasingly technofascist industry created, and when you grow up being told (by some of the same techfash fucks who’re now behind some of the worst use of technology I can think of) that all you do is engineering, it’s easy to feel helpless.
but we aren’t helpless. technofascism is structured to produce and exploit that feeling when it’s engaged with on a purely technical level, but the systems established by technofascism (LLMs, generative AI, cryptocurrencies, and others) are plainly ridiculous when viewed through any other lens. the technofascist goal isn’t to win on technical merit (there isn’t any), but to normalize ridiculousness. the only way I know to push back against that is social. on a small scale, that’s part of what sneering is — any asshole pushing this ridiculous shit should feel ridiculous doing it, as a lot of the crypto grifters felt when the public at large started sneering at crypto (thanks to the efforts of David, Amy, Molly, and many others). on a larger scale, we desperately need systemic change. as engineers, we’re constantly told we don’t need unions or solidarity, particularly with folks like artists who we’re told are unimportant. it is very intentional that attitudes like that enable technofascism.
if and when we have those social factors established, a version of these tools with less potential for exploitation might be worth considering. but I see it kind of like the relationship between fediverse software and its community — federation is generally a good thing, but it’s absolutely nothing (and would probably be a net negative) without posters who generally want the fediverse to be a cozy place to make good posts; the polar opposite of the utterly hostile commercialized thing the internet at large has become.
I don’t think they were defending ai necessarily, just saying they had objections to the specific technique used by these tools. I do think that not open-sourcing the thing is probably defensible given that it exists in an adversarial context, but the technical concerns are worth being aware of
I certainly comply with software licenses so yes, that does slow me down. As they pointed out this precludes it from appearing in Linux distros and such.
Incidentally, I’ve gotten into very long and stupid arguments with people about Stable Diffusion’s Definitely Not Open Fucking Source license.
I’ll bite…
Why do argue is it okay to use this “poison pill” which was created using the same “stolen data” you are railing against and not other, more open, tools?
You underestime how much labor already goes into cleaning/filtering datasets, this does not change the economics nearly as much as you want it to. Do you not think there are any filters in place?
If you want to get into it everything in the modern world is built on labor exploitation. When this issue is finally settled by the law™ it will be in the favor of increased capital consolidation built upon the arguments and misinformation you are repeating. If the law™ expands copyright protections who do you think that benefits the most?
Your tone, lack of research into the topic, calling others grifters, and attempts to silence any contradictory viewpoints isn’t great… Why are you having such a a visceral response to this topic and where do your talking points come from?
I don’t care whether you agree with me. I do think it’s quite interesting that my critique of the topic of your post has you replying with personal insults; I think that you’ve incorrectly assumed that I am one of the capitalists who build these models, rather than an anti-capitalist who encourages destroying corporations.
Who are you, Lars Ulrich? Copyright is not compatible with information theory, and so far, information theory has won every contest in the court of public opinion.
Help defray or establish artists’ funds. Help uncover and prosecute wage theft. Advocate for basic income. If your goal is to remunerate artists, then focus on the efforts which actually help them; don’t support copyright, as it is neither designed nor implemented to help individual artists.
Which one of us is defending a paper which explicitly offers itself as a way for Disney to protect art which it appropriated from artists?