Artist Jingna Zhang built Cara, a platform explicitly intended to protect artists from having their art stolen by AI…and yet these people just won’t leave them well enough alone.
Artist Jingna Zhang built Cara, a platform explicitly intended to protect artists from having their art stolen by AI…and yet these people just won’t leave them well enough alone.
Yeah good luck asshole.
PSA to anyone who uploads art online, anywhere. Poison it first. Use the latest version. I’m sorry but you are in an arms race now. Dickhead over here is full of shit, but even if they’re not, Nightshade is updated fast.
Dang, so… Is there a before and after comparison of an image being poisoned? Or would that defeat the purpose because an AI could then learn to recognize poison?
All the images on that page have been Nightshaded. Apparently it’s noticeable on images with large flat sections of a single color, but only barely. You probably won’t find many before and after examples fir obvious reasons, but it’s free though, so nothing to lose in trying it out yourself
Nightshade and Glaze only run on Windows. Cara has a partnership with them so that your work automatically gets Glazed before being published, hence the anti-AI fame.
Oh cool! Didn’t know that!
Poisoning it is sadly an Arms Race, one that will inevitably be lost
And then won, and then lost, and then won. That’s how an arms race works. And we have the advantage in numbers. There are far more people who don’t want their shit scalped than people working for the morons at OpenAI, or Google or Anthropic. And yes, they ARE morons. The battle is only lost when one side gives up. You only lose when you accept defeat.
Unfortunately the anti-AI side is guaranteed to lose eventually, because there’s a definitive lower bound: humans must be able to view and understand the image, ideally with no perceptible loss in quality compared to the original.
This severely limits how far poisoning can go, whereas the theoretical limit for those stealing the art is at least the same level of understanding that humans have (in the sense that our universe physically allows such a mechanism to be implemented, as demonstrated by our bodies).
At the very least it makes the case that the artist believed strongly enough to prevent people from using their art without permission that they use increasingly radical methods.
It makes it so crystal clear that these shitheads are acting in bad faith when they’re circumventing custom made solutions to keep them from using the art.
As if a sign on your website saying “not for use in AI training” shouldn’t be enough. Like imagine I posted art with the message “not for use in movies” and then it got used in a movie. Open and shut.
This. I don’t believe it’s a losing battle like some in this thread, but even if it is, it’s still worth doing.
Yeah, I’m with you there. IMO things like
robots.txtshould have legal weight.The problem there is that every cycle the old ‘protected’ images become useable in training.
The issue is that they can just reuse the existing art once they’ve “removed” the poison that was used on it. The artworks with poison in them aren’t forever inaccessible to them, they just get them slightly later.
That doesn’t mean we should stop, but it still is a war that we ultimately can’t win. Even if the poison technology advances faster than the AI companies are, they can simply use slightly older art with a less sophisticated version of poison.
I mean, is the goal to prevent training or poison training. Yes nightshade will never prevent your work from being trained on, but hopefully models don’t know it’s poisoned and it sows model dysfunction.
There is no numbers advantage when less than .001% of images are poisoned and of those 99.99% are using outdated tools.
Besides, what’s the goal? Slow its progress down by a few weeks? The inevitable will happen regardless. Do you think people who edited their reddit comments before coming to lemmy made any difference? It’s laughably easy for an AI to tell is garbage data, even easier when you’re using a tool whose source the AI can view.
Yeah fuck this defeatism. Anything you can do you do it. Same thing with anything you strongly disagree with.
Even if you’ve failed don’t give the impression that this was something you went along with. Don’t make it seem like this was just the natural zeitgeist and no one fought it.
Hopefully there will be history in the future, and if there is it should record the fact that there were once humans who tried to be a nuisance over what they believed in even when all the money in the world was against them.
An Arms Race always reaches an end, and sadly there’s more money and resources on the side of AI (but less people) than the side of anti (which has lots of people, but barely any money, and barely any resources)
It will end in defeat, but a slow gradual one at least
2 things.
Master locks are the easiest to pick and still the most popular. I don’t see you advocating for leaving your bike unlocked because it’s inevitable that the bike thieves win.
That money is fake and recirculated like the air in an airplane. That’s literally what people mean when they talk about the AI bubble. All these companies are incestuous as fuck. Google invests in Nvidia who invests in Anthropic who invests in Microsoft who invest in OpenAI who invest in Nvidia who invest in Google. There’s no NEW money coming in. That has ALREADY dried up. They’re literally passing money in a circle. That’s WHY it’s a bubble, and that’s how it’s already popping.
Personally I think we should watch Oracle. They’re by far the most over leveraged. At least publicly. When they fall over, everyone else will follow. The big ones like Microsoft and Google are at least diversified and might survive, but the bubble pop will eventually catch up. OpenAI already spend $1.22 for ever $1 they make, and their expenses grow in lock step with their demand. Sam Altman’s plan to make it all profitable is to build AGI and ask it to figure it out for him because, and I cannot stress this enough, he. is. a. moron.
Your defeatist attitude is what they want because the more people that believe that, the longer the bubble is sustained.
Take heart. They are weaker than you realise. They are not infallible, their armor is paper.
I’m drunk and apologize for my flowery language.
Happy Friday, the fight is winnable.
I love you all.
I’m with you, Kraiden.
I admire your optimism, and I really do hope you’re right!
Also I’m a fan of flowery language, so keep it up
I agree we are in a bubble, but I think we should look at what is allowing us to be in a bubble.
My best guess is that, as a society, we are completely structured around greed. If you have a way to make money, nobody is going to tell you to stop it. Obvious things that previous generations would have put a kabosh on that we are unable to, things that make some individuals money, put are socially parasitic: cryptocurrency and online gambling (that is GAMBLING spelled G-A-M-B-L-I-N-G: GAMBLING. In a sentence: Audrey spent her entire retirement GAMBLING at the casino).
In the run up to the subprime mortgate crisis, Senator Diane Warren tried to raise the alarm that these loans were structurally unsound. She got calls back from colleagues all basically saying the same thing:
They’re all making money! Therefore, the system is working as intended.
If you want to look at it more structurally, we are also ruled an elite group of mega-rich. They have more to gain and less to lose by keeping the AI train going, but they are going to crash the world economy in doing so: a great opportunity to snap up real estate for cheap for them. The rich never lose.
I could not agree more. I’ve been having conversations with people about this since late last year and predicted that the bubble would burst some time in 2026 but I’m thinking I may be off and it could be early 2027 now. This accounting nonsense makes up a very significant portion of the US stock market, so the bubble bursting is gonna have some pretty big flow-on effects as well. I’d recommend anyone with any form of investment look into what they have and plan for a significant market event.
You are delusional to think next year is gonna be curtains for a trillion dollar industry backed by billionaire CEOs, military and government influence. The AI bubble won’t pop, if it ever does, until well into beyond the next decade.
For the AI bubble to not pop, an AI company needs to make money off of AI first. Not raising investments. They need to actually make at least $1 off of using/selling AI services.
This hasn’t happened yet, and I don’t think it ever will. The AI machines simply cost more to run than they produce, so when the investment money runs out, they’ll be shut down.
In 2007: “How could the housing bubble pop? After all, the real estate industry is backed by billionaire CEOs, military, and government influence.”
In 1999: “How could the dot com bubble pop? After all, the internet is backed by billionaire CEOs, military, and government influence.”
In 1929: “How could the Stock Market crash? After all, international capitalism is backed by millionaire CEOs, military, and government influence.”
I really hope you’re right and not me. I’ll check back in with a comment in a year from now so you and I can talk about what happened and what to do at an individual level next.
This rocks and you’re 100% correct. Bless you.
Paper armor and glass cannons.
Which is why I take a completely different direction. I poison my artwork I post online through the one technique they can’t account for.
See, I’m not super great at drawing, and in order to protect their precious models, they’d literally have to go take a look and say “yeah, that drawing wasn’t very good”. They can’t detect that automatically yet!
/sorry for a very predictable joke btw