|
Tei posted:If your poo poo uses my work and make derivative work from it, and I can prove it, the problem is not that your servers hold a copy of my work, but that your work is derivative work from mine. You need me to license you to make derivative stuff from my work, or is illegal* Okay, and what if my poo poo doesn't use your work but does make stuff derivative from it? This is the question y'all seem to be ignoring and is, largely, the current state of play already. Reveilled posted:If that’s so it seems even more important to focus on given that it the question that’s most likely to be legislated and litigated on on the near future! Yeah but the discussion of it is "artists" in this thread who are angry, ignorant, shortsighted and unwilling to listen to or engage in any kind of meaning discussion, using a laser focus on it to beat down anything more interesting. They haven't done an iota of work answering questions as to how what they are saying makes any actual sense, how it solves any of their actual problems, or why it would be worth the likely consequences. Focusing on it doesn't seem to be getting us anywhere because the only people pushing for it refuse to think about it. It's the question most likely to be litigated soon, sure, but in terms of the actual AI field it's nearly irrelevant, at least on the art front, and no one seems to be pushing similar litigation for something like chatGPT where it would matter a lot more. Also, I know it's not something the "artists" in this thread like to hear, but the artists here on all sides of this issue and others and calling fellow artists "techbro evangelists" and denying they are real artists if they disagree about something really isn't helping the discussion. As an artist who has zero real skin in the game (AI isn't coming to replace my medium any time soon, and I'm not sure it will ever be able to - most I can see is being able to use it to push the envelope in some ways) I recognize this is less... Personal... For me. But considering "maybe we should just ban it and it's products" is one of my main arguments, being dismissed as a "techbro evangelist" for calling out the flaws in someone's argument doesn't incline me to want to focus on the topics where that's the only likely outcome.
|
#
?
May 27, 2023 13:01
|
|
|
|
| # ? May 26, 2024 16:14 |
|
|
KillHour posted:Replace "artist" with "coal miner" and you will have the answer to both what will happen and why trying to blindly hold on to the way things are now is doomed to fail and will ultimately be destructive in its own ways. And this is why we've realized that copyright is a lesser evil then you. Gentleman Baller posted:Adobe paid for and owned those images well before the new AI stuff came out as part of their stock images collection, and is a company with a market cap of 190 billion dollars. I have no idea if openAI could pay or not, but if they had to pay for it I'm sure the model wouldn't be available to people like you and me. They were not necessarily licensed for this use. Given how this crapola skyhooks off lots of money and rampantly plays games with IP law, I think you plagiarists are massively overestimating yourselves.
|
|
#
?
May 27, 2023 13:06
|
|
|
Reveilled posted:And if the only option AI advocates are willing to put forward is the last one, is there any reason for artists not to line up behind larger copyright holders and do everything in their power to spite you and bring you down with them? I think people are cautious to start talking about their idea of a solution, from the pro-ai side, because they don't want to look radical when they feel like they're walking a fine line with a currently controversial thing. Not on SA, probably, but it's other places I noticed it because I talk to people in public spaces and then in private spaces and see the difference. You can stereotype them as "haha techbros everyone's Elon Musk" if it makes you feel better but that's not really what most people actually doing stuff with AI are like other than the literal people running some of the companies like OpenAI because of course OpenAI is like that. There are many, many leftists in that space (and other things.) I have not seen a lot of crossover with crypto guys, other than them just talking about it but they're not, from what I've seen, actually there and active and contributing in AI spaces. So you get the lukewarm "well, UBI!" because that's palatable to the mainstream. But the real thing is, the AI generated art thing is an incredibly small part of something much bigger that only appears large because artists, generally, have a larger audience than all the other people being automated out of a job in a system where they aren't the ones who get any of the benefit of that. Automation can be a very good thing. The machines just do most things and I spend a few hours a week checking over what they did in my industry, and everythings still productive etc. And a lot of people say basically that, but then they don't go further because further is some form of communism/anarchism. If the advancement of new technology, the thing humans have done since the very beginning, causes your system you got going on to collapse, the problem probably isn't the technology, it's your system. And that's what actually happening, in my opinion. AI isn't causing problems, capitalism is failing because of its problems.
|
|
#
?
May 27, 2023 13:06
|
|
|
With all this copyright talk I am curious how it would work with AI programs if stronger copyright law was implemented. How are artists normally paid for things like this, would they get a flat fee offer from the company or set their own rates? I have to imagine that a company is developing an AI art program they aren't going to offer much to artists because the program is trained on hundreds of millions of images. Basically, is it going to be a race to the bottom of offering the cheapest bulk images available. I don't think the quality of one individual piece of art influences the system enough to be worth paying more than like a dollar to use it and a trained artist can imitate another artist's style. I guess a small amount is better than nothing, but it's effectively nothing to the individual either way. The only way I could think of an individual making art to be licensed by AI companies would make money would be if they were paid each time the program was used, like a Spotify stream or something.
|
|
#
?
May 27, 2023 13:14
|
|
|
StratGoatCom posted:And this is why we've realized that copyright is a lesser evil then you. "We", here, does not mean artists, but rather means you and people who think like you. But the fact that you likely find your fellow artists evil and hold them in contempt is... Not a surprise. Now - why aren't you answering any of my questions or engaging with me at all?
|
|
#
?
May 27, 2023 13:15
|
|
|
StratGoatCom posted:They were not necessarily licensed for this use. Given how this crapola skyhooks off lots of money and rampantly plays games with IP law, I think you plagiarists are massively overestimating yourselves. If you have any evidence or legal analysis that points to this I would love to read it.
|
|
#
?
May 27, 2023 13:40
|
|
|
Gentleman Baller posted:If you have any evidence or legal analysis that points to this I would love to read it.
|
|
#
?
May 27, 2023 15:39
|
|
|
StratGoatCom posted:Oh please, you always have to be clear in chain of title you dork. The AI wierdos made this problem, not us. And yes, we will watermark, yes we will do things to choke your models because you did not ask permission for usage and theft invites retaliation. Speaking of this, a lawyer went and did just that. https://twitter.com/questauthority/status/1662273759259295746 Peter LoDuca, arguing a case before the Southern District of New York entered an affidavit generated by ChatGPT that outright made up case citations to cases that never existed. When pressed by the Court, he then generated those decisions with ChatGPT and submitted them to the court, complete with a forged notary stamp. To say the judge is a bit upset would be the understatement of the year.
|
|
#
?
May 27, 2023 17:48
|
|
|
GlyphGryph posted:Okay, and what if my poo poo doesn't use your work but does make stuff derivative from it? That don't exists. If my style is painting red big noses, you get that information from somewhere. Sure, you don't have to store a copy of the original artwork, you learned from it, but is still a machine that need my work to create yours, using my style.
|
|
#
?
May 27, 2023 18:08
|
|
|
Tei posted:That don't exists.
|
|
#
?
May 27, 2023 18:10
|
|
|
StratGoatCom posted:And this is why we've realized that copyright is a lesser evil then you. The fact that you could look at that, agree with my take, and go "The coal mining companies are the lesser evil" is astounding.
|
|
#
?
May 27, 2023 18:12
|
|
|
Tei posted:That don't exists. I think it would also be hard to prove that its your style. How many pictures of clowns with big red noses exists, Rudolf the reindeer, people with rashes on their nose, or people with red paint on their face? It could easily make something with red noses without looking at a particular artists red nose style.
|
|
#
?
May 27, 2023 18:15
|
|
|
gurragadon posted:I think it would also be hard to prove that its your style. How many pictures of clowns with big red noses exists, Rudolf the reindeer, people with rashes on their nose, or people with red paint on their face? It could easily make something with red noses without looking at a particular artists red nose style. I think the prompt of the image being "Person draw in the style of Tei" pretty much clears out whos style is being copies. Is not it? It also shows the author of the AI BOT had images with metadata "Created by Tei" in their learning corpus.
|
|
#
?
May 27, 2023 18:21
|
|
|
Tei posted:I think the prompt of the image being "Person draw in the style of Tei" pretty much clears out whos style is being copies. Is not it? That does say that you want it in the style of Tei, but it doesn't tell you how the AI bot "learned" what Tei's style is. I guess that would put the blame on the person who is prompting the AI though I guess you can't copyright style anyway apparently. That leads to another question I have. How many artists have a distinct enough style that they can be easily known and imitated by AI? Wouldn't most artists styles already be classified under impressionism, art deco, surrealism or whatever else? And how many of those artists have works that are new enough to not be public.
|
|
#
?
May 27, 2023 18:34
|
|
|
Tei posted:That don't exists. Of course it exists. Your style is not built out of things unique to you. It can get that information in a ton of ways without having any access to your work. It can synthesize the styles you were inspired by, for example, under my guidance, to add and remove and tweak elements until it matches what you do. Or maybe I hire an artist to do ten original pieces in your style and feed that to the machine, tell it it's in your style, and there we go - I'm genning works without the machine ever seeing anything of yours. Maybe my machine just has a bunch of art that is labeled as *inspired* by Tei, and it works backwards to figure out your style from those pieces. Would that all be fine, even if it produces images eerily similar to ones you would make an extent that a lay person would think you had done it because it's so distinctively you?
|
|
#
?
May 27, 2023 18:43
|
|
|
Let's see if I can condense all this.Reveilled posted:Is it the case then that these models trained on non-copyrighted content are uniformly worse than the ones trained on the copyrighted content? If that’s so it does seem to imply that the copyrighted content does provide direct commercial benefit to the models which use them, in which case it seems very reasonable to at least discuss whether these models should pay to license them. Model training is currently a black magic, it's all entirely way too new and there are no 'best practices' yet, just spaghetti stuck to a wall and some of it doesn't fall off. Overall things are getting better as people share results and learn from each other, for instance it's looking more like largest model isn't necessarily best, there might not be a reason to hoover up the entire internet to create a well working model. Working with a 'limited' dataset of 'just' public domain images may end up being worse at the start but with multiple passes over months end up just as good as todays models, currently the trade-off is time. I know that doesn't sit well with someone worried about today, but it's true. Reveilled posted:Fair enough. At the start, like back in September, I did actually worry their might be a push to outright ban AI image gen, but that doesn't look like it's going to happen. This is important to say so I'll do it again. It does not look like anyone is trying to actually outright ban AI image gen. I don't know the shape of things to actually come, what a potential legislation push will actually look like, this means nothing is guaranteed to happen. Their could end up being no laws, I think this is doubtful but it very much means the situation is fluid. This makes it all the more important for people to say what they actually want out of things, not the half compromises they've already talked themselves into. If you want UBI, you need to be out there talking about UBI. If you want UBI but think it's not very likely that UBI will every actually come to pass... YOU NEED TO BE OUT THERE TALKING ABOUT UBI ANYWAYS, this is the only way it can EVER come to pass. I'm in huge favor of taxing the poo poo out of automation to fund it, this is the way. I've only taken a cursory look at the EU legislation, specifically the open-source parts, those are left open enough that I don't have to raise any alarms. I don't know about commercial restrictions because I'm non-commercial. I'm personally less inclined to care about those. They might be a good thing in the end, I don't know yet, I'd have to carefully read the proposals and that takes time. gurragadon posted:With all this copyright talk I am curious how it would work with AI programs if stronger copyright law was implemented. How are artists normally paid for things like this, would they get a flat fee offer from the company or set their own rates? I have to imagine that a company is developing an AI art program they aren't going to offer much to artists because the program is trained on hundreds of millions of images. Basically, is it going to be a race to the bottom of offering the cheapest bulk images available. There is no current framework for payments, just a lot of wishful thinking over how things could be. I don't know if some of the vitriol in these arguments is because people think they'll be paid fat residuals off this but they're in for a very rude awakening over how things are currently more likely to be written. The Spotify model IS the nightmare scenario. Currently artists get paid somewhere between gently caress and All. This is literally the way to screw artists the absolute most possible, take the number of pixels in a generated image and straight divide them by the total number of pixels trained IN THE ENTIRE MODEL. Hello getting a full hours worth of animation out of the parts in the model trained on your stuff and millions of other artists and getting paid $0.0000000000042 while the movie goes on to make billions. If you think hollywood wouldn't drool over getting that written into law then I don't know what to say. Yes I do, go read a history book. I have a much longer piece on copyright in general that I thought might be outside the scope of this thread but with the conversations going on I've made up my mind to write it out. I need some time but I should be able to have it done by later tonight. It will be on an actual recent change in copyright law so people can get a better idea of why I'm so vehemently against further changes. Tei posted:That don't exists. This explains a lot, you don't understand that I don't need to train on anything you've done to be able to pixel-by-pixel output something you have made. Stable Diffusion is an Infinite Image Generator. Every possible image can be made with it. All of them. Please don't confuse that with the amount of work that would be required to go from something that just looks 'like' something you've done with going the extra extra mile and making it pixel-by-pixel, but from a pure math standpoint it's entirely doable. This also has no bearing on the already exists laws for if something used in a commercial setting looks similar to your work and not outright pixel-by-pixel. There already exists legal framework for dealing with that.
|
|
#
?
May 27, 2023 19:26
|
|
|
KillHour posted:The fact that you could look at that, agree with my take, and go "The coal mining companies are the lesser evil" is astounding. This is why Trump won, and will win again...
|
|
#
?
May 27, 2023 23:12
|
|
|
Lol at the AI dorks turning concept artists into the next displaced cohort with nowhere to shelter but right wing corporatism.
|
|
#
?
May 27, 2023 23:14
|
|
|
porfiria posted:Lol at the AI dorks turning concept artists into the next displaced cohort with nowhere to shelter but right wing corporatism. I guess I’m an AI artist dork and I’m saying you all should stop attacking artists for using AI and instead be pushing for UBI
|
|
#
?
May 27, 2023 23:31
|
|
|
Liquid Communism posted:Speaking of this, a lawyer went and did just that. This story is really funny, the guy kept asking ChatGPT if it was lying to him. I mean you would figure he would at least take the time to make sure citations are correct or real but apparently not. https://www.nytimes.com/2023/05/27/nyregion/avianca-airline-lawsuit-chatgpt.html NYT posted:Mr. Schwartz said that he had consulted ChatGPT “to supplement” his own work and that, “in consultation” with it, found and cited the half-dozen nonexistent cases. He said ChatGPT had provided reassurances.
|
|
#
?
May 27, 2023 23:42
|
|
|
porfiria posted:Lol at the AI dorks turning concept artists into the next displaced cohort with nowhere to shelter but right wing corporatism. This is bullshit, there is absolutely no reason to go right wing anything, ever.
|
|
#
?
May 27, 2023 23:53
|
|
|
KwegiboHB posted:This is bullshit, there is absolutely no reason to go right wing anything, ever. EXT. RUINED RED HOOK - DAY Barron Trump And when I am elected president, I vow to ban AI and bring HR and design jobs back to the people! The crowd, made up of middle aged zoomers, their bodies ravaged by poor nutrition, drugs, and vaping, cheers wildly. INT. MSNBCNN STUDIO - DAY "Expert" (Dorkishly) Um, I think you'll find that the automation of white collar jobs actually increased utility in aggregate and empowered consumers. (USER WAS PUT ON PROBATION FOR THIS POST)
|
|
#
?
May 28, 2023 00:06
|
|
|
It must be wild in your head.
|
|
#
?
May 28, 2023 00:10
|
|
|
gurragadon posted:That leads to another question I have. How many artists have a distinct enough style that they can be easily known and imitated by AI? Wouldn't most artists styles already be classified under impressionism, art deco, surrealism or whatever else? And how many of those artists have works that are new enough to not be public. I think an overlooked area where image generators can improve is in the prompting itself. We're dealing with very simple text descriptions right now, relatively, that lack most/anything but the simplest grammar just based around what the pictures were tagged as. Like, SD barely understands the concept of something being "on" something. Or if you have "x fighting y" it often doesn't get that x is one thing and y is another and that they are fighting each other. This makes it hard (but not impossible, especially with other tools like inpainting and control net) to really build explicitly what's in your head for a lot of things. People are aware of this and have been working to build systems that better understand more complex language. And I think, with multimodal AI it's not going to be impossible to tag each training image with a really detailed description so you can get more direct control over the composition in natural language. At that point I think pretty much any distinct style could be pretty easily imitated (or created!) in AI, you just describe it. Even at this point most can be described without specifically calling on that material in the prompt, it's just a little weirder and less "I'm just saying what I want." I actually dont know as much about how foundation image gen models are trained, there are relatively fewer of them than language models. With language models the foundation models are almost always trained for just 1 epoch, and people are starting to wonder "what happens if you do more?" You actually might be able to get more out of less training data. We did pretty recently learn that parameter size really isn't the be all end all, so I'm pretty sure our training methods are still really inefficient compared to what else we'll figure out in the next 5 years. And maybe that's all true for image models too. I have no idea how paying a person for their stuff showing up in an output would work or make sense. It's not pulling from the "number I made to represent this artwork in the model", it's pulling from "averages but not really averages I got from denoising into a whole bunch of images that seemed related in a way to this token" so, what, you pay every artist who ever drew or photographed a cat a billionth of a penny every time someone uses AI to generate anything that may have been influenced by its understanding of a cat, which actually includes an incredibly amount more (which we probably cant tell) than just the prompt "cat?" Commercial AI use is never going to be using specific artist names, I'd imagine, and I doubt that will even ever really be a thing in future models again anyway. BrainDance fucked around with this message at 06:08 on May 28, 2023 |
|
#
?
May 28, 2023 06:03
|
|
|
BrainDance posted:I have no idea how paying a person for their stuff showing up in an output would work or make sense. It's not pulling from the "number I made to represent this artwork in the model", it's pulling from "averages but not really averages I got from denoising into a whole bunch of images that seemed related in a way to this token" so, what, you pay every artist who ever drew or photographed a cat a billionth of a penny every time someone uses AI to generate anything that may have been influenced by its understanding of a cat, which actually includes an incredibly amount more (which we probably cant tell) than just the prompt "cat?" It doesn't make sense, which is why that isn't what anyone is angling for, but rather requiring licensing the works used to train the model in the first place.
|
|
#
?
May 28, 2023 07:14
|
|
|
Liquid Communism posted:It doesn't make sense, which is why that isn't what anyone is angling for, but rather requiring licensing the works used to train the model in the first place. And even then, there's no way it will ever make economic sense for either party. The overall impact of a single piece of art on a trained model is so small that the incremental value of that single piece of art is barely anything. While on the other hand, no artist would ever agree to "here's 15 cents to use your piece of art in our model in perpetuity." Stable diffusion was trained on ~160 million images.
|
|
#
?
May 28, 2023 10:20
|
|
|
Gee, almost like there's a reason ethics of technology has a lot to say about why it's a net negative for society to create systems that rely on mass appropriation of others' work for commercial purposes.
|
|
#
?
May 28, 2023 12:32
|
|
|
It's why I don't think copyright law is a good way to limit the effects of AI generators on artists income. The way the model is trained basically forces each individual piece it is being trained on to be nearly worthless. You either get an insultingly small, but reasonable from the point of view of the AI company, offer to use the image or you get an even more insultingly small cut each time the program is used. Either way, it's not going to compensate people for their work. That also side steps the issue of somebody making an open-source AI image generator that leaves artists with nothing but doesn't run afoul of copyright at all. I think it would be more effective to either try to use AI generated images in artists own work, or just call for the outright ban of this technology. There definitely needs to be more regulation of AI text generators like ChatGPT. The lawyer's story from a few posts up is basically malpractice and that needs to be cracked down on. The entire idea of machine learning seems to be based on mass appropriation of work, do we ban machine learning? Or do we only ban it if the machine is used for commercial purposes?
|
|
#
?
May 28, 2023 13:50
|
|
|
If you let the AI company set your rates, sure. They need your content, you don't necessarily need to sell to them, and should be pricing in the damage their business will do to your future income if you choose to do so.
|
|
#
?
May 28, 2023 14:56
|
|
|
That sidesteps open-source AI image generation. Artists could try to prevent companies from using their work by not selling at a low price, but art is a commodity and no one artist is necessary for an AI image generator. I just see it as a race to the bottom of someone offering the cheapest images. Collective action from artist's could work, but I don't think artists are a group are unified enough to keep the price of their work high. If were worried about machine learning taking human jobs, then we need to ban machine learning. It would be easier to do it now than waiting to do a Butlerian Jihad in a 100 years or whatever.
|
|
#
?
May 28, 2023 15:17
|
|
|
I'm not against banning its application as a content generation machine, but given that we haven't managed to stop running illegal hotels in residential areas (AirB&B et al) or scab taxi services in places where taxis are regulated despite there being existing laws against them that could be enforced, I'm not sure if regulation can get there fast enough.
|
|
#
?
May 28, 2023 16:27
|
|
|
gurragadon posted:or just call for the outright ban of this technology. Liquid Communism posted:I'm not against banning its application There we go, I knew you were in there somewhere. How are you going to ban || ε - εθ (xt, t) ||2 ? The mean squared error between the actual noise at time t and the predicted noise at time t given some image. This funny equation is the beating heart of what's now called AI Image Gen. If you don't understand it, grab two mirrors and play with them until you do. (Look at this bougie gently caress who can afford TWO WHOLE MIRRORS in this economy)
|
|
#
?
May 28, 2023 16:59
|
|
|
KwegiboHB posted:There we go, I knew you were in there somewhere. I'm not. One doesn't need to ban the underlying technology to make a given use of it verboten.
|
|
#
?
May 28, 2023 17:19
|
|
|
Liquid Communism posted:I'm not. You don't have to ban it to ban it. How very Zen, I will have to meditate on this some.
|
|
#
?
May 28, 2023 17:28
|
|
|
KwegiboHB posted:There we go, I knew you were in there somewhere. Well, I don't really care whether it is banned or not, I just find it interesting. I just don't think this whole copyright conversation is really serving our purpose and is just sidestepping the major problems from AI that people are really talking about, which is displacing human workers. But you don't and can't ban the math, you fine anyone who uses it. Obviously, the regulation regime in the United States is garbage so it wouldn't really be effective IMO. We don't ban the knowledge of how nuclear reactors work, but you can't be holding on to a bunch of radioactive materials. Earlier in the thread I mentioned that my organic chemistry book has a synthesis for methamphetamine, but I can't just get methylamine to make it that way.
|
|
#
?
May 28, 2023 17:35
|
|
|
gurragadon posted:I just don't think this whole copyright conversation is really serving our purpose and is just sidestepping the major problems from AI that people are really talking about, which is displacing human workers. Well you just think that way because this whole copyright conversation really isn't serving your purpose and is just sidestepping the major problems from AI that people are really talking about, displacing human workers.
|
|
#
?
May 28, 2023 17:43
|
|
|
Thats why I'm trying to move the conversation off copyright and onto talking about worker displacement and banning AI. When I said "our purpose" I mean our purpose as a thread having a conversation on AI. Not that I have a purpose of banning AI.
|
|
#
?
May 28, 2023 17:47
|
|
|
You could absolutely ban certain types of AI, just like we ban lots of the other applications of math.
|
|
#
?
May 28, 2023 19:10
|
|
|
Here's an interesting use: https://twitter.com/bengaliyaoi/status/1662546889915486210 Guy generates an image, and traces or uses it as a reference for their own work, and sells the resulting work. There are obvious ethical issues with selling traced work, but the comments claiming plagiarism confuse me. If generated images are effectively public domain, then I don't see how referencing or even tracing runs afoul of copyright infringement? The only violation that might have occurred is of course the training/generation of the image in the first place. Any use of that image after that seems like fair game, legally speaking.
|
|
#
?
May 28, 2023 19:13
|
|
|
|
| # ? May 26, 2024 16:14 |
|
|
Plagiarism is legal and normal in US creative industries so talking about it doesn't mean much outside of situations like academics, where it is against organization rules and does still matter in regards to AI. It is arguably plagiarism, sure, but it doesn't matter legally because thats not something the law cares about.
|
|
#
?
May 28, 2023 19:19
|
|