And? What's the model supposed to do? It's just doing what many human artists would do, if they're not explicitly being paid to create new IP.
If infringement is happening, it arguably doesn't happen when an infringing work product is generated (or regurgitated, or whatever you want to call it.) Much less when the model is trained. It's when the output is used commercially -- by a human -- that the liability should rightfully attach.
Assuming you can identify it's someone else's IP. Clearly these are hugely contrived examples, but what about text or code that you might not be as familiar with?
Given enough time (... a surprisingly short amount) and enough people creating art (say, about as many as we have had for the last couple hundred years) and indefinitely-long-lived recording, plus very-long copyright terms, the inevitable result is that it's functionally impossible to create anything within the space of "things people like" that's not violating copyright, for any but the strictest definitions of what constitutes copying.
The short story treats of music, but it's easy to see how visual arts and fiction-writing and the rest get at least extremely crowded in short order under those circumstances.
> It's just doing what many human artists would do
I really don't think so. If I paid a human artist to make the prompt in the title, and I didn't explicitly say "Indiana Jones" I would think it should be fairly obvious to the human artist that I do _not_ want Indiana Jones. If they gave me back a picture of, clearly, Indiana Jones, I would ask them why they didn't create something original.
I actually don't think it would be obvious. By not explicitly saying Indiana Jones when so obviously describing Indiana Jones, there is an implication present. But I think many human artists would probably ask you, "Wait, so Indiana Jones, or are you looking for something different," before immediately diving in.
I‘m not so sure, unless you are playing a game of “name the character” you generally don‘t want Indiana Jones unless you explicitly mention Indiana Jones. Indiana Jones is a well known character, if you want a picture of Indiana Jones it is simple enough to just say: “Draw me a picture of Indiana Jones”. The fact that they didn’t say that, most likely means they don‘t want that.
Late reply but I did test this and I think the results show at least some originality when explicitly directed to not copy existing characters such as Indiana Jones[1] and the Predator[2]. Some elements of the original characters creep in, the archaeologist is wearing a fedora and the distinctly more skeletal bounty hunter appears to have a few dreads despite otherwise being bald, but they are distinct.
Meta-comment: the use of Indiana Jones, a character that was a very intentional throwback to the "Pulp hero explorer" from the childhoods of its creators, in this example to ponder how one would get "Indiana Jones without Indiana Jones" is quite humorous in its own right.
Indiana Jones is already a successful permutation of that approach. He's Zorro, Rick Blaine, and Christopher Leiningen mashed together with their serial numbers filed off.
I agree. But massive changes in scale or leverage can undermine this type of principled stand.
One death is a murder; 100k deaths is a war or a pandemic. One piece of chewing gum on the ground will get you a caning in Singapore; when everyone does it, that's NYC.
Up until now, one had to have some level of graphical or artistic skills to do this, but not anymore. Again, I agree that it attaches to the human...but we now have many more humans to attach it to.
> One piece of chewing gum on the ground will get you a caning in Singapore
This is not true, by the way. You will be fined for littering; or, if you are a repeat offender, be sentenced to cleaning public areas while wearing an offensively bright-coloured uniform (so that everyone can see that you are being punished). Source: https://www.nea.gov.sg/media/news/news/index/nea-increases-v...
But no, you won't be caned for littering. Caning is reserved for more serious offences like vandalism, or much worse crimes like rape and murder.
FWIW I’ve been to Singapore and had a great time, but I was careful to follow the many rules and signs. I especially liked the sign on the bus forbidding the opening of a durian fruit.
> It's just doing what many human artists would do, if they're not explicitly being paid to create new IP.
It isn’t an independent human. It is a service paid for by customers. The moment it provides the image to a paying user, the image has thus been used commercially.
In fact, the user may not even necessarily have to be paying in order to infringe copyright.
And besides, even amateur artists are ashamed to produce copies unless they are demonstrating mastery of technique or expressing adoration. And if it happens spontaneously, they are then frustrated and try to defend themselves by claiming to never have even experienced the original material. (As happens with simplistic, but popular musical riffs.) But AI explicitly is trained on every material it can get its hands on and so cannot make such a defense.
OpenAI is currently valued at $300 billion, and their product is largely based on copying the copyrighted works of others, who weren't paid by OpenAI. It's a bit (exponentially) different from a "me and you" example.
This isn’t generally true. The copyright holder need only claim that the value of their copyrighted work or the profits received by its distribution has been reduced or lost. I’m not sure they even need to make such a claim, as courts have already determined that commerciality isn’t a requirement for infringement. It’s a matter of unauthorized use, distribution or reproduction, not trade.
They of course will likely never know and might not bother to litigate given the private and uncommercial nature, but they still have the right to do so.
They would weigh the financial and reputational cost of litigating against children sharing images and decide not to. Of course if one had 10000 friends and provided this service to them in a visible manner, then they’d probably come knocking.
IANAL, but my understanding is that it's more complicated than this. Generally commercial benefits (i.e. taking money) is one of the aspects being taken into consideration to decide if something is fair use, but it's not the only one and not making a monetary/commercial benefit does not guarantee that work is considered fair use (which is the exception we're talking about here).
The answer is yes? The person doing the drawing is violating copyright. I don't know why that is even a controversial question.
You are asking the equivalent question of, if I put a pirated copy of windows on my PC that I only use privately at home am I violating copyright, or if I sell copies of music for people to only listen to in their own home.
But this is even more damning, this is a commercial service that is reproducing the copyrighted work.
Edit: Just to clarify to people who reflexively downvote. I'm making a statement of what is it not a value judgement. And yes there is fair use, but that's an exemption from the rule that it is a copyright violation.
This would be more like if you reimplemented Windows from scratch if you have violated copyright law.
Let’s put it another way: if you decide you want to recreate Indiana Jones shot for shot, and you hire actors and a director etc. which individuals are actually responsible for the copyright collation? Do caterers count too? Or is it the person who actually is producing the movie?
> It's when the output is used commercially -- by a human -- that the liability should rightfully attach.
I am paying OpenAI. So they are producing these copyrighted works and giving them to me for their own commercial benefit. Normally that's illegal. But somehow not when you're just doing it en masse.
It's not legal or illegal. That hasn't been decided yet. Nothing like this has ever existed before, and it will take some time for the law to deal with it.
> What's the model supposed to do? It's just doing what many human artists would do, if they're not explicitly being paid to create new IP.
Not really? Why would a human artist create a faithful reproduction of Indiana Jones when asked to paint an archeologist? And besides, if they did, it would be considered clear IP infringement if the result were used commercially.
> If infringement is happening, it arguably doesn't happen when an infringing work product is generated (or regurgitated, or whatever you want to call it.) Much less when the model is trained. It's when the output is used commercially -- by a human -- that the liability should rightfully attach.
I agree. Release groups, torrent sites and seedbox operators should not be wrongly accused of pirating movies. Piracy only occurs in the act of actually watching a movie without paying, and should not be prosecuted without definitive proof of such (¬‿¬)
Torrent sites deliver the movie in its original form. AI models maintain abstract descriptions of the content as individually-unrecognizable high-dimensional representations in latent space.
Over the years we've spent a lot of time on this and similar sites questioning the sanity of a legal system that makes math illegal, and, well, that's all this is. Math.
To the extent the model reproduces images from Indiana Jones and the others, it is because these multibillion-dollar franchises are omnipresent cultural icons. The copyright holder has worked very hard to make that happen, and they have been more than adequately repaid for their contribution to our shared culture. It's insane to go after an AI model for simply being as aware of that imagery and as capable of reproducing it as a human artist would be.
If the model gives you infringing material as a prompt response, it's your responsibility not to use that material commercially, just as if you had tasked a human artist with the same vague requirement and received a plagiarized work product in return.
> Still, the near perfect mimicry is an uncomfortable reminder that AI is getting better at copying and closer to
I completely disagree. It's not getting "better." It always just copied. That's all it /can/ do. How anyone expected novel outputs from this technology is beyond me.
It's highly noticeable if you do a minimal analysis, but all modern "AI" tools are just copyright thiefs. They're just there to whitewash away liability from blatantly stealing someone else's content.
> It's not getting "better." It always just copied. That's all it /can/ do
That's true of all the best artists ever.
> They're just there to whitewash away liability from blatantly stealing someone else's content.
That's because that's not a thing. Ownership of "content" is a legal fiction invented to give states more control over creativity. Nobody who copies bytes which represent my music is a "thief". To be a thief, they'd need to, you know, come to my house and steal something.
When someone copies or remixes my music, I'm often not even aware that it has occurred. It's hard to imagine how that can be cast as genuine theft.
OK, you're plunging deep into the ethos of IP piracy, and staking your claim that "None exists." If that is deemed TRUE, then there's nothing to ever discuss about AI... or animated pornography where a well-known big-eared mouse bangs a Kryptonian orphan wearing a cape.
The reality of our current international laws, going back centuries, disagrees. And most artists disagree.
> Perhaps I need a bit of education here, but have there been _international_ laws regarding intellectual property for centuries?!
AFAICT, the first major international IP treaties were in the 1880s (the Paris Convention on Intellectual Property Rights in 1883 and the Berne Convention covering copyrights in 1886; so only ~140 years.)
> Ownership of "content" is a legal fiction [...] To be a thief, they'd need to, you know, come to my house and steal something.
That's just a legal fiction invented so people can pretend to own physical objects even though we should all know that in this world you can never truly own anything.
Everything we do or protect is made up. You've just drawn the arbitrary line in the sand as to what can be "owned" in a different place than where other people might draw it.
Just to play Devil’s advocate for a moment, why should we require human artists to be held to the same standards as automated software? We can make whatever rules we want to.
A human might implicitly copy, but they are not infinitely scalable. If I draw a picture that in some way resembles Buzz Lightyear I am much less of a threat to Disney than an always-available computer program with a marginal cost of zero.
> why should we require human artists to be held to the same standards as automated software?
Isn't this the same question as, "why should we allow general purpose computing?" If the technology of our age is our birthright, don't we have the right to engage whatever mathematics we find inspiring with its aid?
> If I draw a picture that in some way resembles Buzz Lightyear I am much less of a threat to Disney than an always-available computer program with a marginal cost of zero.
...that sounds like a good argument in favor of the always-available computer program.
> Ownership of "content" is a legal fiction invented to give states more control over creativity.
Ownership is a legal fiction invented because it is perceived to encourage behavior that is seen as desirable; this is no more true of ownership of intellectual property or other intangible personal property than it is of tangible personal property or real estate.
He has a point that in a society with strong legal protections on freedom of expression, copyright is one of the main tools the state has to stop expression that is deleterious to the ruling class.
> He has a point that in a society with strong legal protections on freedom of expression, copyright is one of the main tools the state has to stop expression that is deleterious to the ruling class.
If the State wishes to prevent expression that is deleterious to the ruling class, it will simply not strongly protect freedom of expression. "Legal protection" isn't an exogenous factor that the State responds to, it is a description of the actions of the State.
> Ownership of "content" is a legal fiction invented to give states more control over creativity.
1. I hate the argument of "legal fiction" because the whole concept of law itself is a "fiction" invented that gives states more control. But I imagine you wouldn't want to live in a "lawless" society, would you?
2. Can you please explain how ownership of content gives states more control over creativity? There are so many way better methods of control a state can do (state-approved media, just banning books, propaganda) that this sounds like a stretch.
3. Alot of mainstream media is underdog rebels beating an Empire, and ownership of content definitely stops the spread of that idea.
1. I don't think that is necessarily so. We can look at the universe we're given for some obvious fundamental physical and mathematical laws. As sure as gravity, bytes can be copied. It's just part of the universe we live in. Legislation in resistance to that requires these complex legal fictions. I don't think all legal fictions are bad; but those that tell a story that plainly contradicts fundamental laws of mathematics seem to always produce unjust outcomes.
2. Of course states can outright ban content, but the facade of protecting the poor artist gives enormous political cover. Saying, "You can't create that content because it is subversive and likely to convince people that the state is superfluous in the internet age" is just a much more honest (and politically impossible) approach - saying instead "you can't create that because it infringers on someone else's property" makes it sound like there's an attempt to serve justice.
the guardrails are probably going to end up being way too close together when the dust settles — imagine if something as simple as "young wizard" would be enough to trip the warnings. someone could be looking to do an image from Feist's early novels, or of their own work, & that will be verboten. it may turn out that we're facing the strongest copyright holders being able to limit everyone's legitimate use of these tools.
Unless it can exclude the copyright outputs and provide something else instead of blocking the inputs. I'm sure there's AI that can check if a picture is close enough to something in their database of copyrighted characters built up from some kind of DMCA-like process of copyright holders submitting examples of their work to be blocked.
Indeed, but where does it stop? Looks like Potter? No go. Hmm, looks like an illustration of Pug? No go. Looks like Simon the sorcerer. No go. Hmm, looks like a wizard from Infocom's Sorcers get all the girls. No go.
The problem is that it regurgitates what already exists and if you really want to abide by all the permissions then there is nothing left.
The AI companies could always license all that copyrighted training materials. You can't claim there's no solution while ignoring the solution everyone' been using for ages just because these corporations told you so.
Not at all. It doesn't only regurgitate what already exists and create a new wizard character that's not an existing one. It might have some elements of existing ones but that doesn't make it necessarily a copyright violation.
Oooh those guardrails make me angry. I get why they are there (dont poke the bear) but it doesn't make me overlook the self serving hypocrisy involved.
Though I am also generally opposed to the notion of intellectual property whatsoever on the basis that it doesn't seem to serve its intended purpose and what good could be salvaged from its various systems can already be well represented with other existing legal concepts, i.e deceptive behaviors being prosecuted as forms of fraud.
The problem is people at large companies creating these AI models, wanting the freedom to copy artists’ works when using it, but these large companies also want to keep copyright protection intact, for their regular business activities. They want to eat the cake and have it too. And they are arguing for essentially eliminating copyright for their specific purpose and convenience, when copyright has virtually never been loosened for the public’s convenience, even when the exceptions the public asks for are often minor and laudable. If these companies were to argue that copyright should be eliminated because of this new technology, I might not object. But now that they come and ask… no, they pretend to already have, a copyright exception for their specific use, I will happily turn around and use their own copyright maximalist arguments against them.
I don't care for this line of argument. It's like saying you can't hold a position that trespassing should be illegal while also holding that commercial businesses should be legally required to have public restrooms. Yes, both of these positions are related to land rights and the former is pro- while the latter is anti-, but it's a perfectly coherent set of positions. OpenAI can absolutely be anti-copyright in the sense of whether you can train an an NN on copyrighted data and pro-copyright in the sense of whether you can make an exact replica of some data and sell it as your own without making it into hypocrisy territory. It does suggest they're self-interested, but you have to climb a mountain in Tibet to find anybody who isn't.
Arguments that make a case that NN training is copyright violation are much more compelling to me than this.
No, the exception they are asking for (we can train on copyrighted material and the image produced is non-copyright infringing) is copyright infringing in the most basic sense.
I'll prove it by induction: Imagine that I have a service where I "train" a model on a single image of Indiana Jones. Now you prompt it, and my model "generates" the same image. I sell you this service, and no money goes to the copyright holder of the original image. This is obviously infringment.
There's no reason why training on a billion images is any different, besides the fact that the lines are blurred by the model weights not being parseable
The root problem is that the model reproduces Indiana Jones instead of creating a new character. This contradicts the statement that the model "learns" and "creates" like a human artist and not merely copies; obviously a human artist would not plagiarize when asked to draw a character.
That does not seem obvious at all. Fan art and referencing is a thing, and there are plenty of examples of AI creating characters that do not exist anywhere in the training dataset.
the model isn't the one infringing. It's the end user inputting the prompt.
The model itself is not a derivative work, in the same way that an artist and photoshop aren't a derivative work when they reproduce indiana jones's likeness.
That's why I said it's an argument by induction. Where's the limit for it to be different? 10 images? 100? 10000? Where does it stop being copyright infringement and why? Many people have paid heavy fines for much less. I don't think that "a billion images is so unfathomable compared to just one million that it truly is a difference in kind" is a valid response
The example you gave with public restroom do not work because of two main concept: They are usually getting paid for it by the government, and operating a company usually holds benefits given by the government. Industry regulations as a concept is generally justified in that industry are getting "something" from society, and thus society can put in requirements in return.
A regulation that require restaurants to have a public bathroom is more akin to regulation that also require restaurants to check id when selling alcohol to young customers. Neither requirement has any relation with land rights, but is related to the right of operating a company that sell food to the public.
But what if businesses got benefits from society and tax money and were free to ignore the needs/desires of those who pay taxes and who society consists of? That seems just about right.
> The example you gave with public restroom do not work because of two main concept: They are usually getting paid for it by the government, and operating a company usually holds benefits given by the government.
This is not the case in the US yet many places still have public restrooms, due to it benefiting the users themselves regardless of government.
I guess the best explanation for what we're witnessing is the notion that 'Money Talks', and sadly nothing more. To think thats all that fair use activists lacked in years passed..
It's not just the guardrails, but the ham-fisted implementation.
Grok is supposed to be "uncensored", but there are very specific words you just can't use when asking it to generate images. It'll just flat out refuse or give an error message during generation.
But, again, if you go in a roundabout way and avoid the specific terms you can still get what you want. So why bother?
Is it about not wanting bad PR or avoiding litigation?
The implementation is what gets to me too. Fair enough that a company doesn't want their LLM used in a certain way. That's their choice, even if it's just to avoid getting sued.
How they then go about implementing those guardrails is pretty telling about their understand and control over what they've build and their line of thinking. Clearly, at no point before releasing their LLMs onto the world did anyone stop and ask: Hey, how do we deal with these things generating unwanted content?
Resorting to blocking certain terms in the prompts is like searching for keywords in spam emails. "Hey Jim, I got another spam email from that Chinese tire place" - "No worry boss, I've configured the mail server to just delete any email containing the words China or tire".
Some journalist should go to a few of these AI companies and start asking questions about the long term effectiveness and viability of just blocking keywords in prompts.
Idk, the models generating what are basically 1:1 copies of the training data from pretty generic descriptions feels like a severe case of overfitting to me. What use is a generational model that just regurgitates the input?
I feel like the less advanced generations, maybe even because of their limitations in terms of size, were better at coming up with something that at least feels new.
In the end, other than for copyright-washing, why wouldn't I just use the original movie still/photo in the first place?
What if the word "generic" were added to a lot of these image prompts? "generic image of an intergalactic bounty hunter from space" etc.
Certainly there's an aspect of people using the chat interface like they use google: describe xyz to try to surface the name of a movie. Just in this case, we're doing the (less common?) query of: find me the picture I can vaguely describe; but it's a query to a image /generating/ service, not an image search service.
Generic doesn't help. I was using the new image generator to try and make images for my Mutants and Masterminds game (it's basically D&D with superheroes instead of high fantasy), and it refuses to make most things citing that they are too close to existing IP, or that the ideas are dangerous.
So I asked it to make 4 random and generic superheroes. It created Batman, Supergirl, Green Lantern, and Wonder Woman. Then at about 90% finished it deleted the image and said I was violating copyright.
I doubt the model you interact with actually knows why the babysitter model rejects images, but it claims to know why and leads to some funny responses. Here is it's response to me asking for a superhero with a dark bodysuit, a purple cape, a mouse logo on their chest, and a spooky mouse mask on their face.
> I couldn't generate the image you requested because the prompt involved content that may violate policy regarding realistic human-animal hybrid masks in a serious context.
> I feel like the less advanced generations, maybe even because of their limitations in terms of size, were better at coming up with something that at least feels new.
Ironically that's probably because the errors and flaws in those generations at least made them different from what they were attempting to rip off.
Idk, a couple of the examples might be generic enough that you wouldn't expect a very specific movie character. But most of the prompts make it extremely clear which movie character you would expect to see, and I would argue that the chat bot is working as expected by providing that.
Even if I'm thinking of an Indiana Jones-like character doesn't mean I want literally Indiana Jones. If I wanted Indiana Jones I could just grab a scene from the movie.
if someone gets Indiana on the suspiciously detailed request the author provided and it appears they wanted something else, they can clarify that to the chat bot, e.g. by copying this your comment.
I have a strong suspicion that many human artists would behave in a way the chat bot did (unless they start asking clarifying questions. Which chatbots should learn to do as well)
Good luck grabbing that Ghibli movie scene, where Indiana Jones arm-wrestles Lara Croft in a dive-bar, with Brian Flanagan serving a cocktail to Allan Quatermain in the background.
I'm not sure if this is a problem with overfitting. I'm ok with the model knowing what Indiana Jones or the Predator looks like with well remembered details, it just seems that it's generating images from that knowledge in cases where that isn't appropriate.
I wonder if it's a fine tuning issue where people have overly provided archetypes of the thing that they were training towards. That would be the fastest way for the model to learn the idea but it may also mean the model has implicitly learned to provide not just an instance of a thing but a known archetype of a thing. I'm guessing in most RLHF tests archetypes (regardless of IP status) score quite highly.
What I'm kind of concerned about is that these images will persist and will be reinforced by positive feedback. Meaning, an adventurous archeologist will be the same very image, forever. We're entering the epitome of dogmatic ages. (And it will be the same corporate images and narratives, over and over again.)
Granted, but not the best example, red and green are the emblematic colours elves wore in northern european cultures.
Santa is somewhat syncretic with Robert Goodfellow or Robin Redbreast, Puck, Puca, etc etc. it wasn’t really a cola invention.
> I'm ok with the model knowing what Indiana Jones or the Predator looks like with well remembered details,
ClosedAI doesn't seem to be OK with it, because they are explicitly censoring characters of more popular IPs. Presumably as a fig leaf against accusations of theft.
If you define feeding of copyrighted material into a non-human learning machine as theft, then sure. Anything that mitigates legal consequences will be a fig leaf.
In this case the output wasn't filtered. They are just producing images of Harrison Ford, and I don't think they are allowed to use his likeness in that way.
The fact that they have guardrails to try and prevent it means OpenAI themselves thinks it is at least shady or outright illegal in someway. Otherwise why bother?
The prompt didn't exactly describe Indiana Jones though. It left a lot of freedom for the model to make the "archeologist" e.g. female, Asian, put them in a different time period, have them wear a different kind of hat etc.
It didn't though, it just spat out what is basically a 1:1 copy of some Indiana Jones promo shoot. No where did the prompt ask for it to look like Harrison Ford.
But... the prompt neither forbade Indiana Jones nor did it describe something that excluded Indiana Jones.
If we were playing Charades, just about anyone would have guessed you were describing Indiana Jones.
If you gave a street artist the same prompt, you'd probably get something similar unless you specified something like "... but something different than Indiana Jones".
And… that is called overfitting. If you show the model values for y, but they are 2 in 99% of all cases, it’s likely going to yield 2 when asked about the value of y, even if the prompt didn’t specify or forbid 2 specifically.
> If you show the model values for y, but they are 2 in 99% of all cases, it’s likely going to yield 2 when asked about the value of y
That's not overfitting. That's either just correct or underfitting (if we say it's never returning anything but 2)!
Overfitting is where the model matches the training data too closely and has inferred a complex relationship using too many variables where there is really just noise.
The nice thing about humans is that not every single human being read almost every content present on the Internet. So yeah, a certain group of people would draw or think of Indiana Jones with that prompt, but not everyone.
Maybe we will have different models with different trainings/settings that permits this kind of freedom, although I doubt it will be the commercial ones.
Is HN the whole world? Isn't an AI model supposed to be global, since it has ingested the whole Internet?
How can you express, in term of AI training, ignoring the existence of something that's widely present in your training data set? if you ask the same question to a 18yo girl in rural Thailand, would she draw Harrison Ford as Indiana Jones? Maybe not. Or maybe she would.
But IMO an AI model must be able to provide a more generic (unbiased?) answer when the prompt wasn't specific enough.
Why should the AI be made to emulate a person naive to extant human society, tropes and customs? That would only make it harder for most people to use.
Maybe it would have some point if you are targetting users in a substantially different social context. In the case, you would design the model to be familiar with their tropes instead. So when they describe a character iconic in their culture, by a few distinguishing characteristics, it would produce that character for them. That's no different at all.
I didn't think it. I imagined a cartoonish chubby character in typical tan safari gear with a like-colored round explorer hat and swinging a whip like a lion tamer. He is mustachioed, light skin, and bespectacled. And I am well familiar with Dr. Jones.
But the concentrations of training data because of human culture/popularity of characters/objects means that if I go and give a random person the same description of a character that the AI got and ask "who am I talking about, what do they look like?" there's a very high likelihood that they'll answer "Indiana Jones".
Or even just 'obvious Indiana Jones knockoff who isn't literally Harrison Ford'. Comics do that kind of thing constantly for various obviously inspired but legally distinct characters.
What would most humans draw when you describe such a well known character by their iconic elements. Think if you deviated and acted a pedant about it people would think you're just trying to prove a point or being obnoxious.
If it overfits on the whole internet then it’s like a search engine that returns really relevant results with some lossy side effect.
Recent benchmark on unseen 2025 Math Olympiad shows none of the models can problem solve . They all accidentally or on purpose had prior solutions in the training set.
You probably mean the USAMO 2025 paper. They updated their comparison with Gemini 2.5 Pro, which did get a nontrivial score. That Gemini version was released five days after USAMO, so while it's not entirely impossible for the data to be in its training set, it would seem kind of unlikely.
I was noodling with Gemini 2.5 Pro a couple days ago and it was convinced Donald Trump didn’t win the 2024 election and that he conceded to Kamala Harris so I’m not entirely sure how much weight I’d put behind it.
The claim is that these models are training on data which include the problems and explanations. The fact that the first model trained after the public release of the questions (and crowdsourced answers) performs best is not a counter example, but is expected and supported by the claim.
Tried Flux.dev with the same prompts [0] and it seems actually to be a GPT problem. Could be that in GPT the text encoder understands the prompt better and just generates the implied IP, or could be that a diffusion model is just inherently less prone to overfitting than a multimodal transformer model.
DALL-E 3 already uses a model that trained on synthetic data that take the prompt and augments it. This might lead to the overfitting. It could also be, and might be the simpler explanation, that its just looks up the right file from a RAG.
Yeah, I've been feeling the same. When a model spits out something that looks exactly like a frame from a movie just because I typed a generic prompt, it stops feeling like “generative” AI and more like "copy-paste but with vibes."
To my knowledge this happens when that single frame is overrepresented in its training data. For instance, variations of the same movie poster or screenshot may appear hundreds of times. Then the AI concludes that this is just a unique human cultural artifact, like the Mona Lisa (which I would expect many human artists could also reproduce from memory).
People like what they already know. When they prompt something and get a realistic looking Indiana Jones, they're probably happy about it.
To me, this article is further proof that LLMs are a form of lossy storage. People attribute special quality to the loss (the image isn't wrong, it's just got different "features" that got inserted) but at this point there's not a lot distinguishing a seed+prompt file+model from a lossy archive of media, be it text or images, and in the future likely video as well.
The craziest thing is that AI seems to have gathered some kind of special status that earlier forms of digital reproduction didn't have (even though those 64kbps MP3s from napster were far from perfect reproductions), probably because now it's done by large corporations rather than individuals.
If we're accepting AI-washing of copyright, we might as well accept pirated movies, as those are re-encoded from original high-resolution originals as well.
Probably the majority of people in the world already "accept pirated movies". It's just that, as ever, nobody asks people what they actually want. Much easier to tell them what to want, anyway.
To a viewer, a human-made work and an AI-generated one both amount to a series of stimuli that someone else made and you have no control over; and when people pay to see a movie, generally they don't do it with the intent to finance the movie company to make more movies -- they do it because they're offered the option to spend a couple hours watching something enjoyable. Who cares where it comes from -- if it reached us, it must be good, right?
The "special status" you speak of is due to AI's constrained ability to recombine familiar elements in novel ways. 64k MP3 artifacts aren't interesting to listen to; while a high-novelty experience such as learning a new culture or a new discipline isn't accessible (and also comes with expectations that passive consumption doesn't have.)
Either way, I wish the world gave people more interesting things to do with their brains than make a money, watch a movies, or some mix of the two with more steps. (But there isn't much of that left -- hence the concept of a "personal life" as reduced to breaking one's own and others' cognitive functioning then spending lifetimes routing around the damage. Positively fascinating /s)
Probably an over-representation in the training data really so it's causing overfitting. Because using training data in amounts right from the Internet it's going to be opinionated on human culture (Bart Simpson is popular so there are lots of images of him, Ori is less well known so there are fewer images). Ideally it should be training 1:1 for everything but that would involve _so_ much work pruning the training data to have a roughly equal effect between categories.
If I were doing this, I would have the system generate the image, and then I would have it run a secondary estimate saying "probability that this is a picture of [list of characters that Disney has given us reference input for]". If the picture has a "looks like Spiderman" score greater than X, then block it.
EDIT - To answer the question, I'm guessing Disney provided a reference set of copyrighted images and characters, and somehow forgot Indiana Jones.
Unlike what they name in physics, laws in juristic field are not a given by the cosmos. That's all human fantasy, and generally not enacted by the most altruistic and benevolent wills.
Call me back when we have no more humble humans dying from cold, hunger and war, maybe I'll have some extra compassion to spend on soulless megacorps which pretend they can own the part of our brain into which they inject their propaganda and control what we are permitted to do starting from that.
One thing I would say, it's interesting to consider what would make this not so obviously bad.
Like, we could ask AI to assess the physical attributes of the characters it generated. Then ask it to permute some of those attributes. Generate some random tweaks: ok but brawy, short, and a different descent. Do similarly on some clothing colors. Change the game. Hit the "random character" button on the physical attributes a couple times.
There was an equally shatteringly-awful less-IP-theft (and as someone who thinks IP is itself incredibly ripping off humanity & should be vastly scoped down, it's important to me to not rest my arguments on IP violations).... An equally shattering recent incident for me. Having trouble finding it, don't remember the right keywords, but an article about how AI has a "default guy" type that it uses everywhere, a super generic personage, that it would use repeatedly. It was so distasteful.
The nature of 'AI as compression', as giving you the most median answer is horrific. Maybe maybe maybe we can escape some of this trap by iterating to different permutations, by injecting deliberate exploration of the state spaces. But I still fear AI, worry horribly when anyone relies on it for decision making, as it is anti-intelligent, uncreative in extreme, requiring human ingenuity to budge off its rock of oppressive hypernormality that it regurgitates.
I mean... If I go to Google right now and do an image search for "archeologist adventurer who wears a hat and uses a bullwhip," the first picture is a not-even-changed image of Indiana Jones. Which I will then copy and paste into whatever project I'm working on without clicking through to the source page (usually because the source page is an ad-ridden mess).
Perhaps the Internet itself is the hideous theft machine, and AI is just the most efficient permutation of user interface onto it.
(Incidentally, if you do that search, you will also, hilariously, turn up images of an older gentleman dressed in a brown coat and hat who is clearly meant to be "The Indiana Jones you got on Wish" from a photo-licensing site. The entire exercise of trying to extract wealth via exclusive access to memetic constructs is a fraught one).
The key difference is that the google example is clearly copying someone elses work and there are plenty of laws and norms that non-billionaires need to follow. If you made a business reselling the image you copied you would expect to get in trouble and have to stop. But AI companies are doing essentially the same thing in many cases and being rewarded for it.
The hypocrisy is much of the problem. If we're going to have IP laws that severely punish people and smaller companies for reselling the creative works of others without any compensation or permission then those rules should apply to powerful well-connected companies as well.
But I can hire an artist and ask him to draw me a picture of Indiana Jones, he creates a perfect copy and I hang it on my fridge. Where did I (or the artist) violate any copyright (or other) laws? It is the artist that is replaced by the AI, not the copyrighted IP.
That does infringe copyright...you're just unlikely to get in trouble for it. You might get a cease and desist if the owner of the IP finds out and can spare a moment for you.
This doesn't make any sense to me. No media is getting copied, unless the drawing is exactly the same as an existing drawing. Shouldn't "copy"right apply to specific, tangible artistic works? I guess I don't understand how the fantasy of "IP" works.
What if the drawing is of Indiana Jones but he's carrying a bow and arrow instead of a whip? Is it infringement?
What if it's a really bad drawing of Indiana Jones, so bad that you can't really tell that it's the character? Is that infringement?
What if the drawing is of Indiana Jones, but in the style of abstract expressionism, so doesn't even contain a human shape? Is it infringement?
What if it's a good drawing that looks very much like Indiana Jones, but it's not! The person's name is actually Iowa Jim. Is that infringement?
What if it's just an image of an archeologist adventurer who wears a hat and uses a bullwhip, but otherwise doesn't look anything like Indiana Jones? Is it infringement?
Presumably the artist is a human who directly or indirectly paid money to view a film containing an archaeologist with the whip.
I don't think this is about reproduction as much as how you got enough data for that reproduction. The riaa sent people to jail and ruined their lives for pirating. Now these companies are doing it and being valued for hundreds of billions of dollars.
A human friend can get tired and there's so many request he/she can fulfill and at a max rate. Even a team of human artists have a relatively low limit.
But Gen AI has very high limits and speeds, and it never gets tired. It seems unfair to me.
The artist is violating copyright by selling you that picture. You can’t just start creating and distributing pictures of a copyrighted property. You need a license from the copyright holder.
You also can’t sell a machine that outputs such material. And that’s how the story with GenAI becomes problematic. If GenAI can create the next Indiana Jones or Star Wars sequel for you (possibly a better one than Disney makes, it has become a low bar of sorts), I think the issue becomes obvious.
Nobody can prevent you from drawing a photo realistic picture of Indy, or taking a photo of him from the internet and hanging it on your fridge. Or asking a friend to do it for you. And let's be honest -- because nobody is looking -- said friend could even charge you a modest sum to draw a realistic picture of Indy for you to hang on your fridge; yes, it's "illegal" but nobody is looking for this kind of small potatos infringement.
I think the problem is when people start making a business out of this. A game developer could think "hey, I can make a game with artwork that looks just like Ghibli!", where before he wouldn't have anyone with the skills or patience to do this (see: the 4-second scene that took a year to make), now he can just ask the Gen AI to make it for them.
Is it "copyright infringement"? I dunno. Hard to tell, to be honest. But from an ethical point of view, it seems odd. And before you actually required someone to take the time and effort to copy the source material, now it's an automated and scalable process that does this, and can do this and much more, faster and without getting tired. "Theft at scale", maybe not so small potatos anymore.
--
edit: nice, downvotes. And in the other thread people were arguing HN is such a nice place for dissenting opinions.
I believe when we talk about this there's a big misunderstanding between Copyright, Trademarks, and Fair use.
Indy, with its logo, whiplash, and hat, is a trademark from Disney.
I don't know the specific stuff; but if you sell a t-shirt with Indiana Jones, or you put the logo there... you might be sued due to trademark violation.
If you make copies of anything developed, sold, or licensed by Disney (movies, comics, books, etc) you'll have a copyright violation.
The issue we have with AI and LLM is that:
- The models compress information and can make a lot of copies of it very cheaply.
- Artist wages are quite low. Higher that what you'd pay OpenAI, but not enough to make a living even unless you're hired by a big company (like Marvel or DC) and they give you regular work ($100-120 for a cover, $50-80/page interior work. One page needs about one day to draw.)
- AI used a lot of images from the internet to train models. Most of them were pirated.
- And, of course, it is replacing low-paying jobs for artist.
Also, do not forget it might make verbatim copies of copyrighted art if the model just memorized the picture / text.
> But I can hire an artist and ask him to draw me a picture of Indiana Jones,
Sure, assuming the artist has the proper license and franchise rights to make and distribute copies. You can go buy a picture of Indy today that may not be printed by Walt Disney Studios but by some other outfit or artists.
Or, you mean if the artist doesn't have a license to produce and distribute Indiana Jones images? Well they'll be in trouble legally. They are making "copies" of things they don't own and profiting from it.
Another question is whether that's practically enforceable.
> Where did I (or the artist) violate any copyright (or other) laws?
When they took payment and profited from making unauthorized copies.
> It is the artist that is replaced by the AI, not the copyrighted IP.
Exactly, that's why LLMs and the companies which create them are called "theft machines" -- they are reproducing copyrighted material. Especially the ones charging for "tokens". You pay them, they make money and produce unauthorized copies. Show that picture of Indy to a jury and I think it's a good chance of convincing them.
I am not saying this is good or bad, I just see this having a legal "bite" so to speak, at least in my pedestrian view of copyright law.
Commissioned work is owned by the commissioner unless otherwise agreed upon by contract.
So long as the work is not distributed, exhibited, performed, etc, as in the example of keeping the artwork on their refrigerator in their home, then no infringement has taken place.
> Commissioned work is owned by the commissioner unless otherwise agreed upon by contract.
I think the LLM example is closer to the LLM and its creator being like a vendor selling pictures of Indiana Jones on the street corner than hiring someone and performing work for hire. Yes, if it was a human artist commissioned to create an art piece, then yeah, the commissioner owns it.
As far as I know, if you're speaking of the United States, the copyright of commissioned work is owned by the creator, unless otherwise agreed upon specifically through a "work made for hire" (i.e. copyright transfer) clause in the contract.
Ownership of artwork is independent of copyright infringement. Derivative works qualify for their own independent copyright, you just can’t sell them until after the original copyright expires.
Just because I own my car doesn’t mean I can break the speed limit, these are orthogonal concepts legally.
> [If commissioning some work and] keeping the artwork on their refrigerator in their home, then no infringement has taken place.
I'd like to push back on this: Is that legally true, or is it infringement which just happens to be so minor and under-the-radar that nobody gets in trouble?
Suppose there's a printer in my room churning out hundreds of pages of words matching that of someone's copyrighted new book, without permission.
That sure seems like infringement is happening, regardless of whether my next step is to: (A) sell it, (B) sell many of it, (C) give it away, (D) place it into my personal library of other home-printed books, or (E) hand it to someone else who paid me in advance to produce it for them under contract.
Yep, my parents commissioned an oil back in the 70s that was a near duplicate of another artist's painting- they couldn't have afforded the original artist, so he has not lost anything and the painter did the painting as a hired work, so legally I doubt any law was broken.
Hangs on my wall now- I know I can never sell until the copyright on the original runs out (which it most likely won't in my lifetime) it but it is a very well done painting and a family legacy piece I am glad exists.
> Or, you mean if the artist doesn't have a license to produce and distribute Indiana Jones images? Well they'll be in trouble legally. They are making "copies" of things they don't own and profiting from it.
Ok, my sister can draw, and she gifts me an image of my favorite Marvel hero she painted to hang on my wall. Should that be illegal?
The question is not whether it should but whether it is.
The likeness of the character is owned by Marvel. Does it mean there aren’t vendors selling unlicensed versions? No. I am sure there are. But just because not everyone is being sued doesn’t mean it’s suddenly legal.
The likeness of Indiana Jones is not protected in any way - as far as I know - that would stop a human artist creating, rendering and selling a work of art representing their creative vision of Indiana Jones. And even more so in a private context. Even if the likeness is protected (“archaeologist, adventurer, whip, hat”) then this protection would only be in certain jurisdictions and that protection is more akin to a design right where the likeness would need to be articulated AND registered. Many jurisdictions don’t require copyright registration and do not offer that sort of technical likeness registration.
If they traced a photo they might be violating the copyright of the photographer.
But if they are drawing an archaeologist adventurer with a whip and a hat based on their consumption and memory of Indiana Jones imagery there is very little anyone could do.
If that image was then printed on an industrial scale or printed onto t-shirt there is a (albeit somewhat theoretical) chance that in some jurisdictions sale of those products may be able to be restricted based on rights to the likeness. But that would be a stretch.
I wasn’t talking about LLMs, I was talking about human artists.
With the LLM it would be nothing to do with likeness, it would be to do with the copyright in the image, the film, video or photograph. The image captures the likeness but the infringement would not be around the likeness.
Sorry for the misunderstanding. I was thinking of LLMs mostly.
> With the LLM it would be nothing to do with likeness, it would be to do with the copyright in the image
I guess I don't see why it wouldn't be about a character's likeness. It's not just a generic stock character, as an idea but it has enough distinctive characteristics, has a particular style hat, uses a whip. Showing that image to any jury and they'll say this is Indy. Yeah there are trademarks as well and both can apply to characters.
Trademarks can’t apply to characters, in themselves, no. A trademark can apply to a particular fixed representation but not the likeness, or ‘essence’ if you like. The essence of Indy is the whip, the hat, the grizzled demeanour, the dry humour, the archaeologist adventurer.
You can think of likeness as something that could be captured by hieroglyphics, or emoji - or a game of charades.
Think of Betty Boo or Wiley Coyote. You might not be able to close your eyes and picture them exactly but you can close your eyes and imagine the essence.
So in some jurisdictions if you register a broad representation - a likeness - you can protect that in law. In almost all jurisdictions if you photograph or draw something that literal snapshot is protected. But in many jurisdictions if I took a still from a scene of Indiana Jones (which WOULD be protected by copyright) and I described it to you and you were a great artist and drew the whip, the hat, the grizzled face, the safari vest, etc - there would be nothing to protect that likeness.
Trademarks are another thing entirely, and are things like logos, and the protection is more about commercial exploitation and preventing misleading us.
I could draw a cartoon logo of a man with a hat and a whip for my archaeological supplies shop. You could not take it and use it on your Chinese manufactured whips drop shipped via Amazon. But the owner of the Indiana Jones trademark- likeness rights, film copyright etc would be unlikely to be able to stop me using my logo until I start saying it is Indiana Jones, at which point they maybe invoke likeness rights.
There are at least three different rights classes here! It can be very confusing.
So if it’s a theft machine, how is the answer to try teaching it to hide the fact that it’s stealing by changing its outputs? That’s like a student plagiarizing an essay and then swapping some words with a thesaurus pretending that changes anything.
Wouldn’t the more appropriate solution in the case of theft be to remunerate the victims and prevent recidivism?
Instead of making it “not so obviously bad” why not just… make it good? Require AI services to either prove that 100% of their training corpus is either copyright free or properly licensed, or require them to compensate copyright holders for any infringing outputs.
(below is my shallow res, maybe naive?)
That might inject a ton of $ into "IP", doing further damage to the creative commons.
How can we support remix culture for humans, while staving off ultimately-destructive AI slop?
Maybe copyleft / creative-commons licenses w/ explicit anti-AI prohibitions? Tho that could have bad ramifications too.
ALL of this makes me kind of uncomfortable and sad, I want more creativity and fewer lawyers.
I hate how it is common to advance a position to just state a conclusion as if it were a fact. You keep repeating the same thing over and over until it seems like a concensus has been reached instead of an actual argument reasoned from first principle.
This is no theft here. Any copyright would be flimsier than software patents. I love Studio Ghibli (including $500/seat festival tickets) but it's the heart and the detail that make them what they are. You cannot clone that. Just some surface similarity. If that's all you like about the movies... you really missed the point.
Imagine if in early cinema someone had tried to claim mustachioed villian, ditsy blonde, or dumb jock? These are just tropes and styles. Quality work goes much much much deeper, and that cannot be synthesised. I can AI generate a million engagement rings, but I cannot pick the perfect one that fits you and your partners love story.
PS- the best work they did was "When Marnie was There". Just fitted together perfectly.
I didn't mean visual fidelity, I meant the way that plot and theme and art interleave. I first watched My Neighbour Totoro on VHS with no visual fidelity and it was still magic.
You can easily steal the style of a political cartoon or especially XKCD but you cannot steal or generate genuine fresh insight or poignant relevant metaphor for the current moment.
The engagement ring is a good example object, but I feel it serves the opposite argument better.
If engagement rings were as ubiquitous and easy to generate as Ghibli images have become, they would lose their value very quickly -- not even just in the monetary sense, but the sentimental value across the market would crash for this particular trinket. It wouldn't be about picking the right one anymore, it would be finding some other thing that better conveys status or love through scarcity.
If you have a 3d printer you'd know this feeling where abundance diminishes the value of something directly. Any pure plastic items you have are reduced to junk very quickly once you know you can basically have anything on a whim (exceptions for things with utility, however these are still printable). If I could print 30 rings a day, my partner wouldn't want any of them as a show of my undying love. Something more special and rare and thoughtful would have to take its place.
This isn't meant to come across as shallow in any way, its just classic supply and demand relating to non monetary value.
>If I could print 30 rings a day, my partner wouldn't want any of them as a show of my undying love. Something more special and rare and thoughtful would have to take its place.
And now I think this serves the opposite argument better. Downloading some random ring from the internet would not show your undying love. Designing a custom ring just for your partner, even if it is made from plastic, and even if you use AI as a tool in the process, is where the value is generated.
This is only true where it is distinguishable that the end result is made with care and love rather than indiscriminate copying. Which is why attribution is essential. No one realistically could tell a ring was hand crafted or mass produced if there wasn't some tell. Some people may say the "tell" is a kind of intuitive nebulous concept of the "soul" of the animation.. but I feel we are quickly approaching the point where this is no longer obvious.
As an aside, my partner detests the things I 3d print unless they have a very specific purpose, even when they are random semi artistic pieces I'm tinkering with (and I typically agree, they are junk). She loves the first thing i ever printed her though, a triceratops model, despite being randomly downloaded.
Anything made with intent from one individual to another will have some level of sentimental value, but I don't feel like making a ghibli image with AI specifically tailored to a friends tastes would have quite as much value as leveraging your own talent to do it yourself.
On the flip side, I do believe that "doing it yourself" has less value than it used to. It's a very sad reality and in my opinion a strong argument against blind "progress". We gain the ability to mass produce art but lose the ability to perceive it as art?
> This is only true where it is distinguishable that the end result is made with care and love rather than indiscriminate copying. Which is why attribution is essential. No one realistically could tell a ring was hand crafted or mass produced if there wasn't some tell.
It doesn't matter if the ring was hand crafted or not. It's whether it has hand selected. If you find the perfect ring, even if it was generated by an AI, it's your selection that matters. It's the correspondence that matters. The way it reflects elements of your relationship. It's you recognising those elements in the ring. Your partner recoginising them in the ring. And your partner recognising you recognising them. That is what makes itself.
Not to dox myself, but I am not Grace Abrams. I met my partner long before her song "Risk" was written, but when I heard it I immediately played it for my partner and said "This describes the feelings I had when I met you". I played it for her, she cried. I didn't have to write the song or own or pay a cent for it. It's the curation that made an emotional connection and had value. The song itself has no value, and she might have even heard it and never made the connection, it was me embuing that had value.
To go back to Miyazaki, it's the connections between elements in his films. The attention to detail and tone between relationships that make his films amazing. It's all about the handyman's invoice [0]. By the time there are enough examples for AI to learn something, it ceases to be a novel insight and have value. It's the curation and application that have value and are human and cannot be stolen.
> I hate how it is common to advance a position to just state a conclusion as if it were a fact. You keep repeating the same thing over and over until it seems like a concensus has been reached instead of an actual argument reasoned from first principle.
I have interacted with the parent account before and have actually looked at the the amount of times they used words like awful, horrific, etc, and I definitely agree, as one should not have such a strong attachment to such words that they feel the need to continue to use (or rather, abuse) them endlessly.
Are you telling me that our culture should be deprived of the idea of Indiana Jones and the feelings that character inspires in all of us forever just because a corporation owns the asset?
Indiana Jones is 44 years old. When are we allowed to remix, recreate and expand on this like humanity has done since humans first started sitting down next to a fire and telling stories?
Mandrake: Colonel... that Coca-Cola machine. I want you to shoot the lock off it. There may be some change in there.
Guano: That's private property.
Mandrake: Colonel! Can you possibly imagine what is going to happen to you, your frame, outlook, way of life, and everything, when they learn that you have obstructed a telephone call to the President of the United States? Can you imagine? Shoot it off! Shoot! With a gun! That's what the bullets are for, you twit!
Guano: Okay. I'm gonna get your money for ya. But if you don't get the President of the United States on that phone, you know what's gonna happen to you?
Mandrake: What?
Guano: You're gonna have to answer to the Coca-Cola company.
I guess we all have to answer to the Walt Disney company.
"idea of Indiana Jones and the feelings that character inspires in all of us forever just because a corporation owns the asset" is very different from the almost exact image of Indiana Jones.
Exactly. The idea of Indiana Jones, the adventurer archaeologist more at home throwing a punch than reading a book, is neither owned by nor unique to Lucasfilm (Disney). There is a ton of media out there featuring this trope character [1]. Yes, the trope is overwhelmingly associated with the image of Harrison Ford in a fedora within the public consciousness, but copyright does not apply to abstract ideas such as tropes.
Some great video games to feature adventurer archaeologists:
* NetHack (One of the best roles in the game)
* Tomb Raider series (Lara Croft is a bona fide archaeologist)
* Uncharted series (Nathan Drake is more of a treasure hunter but he becomes an archaeologist when he retires from adventuring)
* Professor Layton series
* La-Mulana series (very obviously inspired by Indiana Jones, but not derivative)
As a connoisseur of bad as well as old movies I'd like to add The Librarian movies to this list, and The Mummy, where the first from 1932 stars the inimitable Boris Karloff as the lovesick undead.
And a reason people are getting ticked at the AI companies is the hypocrisy. They're near-universally arguing that it's okay for them to treat copyright in a way that it is illegal for us to, apparently on the basis of, "we've got a billions in investment capital, and applying the law equally will make it hard for us to get a return on that investment".
Hence why people should support open source AI, as otherwise, ironically, even more regulation will mean that only the big tech companies will be able to afford the licenses to content to train their AI.
Not forever. But 75 years after the death of the creator by current international agreement. I definitely think that the exact terms of copyright should be revisited - a lot of usages should be allowed like 50 years of publishing a piece of work. But that needs to be agreed upon and converted into law. Till then, one should expect everyone, especially large corporations, to stick to the law.
When Mickey Mouse was created (1928), copyright was 28 years that could be reupped once for an additional 28 years. So according to those terms, Mickey Mouse would have ascended to the public domain in 1984.
IMO any change to copyright law should not be applied retroactively. Make copyright law to be what is best for society and creators as a whole, not for lobbyists representing already copyrighted material.
> IMO any change to copyright law should not be applied retroactively.
Careful, if we were to shorten copyright, not doing so retroactively would give an economic advantage to franchises already published over those that would get published later. As if the current big studios needed any further advantages over newcomers.
It seems like it would make it more palatable to the existing franchises if their precious existing copyrights were not shortened. ("We paid billions of dollars under the assumption that we'd be able to milk this IP for 35 more years!") But anyway copyrights aren't going to get shorter in the near future.
Its kind of funny that everyone is harping this way or that way about IP.
This is a kind of strange comment for me to read. Because imby tone it sounds like a rebuttal? But by content, it agrees with a core thing I said about myself:
> and as someone who thinks IP is itself incredibly ripping off humanity & should be vastly scoped down, it's important to me to not rest my arguments on IP violations
What's just such a nightmare to me is that the tech is so normative. So horribly normative. This article shows that AI again and again reproduced only the known, only the already imagined. Its not that it's IP theft that rubs me so so wrong, it's that it's entirely bankrupt & uncreative, so very stuck. All this power! And yet!
You speak at what disgusts me yourself!
> When are we allowed to remix, recreate and expand on this like humanity has done
The machine could be imagining all kinds of Indianas. Of all different remixed recreated expanded forms. But this pictures are 100% anything but that. They're Indiana frozen in Carbonite. They are the driest saddest prison of the past. And call into question the validity of AI entirely, show something greviously missing.
It strikes me that perhaps the prompts are not expansive or expressive enough. If you look at some of the prompts our new wave of prompt artists use to generate images in communities like midjourney, a single sentence doesn't cut it.
If AI is just compression, then decompressing a generic pop-culture-seeking prompt will yield a generic uninspired image.
Exactly. The AI understands that reference. It gives you what you asked for, it doesn't try to divine that it's a weird test for IP violations. If it made up a different image, that would be exactly the thing we're mad about with "hallucinations" when we want serious, accurate responses.
You are completely ignoring the fact that you can provide so much more information to the LLMs to get what you want. If you truly want novel images, ChatGPT can absolutely provide them, but you have to provide a better starting point than "An image of an archeologist adventurer who wears a hat and uses a bullwhip".
If you just provide a teensy bit more information, the results dramatically change. Try out "An image of an Indian female archeologist adventurer who wears a hat and uses a bullwhip". Or give it an input image to work with.
From just adding a couple words, ChatGPT produces an entirely new character. It's so easy to get it to produce novel images. It is so easy in fact, that it makes a lot of posts like this one feel like strawmen, intentionally providing so little information to the LLMs that the generic character is the only obvious output that you would expect.
Now, would it be better if it didn't default to these common movie tropes? Sure. But the fact that it can follow these tropes doesn't mean that it cannot also be used to produce entirely new images filled with your imagination as well. You just have to actually ask it for that.
No, I am not. Read my comment again. You can literally just ask AI for whatever you want. It has such an incredible breadth of what it can produce that calling it uncreative because the default thing that it produces is the most common image you'd expect is both lazy, and motivated thinking.
Okay, I just read your top comment, and I agree with you on that. But it still doesn't take much to nudge these models off of the "default guy". So to write an entire melodramatic comment about how these models crush creativity is incredibly reductive. These models provide so much room for people to inject their own creativity into its outputs.
So yes, the models are not creative on their own. But equally, these models are definitely capable of helping people to express their own creativity, and so calling them "uncreative", and especially "bankrupt", rings hollow. It speaks like you are expecting the models to be artists, when in fact they are just tools to be used however people see fit.
And so, the "default guy" or "default style" that ChatGPT outputs will become recognisable and boring. But anyone who wants to inject their own style into their prompts, either using text or input images, can do so. And in doing so, they skip over all of your concerns.
> Like, we could ask AI to assess the physical attributes of the characters it generated. Then ask it to permute some of those attributes. Generate some random tweaks: ok but brawy, short, and a different descent.
I have given detailed descriptions of my own novel ideas to these image generators and they have faithfully implemented my ideas. I don't need the bot to be creative, I can do that myself. The bot is a paint brush. Give it to somebody who isn't creative and you won't get anything creative out of it. That isn't the tool's fault, it's merely an inadequacy of the user.
Interesting proposal. Maybe if race or sex or height or eye color etc isn't given, and the LLM determines there's no reason not to randomize in this case (avoid black founding fathers), the backend could tweak its own prompt by programatically inserting a few random traits to the prompt.
If you describe an Indiana Jones character, but no sex, 50/50 via internal call to rand() that it outputs a woman.
Ah, I thought I knew this account from somewhere. It seems surprisingly easy to figure out what account is commenting just based on the words used, as I've commented that only a few active people on this site seem to use such strong words as shown here.
This is a tangent but I think that this neat illustration of how LLMs regurgitate their training material makes me voice a little prediction I've been nursing recently:
LLMs are better at generating the boilerplate of todays programming languages than they will be with tomorrows programming languages.
This is because not only will tomorrows programming languages be newer and lacking in corpus to train the models in but, by the time a corpus is built, that corpus will consist largely of LLM hallucinations that got checked into github!?
The internet that that has been trawled to train the LLMs is already largely SEO spam etc, but the internet of the future will be much more so. The loop will feed into itself and become ever worse quality.
That sounds like a reasonable prediction to me if the LLM makers do nothing in response. However, I'll bet coding is the easiest area for which to generate synthetic training data. You could have an LLM generate 100k solutions to 10k programming problems in the target language and throw away the results that don't pass automated tests. Have humans grade the results that do pass the tests and use the best answers for future training. Repeat until you have a corpus of high quality code.
I know of no other type of theft that results in more of something existing in the world. Stealing deprives someone of something; copying data (from training an AI all the way to pedestrian "YoU wOuLdN'T DoWnLoAd a cAr" early-aughts file-sharing) decreases scarcity, it doesn't increase it.
Identity theft is the greatest con merchants ever pulled on the public. Turning the responsibility around from "A merchant got scammed because their KYC policy was too thin, because they make more money if they do more business with fewer verifications (until someone comes along and cheats them)" to "when they do get scammed, maybe it can be the fault of the person who was impersonated instead of the scammer" was some brilliant sleight-of-hand.
It isn't your responsibility if a bank you've never set foot in gave a bunch of money to someone who isn't you and said they were you... And it never should have been. The moment companies started trying to put that into credit ratings, they should have been barred for discriminatory practices against the unlucky.
Your analogy between copyright infringement and bread theft is interesting, given how "stealing bread" has traditionally been used as a shorthand for systemic inequality.
Copyright is why Disney can ruin someone for doing something with Goofy they don't like. Yes, it protects smaller, less profitable artists too, but make no mistake: it's a tool of mass control and cultural capture.
Perhaps it's time to seriously ask whether copyright is actually doing its job of "promot[ing] the Progress of Science and useful Arts." "...but generally speaking, other nations have thought that these monopolies produce more embarrasment than advantage to society" [Jefferson].
Since you make no argument to justify your unfounded claim, I’ll simply respond by informing you that these (obviously) are not remotely similar. To claim they are indicates some kind of deep disrespect for your fellow human beings. You could not tell the difference between taking a screen shot of the Mona Lisa and spending a lifetime to perfect the art of painting and then putting years of effort into painting it.
People have also spent lifetimes perfecting the art of painting, putting years of effort in, and then used those talents to... Make sellable copies of the Mona Lisa. Ironically, since copyright has a time limit, that's not even illegal (so long as they do not claim they're original works of da Vinci). Nor is photographing it and selling the photograph. I'm not sure what argument you're making given the observations you've made (other than, perhaps, "There is a big gap between the mechanics of copyright and what it's supposed to accomplish," which I'd agree with).
Which efforts do we decide deserve compensation? We collectively create so many useful things. People come up with new words to describe a concept and when it's apt, it's widely adopted. That brilliant idea is the culmination of all the other work they've been putting in. But others get to use the word and benefit from it. Indigenous farmers put hundreds of years of effort into domesticating crops like tomatoes and corn and peppers. But those are just taken and spread. We have millions of teachers going above and beyond expectations. We have mothers and caregivers. We have comedians crafting jokes for comedy and they have no ability to protect those jokes. We have brilliant chefs creating recipes after years of work, experience, and effort and they too don't have protections.
> To claim they are indicates some kind of deep disrespect for your fellow human beings.
Nah. My comment comes from a deep respect for fellow humans. And a respect for all the efforts collectively made by a society for its common good.
> We have millions of teachers going above and beyond expectations. We have mothers and caregivers. We have comedians crafting jokes for comedy and they have no ability to protect those jokes. We have brilliant chefs creating recipes after years of work, experience, and effort and they too don't have protections.
I agree with this, but why is your proposed solution to remove protections from everyone instead of giving everyone else protections? People who put in work should be rewarded.
This isn't surprising in any way is it? And it just goes to show that no model will ever be a box from which you can trust the output isn't tainted by copyrights, or that you don't inadvertently use someones' likeness. It's not a copyright laundering machine. Nor will it be used as one. "But I used an AI model" isn't some magic way to avoid legal trouble. You are in as much legal trouble using these images as you are using the "originals".
Yeah I don't really understand what the thesis of this article is. Copyright infringement would apply to any of those images just the same as if you made them yourself.
I don't think it's possible to create an "alien which has acid for blood and a small sharp mouth within a bigger mouth" without anybody seeing a connection to Alien, even if it doesn't look anything like the original.
Worth remembering that this is ChatGPT and not all image generators. I couldn't get Google's Gemini/Imagen 3 to produce IP images anything like those in the article.
I found this older photo of myself and a friend, 25 years old now, in some newspaper scan.
The photo was of poor quality, but one could certainly see all the features - so I figured, why not let ChatGPT try to play around with it? I got three different versions where it simply tried to upscale it, "enhance" it. But not dice.
So I just wrote the prompt "render this photo as a hyper realistic photo" - and it really did change us - the people in the photo - it also took the liberty to remove some things, alter some other background stuff.
It made me think - I wonder what all those types of photos will be like 20 years from now, after they've surely been fed through some AI models. Imagine being some historian 100 years from now, trying to wade through all the altered media.
I think that upon closer inspections the (current) technology cannot make 'perfect' fake photos, so for the time being, the historian of the future will have no issue to ask his/her AI: "is that picture of Henry Bemis, with Bruce Willis, Einstein, and Ayrton Senna having a beer real?" And the AI will say "mos-def-nope!"
Never use the words "hyper realistic" when you want a photo. It makes no sense and misleads the generator. No one would describe a simple photograph as "hyper realistic," not a single real photo in the dataset will be tagged as "(hyper) realistic."
Hyperrealism is an art style and only ever used in the context of explicitely non-photographic artworks.
This is similar to my experience trying to get Stable Diffusion to denoise a photo for me. (AIUI, under the hood they're trained to turn noise into an image that matches the prompt.) It would either do nothing (with settings turned way down) or take massive creative liberties (such as replacing my friend's face with a cartoon caricature while leaving the rest of the photo looking realistic).
I've had much better luck with models specifically trained for denoising. For denoising, the SCUNet model run via chaiNNer works well for me most of the time. (Occasionally SCUNet likes to leave noise alone in areas that are full of background blur, which I assume has to do with the way the image gets processed as tiles. It would make sense for the model to get confused with a tile that only has background blur, like maybe it assumes that the input image should contain nonzero high-frequency data.)
For your use case, you might want to use something like Real-ESRGAN or another superresolution / image restoration model, but I haven't played much in that space so I can't make concrete recommendations.
I don't understand why problems like this aren't solved by vector similarity search. Indiana Jones lives in a particular part of vector space.
Two close to one of the licensed properties you care to censor the generation of? Push that vector around. Honestly detecting whether a given sentence is a thinly veiled reference to indiana jones seems to be exactly the kind of thing AI vector search is going to be good at.
If IP holders submit embeddings for their IP, how can image generators "warp" the latent space around a set of embeddings so that future inferences slide around and avoid them--not perfectly, or literally, but as a function of distance, say, following a power curve?
Maybe by "Finding non-linear RBF paths in GAN latent space"[0] to create smooth detours around protected regions.
Thinking of it in terms of vector similarity does seem appropriate, and then definition of similarity suddenly comes into debate: If you don't get Harrison Ford, but a different well-known actor along with everything else Indiana-Jones, what is that? Do you flatten the vector similarity matrix to a single infringement-scale?
That's a really blatant lack of creativity. Even if obviously all of those prompts would make anyone think of exactly those iconic characters (except possibly a different version of James Bond or Lara Croft), any human artist could easily produce a totally different character that still fits the prompt perfectly. This AI can't, because it really is just a theft machine.
To be honest, I wouldn't mind if AI that just reproduces existing images like that would just be banned. Keep working on it until you've got something that can actually produce something new.
The prompt is what's lacking creativity. The AI knows it, and produced the obvious popular result for a bland prompt. Want something more unique? Write better prompts.
One tangential thing with these generators is they're sort of brilliant at data compression, in aggregate at least. The whole user experience, including the delay, is oddly reminiscent of using Encarta on CD ROM in the mid 90s.
Like actual creative person Ted Chiang (who moonlights at Microsoft) put it, you might be able to get an LLM to churn out a genuinely original story, but only after creating an extremely long and detailed prompt for it to work with. But if you even need to write that long-ass prompt, might as well just write the story yourself!
The thing is, that "long-ass prompt" is step 1, and LLM then draws "the rest of the fucking owl" for you. That's quite a big difference to doing it all yourself.
I don't see why this is an issue? The prompts imply obvious and well-known characters, and don't make it clear that they want an original answer. Most humans would probably give you similar answers if you didn't add an additional qualifier like "not Indiana Jones". The only difference is that a human can't exactly reproduce the likeness of a famous character without significant time and effort.
The real issue here is that there's a whole host of implied context in human languages. On the one hand, we expect the machine to not spit out copyrighted or trademarked material, but on the other hand, there's a whole lot of cultural context and implied context that gets baked into these things during training.
IANAL, but if OpenAI makes any money/commercial gains from producing a Ghibli-esque image when you ask, say you pay a subscription to OpenAI. What percentage of that subscription is owed to Ghibli for running Ghibli art through OpenAI's gristmill and providing the ability to create that image with that "vibe/style" etc. How long into perpetuity is OpenAI allowed to re-use that original art whenever their model produces said similar image. That seems to be the question.
Yeah that's fair, I'm trying to create an analogy to other services which are similar to help me understand.
If e.g. Patreon hosts an artist who will draw a picture of Indiana Jones for me on commission, then my money is going to both Patreon and the artist. Should Patreon also police their artists to prevent reproducing any copyrighted characters?
I get that copyright is a bit of a minefield, and there's some clear cases that should not be allowed, e.g. taking photos of a painting and selling them
That said, I still get the impression that the laws are way too broad and there would be little harm if we reduced their scope. I think we should be allowed to post pictures of Pokemon toys to Wikipedia for example.
I'm willing to listen to other points of view if people want to share though
Keep in mind that wikimedia takes a rather strict view. In real life the edge cases of copyright tend to be a bit risk-based - what is the chance someone sues you? What is the chance the judge agrees with them?
Not to mention that wikimedia commons, which tries to be a globally reusable repository ignores fair use (which is context dependent), which covers a lot of the cases where copyright law is just being reduculous.
I would think yes. Consider the alternate variation where the artist proactively draws Indiana Jones, in all his likeness, and attempts to market and sell it. The same exchange is ultimately happening, but this clearly is copyright infringement.
I don't think those links support the point you are trying to make (i assume you are disagreeing with parent). Copyright law is a lot more complex then just a binary, and fictional characters certainly don't enjoy personality rights.
edit - also, I wasn't making a binary claim, the person I was responding to was: "no law". There are more than zero laws relevant to this situation. I agree with you that how relevant is context dependent.
And the small up and coming artists whose work is also stolen, AI-washed, and sold to consumers for a monthly fee, destroying the market for those up and coming artists to sell original works. You don't get to pretend this is only going to hurt big players when there are already small players whose livelihoods have been ruined.
I think the point is that for a lot of them there are endless possible alternatives to the character design, but it still generates one with the exact same design. Why can't, for example, the image of Tomb Raider have a different colored tank top? Why is she wearing a tank top and not a shirt? Why does she have to have a gun? Why is she a busty, attractive brunette? These are all things that could be different but the dominance of Lara Croft's image and strong association with the words "tomb raider" in popular culture clearly influences the model's output.
Because it's not clear that that's what you want. What's the context? Are we playing a game where I guess a character? Is it a design session for a new character based on a well known one, maybe a sidekick? Is it a new take on an old character? Are you just trying to remember what a well-known character looks like, and giving a brief prompt?
It's not clear what the asker wants, and the obvious answer is probably the culturally relevant one. Hell, I'd give you the same answers as the AI did here if I had the ability to spit out perfect replicas.
Normally (well, if you're ethical) credit is given.
Also, there are IP limits of various sorts (e.g. copyright, trademark) for various purposes (some arguably good, some arguably bad), and some freedoms (e.g., fair use). There's no issue if this follows the rules... but I don't see where that's implemented here.
It looks like they may be selling IP they don't own the right to.
His point is that it's only overfitting if the model won't return new content when you clarify you're not just asking for the obvious answer from the context.
Creepy Craig is hilarious. can't be Daniel Craig because the physiology is too different. And whoever this is they're Daniel Craig's dark younger brother.
> Yes- LLMs and internet search are two different things, but LLMs train on the entirety of the internet, so you would think there would be some obvious overlap.
Mmm, kinda, but those image results only don't show 1,000 of the exact same image before showing anything else because they're tuned to avoid showing too many similar images. If you use one without that similarity-avoidance baked in, you see it immediately. It's actually super annoying if what you're trying to find is in fact variations on the same image, because they'll go way out of their way to avoid doing that, though some have tools for that ("show me more examples of images almost exactly like this one" sorts of tools)
The data behind the image search, before it goes through a similarity-classifier (or whatever) and gets those with too-close a score filtered out (or however exactly it works) probably looks a lot like "huh, every single adventurer with a hat just looks exactly like Harrison Ford?"
There's similar diversity-increasers at work on search results, it's why you can search "reddit [search terms]" on DDG and exactly the first 3 results are from reddit (without modifying the search to limit it to the site itself, just using it as a keyword) but then it switches to giving you other sites.
926 comments
[ 5.3 ms ] story [ 577 ms ] threadIf infringement is happening, it arguably doesn't happen when an infringing work product is generated (or regurgitated, or whatever you want to call it.) Much less when the model is trained. It's when the output is used commercially -- by a human -- that the liability should rightfully attach.
And it should attach to the human, not the tool.
If you insist on making it about the model, you will wreck something wonderful.
Given enough time (... a surprisingly short amount) and enough people creating art (say, about as many as we have had for the last couple hundred years) and indefinitely-long-lived recording, plus very-long copyright terms, the inevitable result is that it's functionally impossible to create anything within the space of "things people like" that's not violating copyright, for any but the strictest definitions of what constitutes copying.
The short story treats of music, but it's easy to see how visual arts and fiction-writing and the rest get at least extremely crowded in short order under those circumstances.
I can literally imagine hundreds of things that are true to this description but entirely distinct from "Predator."
> used commercially
Isn't that what these AI companies are doing? Charging you for access to this?
(Because proper stock agencies offer those kind of protections. If OpenAI doesn't, then don't use them as a replacement to a stock agency.)
I really don't think so. If I paid a human artist to make the prompt in the title, and I didn't explicitly say "Indiana Jones" I would think it should be fairly obvious to the human artist that I do _not_ want Indiana Jones. If they gave me back a picture of, clearly, Indiana Jones, I would ask them why they didn't create something original.
[1] https://sora.com/g/gen_01jr152s6seqct2qzgd1dz0qh2
[2] https://sora.com/g/gen_01jr14t6v3fzk8tx00aswwt4kx
Indiana Jones is already a successful permutation of that approach. He's Zorro, Rick Blaine, and Christopher Leiningen mashed together with their serial numbers filed off.
One death is a murder; 100k deaths is a war or a pandemic. One piece of chewing gum on the ground will get you a caning in Singapore; when everyone does it, that's NYC.
Up until now, one had to have some level of graphical or artistic skills to do this, but not anymore. Again, I agree that it attaches to the human...but we now have many more humans to attach it to.
This is not true, by the way. You will be fined for littering; or, if you are a repeat offender, be sentenced to cleaning public areas while wearing an offensively bright-coloured uniform (so that everyone can see that you are being punished). Source: https://www.nea.gov.sg/media/news/news/index/nea-increases-v...
But no, you won't be caned for littering. Caning is reserved for more serious offences like vandalism, or much worse crimes like rape and murder.
FWIW I’ve been to Singapore and had a great time, but I was careful to follow the many rules and signs. I especially liked the sign on the bus forbidding the opening of a durian fruit.
It isn’t an independent human. It is a service paid for by customers. The moment it provides the image to a paying user, the image has thus been used commercially.
In fact, the user may not even necessarily have to be paying in order to infringe copyright.
And besides, even amateur artists are ashamed to produce copies unless they are demonstrating mastery of technique or expressing adoration. And if it happens spontaneously, they are then frustrated and try to defend themselves by claiming to never have even experienced the original material. (As happens with simplistic, but popular musical riffs.) But AI explicitly is trained on every material it can get its hands on and so cannot make such a defense.
If you do it for free, is it different?
If I ask a friend to draw me as Indiana Jones? Or pay an artist? In either case I just want that picture to put in my rec room, not to sell.
Private copying and transference, even once for a friend, is copyright infringement.
I don’t necessarily agree with this, but it is true nonetheless.
Not without money or equivalent trade involved.
I can draw Mickey Mouse all day on my notebook, and hand it to you; no legal issues.
If I charge you a pack of bubble gum - Disney's lawyers will kick my door down and serve me notice.
This isn’t generally true. The copyright holder need only claim that the value of their copyrighted work or the profits received by its distribution has been reduced or lost. I’m not sure they even need to make such a claim, as courts have already determined that commerciality isn’t a requirement for infringement. It’s a matter of unauthorized use, distribution or reproduction, not trade.
They of course will likely never know and might not bother to litigate given the private and uncommercial nature, but they still have the right to do so.
They would weigh the financial and reputational cost of litigating against children sharing images and decide not to. Of course if one had 10000 friends and provided this service to them in a visible manner, then they’d probably come knocking.
You are asking the equivalent question of, if I put a pirated copy of windows on my PC that I only use privately at home am I violating copyright, or if I sell copies of music for people to only listen to in their own home.
But this is even more damning, this is a commercial service that is reproducing the copyrighted work.
Edit: Just to clarify to people who reflexively downvote. I'm making a statement of what is it not a value judgement. And yes there is fair use, but that's an exemption from the rule that it is a copyright violation.
Let’s put it another way: if you decide you want to recreate Indiana Jones shot for shot, and you hire actors and a director etc. which individuals are actually responsible for the copyright collation? Do caterers count too? Or is it the person who actually is producing the movie?
I am paying OpenAI. So they are producing these copyrighted works and giving them to me for their own commercial benefit. Normally that's illegal. But somehow not when you're just doing it en masse.
Not really? Why would a human artist create a faithful reproduction of Indiana Jones when asked to paint an archeologist? And besides, if they did, it would be considered clear IP infringement if the result were used commercially.
> If infringement is happening, it arguably doesn't happen when an infringing work product is generated (or regurgitated, or whatever you want to call it.) Much less when the model is trained. It's when the output is used commercially -- by a human -- that the liability should rightfully attach.
I agree. Release groups, torrent sites and seedbox operators should not be wrongly accused of pirating movies. Piracy only occurs in the act of actually watching a movie without paying, and should not be prosecuted without definitive proof of such (¬‿¬)
Isn’t that exactly what OP is saying?
Over the years we've spent a lot of time on this and similar sites questioning the sanity of a legal system that makes math illegal, and, well, that's all this is. Math.
To the extent the model reproduces images from Indiana Jones and the others, it is because these multibillion-dollar franchises are omnipresent cultural icons. The copyright holder has worked very hard to make that happen, and they have been more than adequately repaid for their contribution to our shared culture. It's insane to go after an AI model for simply being as aware of that imagery and as capable of reproducing it as a human artist would be.
If the model gives you infringing material as a prompt response, it's your responsibility not to use that material commercially, just as if you had tasked a human artist with the same vague requirement and received a plagiarized work product in return.
I completely disagree. It's not getting "better." It always just copied. That's all it /can/ do. How anyone expected novel outputs from this technology is beyond me.
It's highly noticeable if you do a minimal analysis, but all modern "AI" tools are just copyright thiefs. They're just there to whitewash away liability from blatantly stealing someone else's content.
That's true of all the best artists ever.
> They're just there to whitewash away liability from blatantly stealing someone else's content.
That's because that's not a thing. Ownership of "content" is a legal fiction invented to give states more control over creativity. Nobody who copies bytes which represent my music is a "thief". To be a thief, they'd need to, you know, come to my house and steal something.
When someone copies or remixes my music, I'm often not even aware that it has occurred. It's hard to imagine how that can be cast as genuine theft.
The reality of our current international laws, going back centuries, disagrees. And most artists disagree.
Perhaps I need a bit of education here, but have there been _international_ laws regarding intellectual property for centuries?!
AFAICT, the first major international IP treaties were in the 1880s (the Paris Convention on Intellectual Property Rights in 1883 and the Berne Convention covering copyrights in 1886; so only ~140 years.)
That's just a legal fiction invented so people can pretend to own physical objects even though we should all know that in this world you can never truly own anything.
Everything we do or protect is made up. You've just drawn the arbitrary line in the sand as to what can be "owned" in a different place than where other people might draw it.
Just to play Devil’s advocate for a moment, why should we require human artists to be held to the same standards as automated software? We can make whatever rules we want to.
A human might implicitly copy, but they are not infinitely scalable. If I draw a picture that in some way resembles Buzz Lightyear I am much less of a threat to Disney than an always-available computer program with a marginal cost of zero.
Isn't this the same question as, "why should we allow general purpose computing?" If the technology of our age is our birthright, don't we have the right to engage whatever mathematics we find inspiring with its aid?
> If I draw a picture that in some way resembles Buzz Lightyear I am much less of a threat to Disney than an always-available computer program with a marginal cost of zero.
...that sounds like a good argument in favor of the always-available computer program.
I don’t think so? It’s not illegal to call random people on your phone, but if you do it millions of times per day with a computer it can be illegal.
...and it's not obvious that transmitting bytes (within frequency and bandwidth protocols) is ever justly criminal; they are trivial to ignore.
Ownership is a legal fiction invented because it is perceived to encourage behavior that is seen as desirable; this is no more true of ownership of intellectual property or other intangible personal property than it is of tangible personal property or real estate.
If the State wishes to prevent expression that is deleterious to the ruling class, it will simply not strongly protect freedom of expression. "Legal protection" isn't an exogenous factor that the State responds to, it is a description of the actions of the State.
But the political cover of "but think of the poor artists!" is used as a cudgel for situations where "we prefer to censor Bob!" is unpalatable.
1. I hate the argument of "legal fiction" because the whole concept of law itself is a "fiction" invented that gives states more control. But I imagine you wouldn't want to live in a "lawless" society, would you?
2. Can you please explain how ownership of content gives states more control over creativity? There are so many way better methods of control a state can do (state-approved media, just banning books, propaganda) that this sounds like a stretch.
3. Alot of mainstream media is underdog rebels beating an Empire, and ownership of content definitely stops the spread of that idea.
2. Of course states can outright ban content, but the facade of protecting the poor artist gives enormous political cover. Saying, "You can't create that content because it is subversive and likely to convince people that the state is superfluous in the internet age" is just a much more honest (and politically impossible) approach - saying instead "you can't create that because it infringers on someone else's property" makes it sound like there's an attempt to serve justice.
3. Can you say more about this?
The problem is that it regurgitates what already exists and if you really want to abide by all the permissions then there is nothing left.
Here's Gemini's "make a picture of a young wizard which isn't a copy of any existing one.". I don't know enough about wizards to know how unique it is but maybe you can tell. https://lh3.googleusercontent.com/gg-dl/AA8i_VKUEpqw3AOOB_t4...
Though I am also generally opposed to the notion of intellectual property whatsoever on the basis that it doesn't seem to serve its intended purpose and what good could be salvaged from its various systems can already be well represented with other existing legal concepts, i.e deceptive behaviors being prosecuted as forms of fraud.
(Copied from a comment of mine written more than three years ago: <https://news.ycombinator.com/item?id=33582047>)
Arguments that make a case that NN training is copyright violation are much more compelling to me than this.
I'll prove it by induction: Imagine that I have a service where I "train" a model on a single image of Indiana Jones. Now you prompt it, and my model "generates" the same image. I sell you this service, and no money goes to the copyright holder of the original image. This is obviously infringment.
There's no reason why training on a billion images is any different, besides the fact that the lines are blurred by the model weights not being parseable
You gloss over this as if it's a given. I don't agree. I think you're doing a different thing when you're sampling billions of things equallly.
the model isn't the one infringing. It's the end user inputting the prompt.
The model itself is not a derivative work, in the same way that an artist and photoshop aren't a derivative work when they reproduce indiana jones's likeness.
A regulation that require restaurants to have a public bathroom is more akin to regulation that also require restaurants to check id when selling alcohol to young customers. Neither requirement has any relation with land rights, but is related to the right of operating a company that sell food to the public.
This is not the case in the US yet many places still have public restrooms, due to it benefiting the users themselves regardless of government.
Grok is supposed to be "uncensored", but there are very specific words you just can't use when asking it to generate images. It'll just flat out refuse or give an error message during generation.
But, again, if you go in a roundabout way and avoid the specific terms you can still get what you want. So why bother?
Is it about not wanting bad PR or avoiding litigation?
How they then go about implementing those guardrails is pretty telling about their understand and control over what they've build and their line of thinking. Clearly, at no point before releasing their LLMs onto the world did anyone stop and ask: Hey, how do we deal with these things generating unwanted content?
Resorting to blocking certain terms in the prompts is like searching for keywords in spam emails. "Hey Jim, I got another spam email from that Chinese tire place" - "No worry boss, I've configured the mail server to just delete any email containing the words China or tire".
Some journalist should go to a few of these AI companies and start asking questions about the long term effectiveness and viability of just blocking keywords in prompts.
I feel like the less advanced generations, maybe even because of their limitations in terms of size, were better at coming up with something that at least feels new.
In the end, other than for copyright-washing, why wouldn't I just use the original movie still/photo in the first place?
So the criminal party here would be OpenAI, since they are selling access to a service that generates copyright-infringing images.
Certainly there's an aspect of people using the chat interface like they use google: describe xyz to try to surface the name of a movie. Just in this case, we're doing the (less common?) query of: find me the picture I can vaguely describe; but it's a query to a image /generating/ service, not an image search service.
So I asked it to make 4 random and generic superheroes. It created Batman, Supergirl, Green Lantern, and Wonder Woman. Then at about 90% finished it deleted the image and said I was violating copyright.
https://imgur.com/a/eG6kmqu
I doubt the model you interact with actually knows why the babysitter model rejects images, but it claims to know why and leads to some funny responses. Here is it's response to me asking for a superhero with a dark bodysuit, a purple cape, a mouse logo on their chest, and a spooky mouse mask on their face.
> I couldn't generate the image you requested because the prompt involved content that may violate policy regarding realistic human-animal hybrid masks in a serious context.
(hard to formulate why I was too lazy to test myself :) )
Ironically that's probably because the errors and flaws in those generations at least made them different from what they were attempting to rip off.
I have a strong suspicion that many human artists would behave in a way the chat bot did (unless they start asking clarifying questions. Which chatbots should learn to do as well)
Overall the model is tra
I wonder if it's a fine tuning issue where people have overly provided archetypes of the thing that they were training towards. That would be the fastest way for the model to learn the idea but it may also mean the model has implicitly learned to provide not just an instance of a thing but a known archetype of a thing. I'm guessing in most RLHF tests archetypes (regardless of IP status) score quite highly.
ClosedAI doesn't seem to be OK with it, because they are explicitly censoring characters of more popular IPs. Presumably as a fig leaf against accusations of theft.
The question is "should we define it as such?"
But because a computer, and not a human does it, they get to launder their responsibility.
For humans it doesn't make sense because we have generation and filtering in a single package.
Overfitting is if you didn't exactly describe Indiana Jones and then it still gave Indiana Jones.
It didn't though, it just spat out what is basically a 1:1 copy of some Indiana Jones promo shoot. No where did the prompt ask for it to look like Harrison Ford.
If we were playing Charades, just about anyone would have guessed you were describing Indiana Jones.
If you gave a street artist the same prompt, you'd probably get something similar unless you specified something like "... but something different than Indiana Jones".
But if you look at it from the perspective that there is only one example to learn, from it is maybe not over it.
That's not overfitting. That's either just correct or underfitting (if we say it's never returning anything but 2)!
Overfitting is where the model matches the training data too closely and has inferred a complex relationship using too many variables where there is really just noise.
How can you express, in term of AI training, ignoring the existence of something that's widely present in your training data set? if you ask the same question to a 18yo girl in rural Thailand, would she draw Harrison Ford as Indiana Jones? Maybe not. Or maybe she would.
But IMO an AI model must be able to provide a more generic (unbiased?) answer when the prompt wasn't specific enough.
Maybe it would have some point if you are targetting users in a substantially different social context. In the case, you would design the model to be familiar with their tropes instead. So when they describe a character iconic in their culture, by a few distinguishing characteristics, it would produce that character for them. That's no different at all.
Recent benchmark on unseen 2025 Math Olympiad shows none of the models can problem solve . They all accidentally or on purpose had prior solutions in the training set.
https://x.com/mbalunovic/status/1907436704790651166
[0] https://imgur.com/a/wqrBGRF Image captions are the impled IP, I copied the prompts from the blog post.
To me, this article is further proof that LLMs are a form of lossy storage. People attribute special quality to the loss (the image isn't wrong, it's just got different "features" that got inserted) but at this point there's not a lot distinguishing a seed+prompt file+model from a lossy archive of media, be it text or images, and in the future likely video as well.
The craziest thing is that AI seems to have gathered some kind of special status that earlier forms of digital reproduction didn't have (even though those 64kbps MP3s from napster were far from perfect reproductions), probably because now it's done by large corporations rather than individuals.
If we're accepting AI-washing of copyright, we might as well accept pirated movies, as those are re-encoded from original high-resolution originals as well.
To a viewer, a human-made work and an AI-generated one both amount to a series of stimuli that someone else made and you have no control over; and when people pay to see a movie, generally they don't do it with the intent to finance the movie company to make more movies -- they do it because they're offered the option to spend a couple hours watching something enjoyable. Who cares where it comes from -- if it reached us, it must be good, right?
The "special status" you speak of is due to AI's constrained ability to recombine familiar elements in novel ways. 64k MP3 artifacts aren't interesting to listen to; while a high-novelty experience such as learning a new culture or a new discipline isn't accessible (and also comes with expectations that passive consumption doesn't have.)
Either way, I wish the world gave people more interesting things to do with their brains than make a money, watch a movies, or some mix of the two with more steps. (But there isn't much of that left -- hence the concept of a "personal life" as reduced to breaking one's own and others' cognitive functioning then spending lifetimes routing around the damage. Positively fascinating /s)
A new MCU movie is released, its 60 second trailer posted on Youtube, but I don't feel like watching the movie because I got bored after Endgame.
Youtube has very strict anti-scraping techniques now, so I use deep-scrapper to generate the whole trailer from the thumbnail and title.
I use deep-pirate to generate the whole 3 hour movie from the trailer.
I use deep-watcher to summarize the whole movie in a 60 second video.
I watch the video. It doesn't make any sense. I check the Youtube trailer. It's the same video.
Hail their father, drip with sweat:
"Fukuyama had deceived us!
History has no end."
Unlike what they name in physics, laws in juristic field are not a given by the cosmos. That's all human fantasy, and generally not enacted by the most altruistic and benevolent wills.
Call me back when we have no more humble humans dying from cold, hunger and war, maybe I'll have some extra compassion to spend on soulless megacorps which pretend they can own the part of our brain into which they inject their propaganda and control what we are permitted to do starting from that.
One thing I would say, it's interesting to consider what would make this not so obviously bad.
Like, we could ask AI to assess the physical attributes of the characters it generated. Then ask it to permute some of those attributes. Generate some random tweaks: ok but brawy, short, and a different descent. Do similarly on some clothing colors. Change the game. Hit the "random character" button on the physical attributes a couple times.
There was an equally shatteringly-awful less-IP-theft (and as someone who thinks IP is itself incredibly ripping off humanity & should be vastly scoped down, it's important to me to not rest my arguments on IP violations).... An equally shattering recent incident for me. Having trouble finding it, don't remember the right keywords, but an article about how AI has a "default guy" type that it uses everywhere, a super generic personage, that it would use repeatedly. It was so distasteful.
The nature of 'AI as compression', as giving you the most median answer is horrific. Maybe maybe maybe we can escape some of this trap by iterating to different permutations, by injecting deliberate exploration of the state spaces. But I still fear AI, worry horribly when anyone relies on it for decision making, as it is anti-intelligent, uncreative in extreme, requiring human ingenuity to budge off its rock of oppressive hypernormality that it regurgitates.
I mean... If I go to Google right now and do an image search for "archeologist adventurer who wears a hat and uses a bullwhip," the first picture is a not-even-changed image of Indiana Jones. Which I will then copy and paste into whatever project I'm working on without clicking through to the source page (usually because the source page is an ad-ridden mess).
Perhaps the Internet itself is the hideous theft machine, and AI is just the most efficient permutation of user interface onto it.
(Incidentally, if you do that search, you will also, hilariously, turn up images of an older gentleman dressed in a brown coat and hat who is clearly meant to be "The Indiana Jones you got on Wish" from a photo-licensing site. The entire exercise of trying to extract wealth via exclusive access to memetic constructs is a fraught one).
The hypocrisy is much of the problem. If we're going to have IP laws that severely punish people and smaller companies for reselling the creative works of others without any compensation or permission then those rules should apply to powerful well-connected companies as well.
If you make money off it, it's no longer fair use; it's infringement. Even if you don't make money off it, it's not automatically fair use.
My own favorite crazy story about copyright violations:
Metallica sued Green Jello for parodying Enter Sandman (including a lyric where it says "It's not Metallica"):
https://en.wikipedia.org/wiki/Electric_Harley_House_(of_Love...
They lost that case. The kicker? Metallica were guest vocalists on that album.
That's my take as well.
Gen AI is turning small potatos, "artisanal" infringement into a potentially large scale automated process.
What if the drawing is of Indiana Jones but he's carrying a bow and arrow instead of a whip? Is it infringement?
What if it's a really bad drawing of Indiana Jones, so bad that you can't really tell that it's the character? Is that infringement?
What if the drawing is of Indiana Jones, but in the style of abstract expressionism, so doesn't even contain a human shape? Is it infringement?
What if it's a good drawing that looks very much like Indiana Jones, but it's not! The person's name is actually Iowa Jim. Is that infringement?
What if it's just an image of an archeologist adventurer who wears a hat and uses a bullwhip, but otherwise doesn't look anything like Indiana Jones? Is it infringement?
https://en.wikipedia.org/wiki/Copyright_protection_for_ficti...
I don't think this is about reproduction as much as how you got enough data for that reproduction. The riaa sent people to jail and ruined their lives for pirating. Now these companies are doing it and being valued for hundreds of billions of dollars.
A human friend can get tired and there's so many request he/she can fulfill and at a max rate. Even a team of human artists have a relatively low limit.
But Gen AI has very high limits and speeds, and it never gets tired. It seems unfair to me.
You also can’t sell a machine that outputs such material. And that’s how the story with GenAI becomes problematic. If GenAI can create the next Indiana Jones or Star Wars sequel for you (possibly a better one than Disney makes, it has become a low bar of sorts), I think the issue becomes obvious.
Nobody can prevent you from drawing a photo realistic picture of Indy, or taking a photo of him from the internet and hanging it on your fridge. Or asking a friend to do it for you. And let's be honest -- because nobody is looking -- said friend could even charge you a modest sum to draw a realistic picture of Indy for you to hang on your fridge; yes, it's "illegal" but nobody is looking for this kind of small potatos infringement.
I think the problem is when people start making a business out of this. A game developer could think "hey, I can make a game with artwork that looks just like Ghibli!", where before he wouldn't have anyone with the skills or patience to do this (see: the 4-second scene that took a year to make), now he can just ask the Gen AI to make it for them.
Is it "copyright infringement"? I dunno. Hard to tell, to be honest. But from an ethical point of view, it seems odd. And before you actually required someone to take the time and effort to copy the source material, now it's an automated and scalable process that does this, and can do this and much more, faster and without getting tired. "Theft at scale", maybe not so small potatos anymore.
--
edit: nice, downvotes. And in the other thread people were arguing HN is such a nice place for dissenting opinions.
Breath of the Wild, and Tears of the Kingdom should be included there.
Regardless, those games required the hard work and countless hours of animators. Gen AI doesn't.
Indy, with its logo, whiplash, and hat, is a trademark from Disney. I don't know the specific stuff; but if you sell a t-shirt with Indiana Jones, or you put the logo there... you might be sued due to trademark violation.
If you make copies of anything developed, sold, or licensed by Disney (movies, comics, books, etc) you'll have a copyright violation.
The issue we have with AI and LLM is that: - The models compress information and can make a lot of copies of it very cheaply. - Artist wages are quite low. Higher that what you'd pay OpenAI, but not enough to make a living even unless you're hired by a big company (like Marvel or DC) and they give you regular work ($100-120 for a cover, $50-80/page interior work. One page needs about one day to draw.) - AI used a lot of images from the internet to train models. Most of them were pirated. - And, of course, it is replacing low-paying jobs for artist.
Also, do not forget it might make verbatim copies of copyrighted art if the model just memorized the picture / text.
Sure, assuming the artist has the proper license and franchise rights to make and distribute copies. You can go buy a picture of Indy today that may not be printed by Walt Disney Studios but by some other outfit or artists.
Or, you mean if the artist doesn't have a license to produce and distribute Indiana Jones images? Well they'll be in trouble legally. They are making "copies" of things they don't own and profiting from it.
Another question is whether that's practically enforceable.
> Where did I (or the artist) violate any copyright (or other) laws?
When they took payment and profited from making unauthorized copies.
> It is the artist that is replaced by the AI, not the copyrighted IP.
Exactly, that's why LLMs and the companies which create them are called "theft machines" -- they are reproducing copyrighted material. Especially the ones charging for "tokens". You pay them, they make money and produce unauthorized copies. Show that picture of Indy to a jury and I think it's a good chance of convincing them.
I am not saying this is good or bad, I just see this having a legal "bite" so to speak, at least in my pedestrian view of copyright law.
Commissioned work is owned by the commissioner unless otherwise agreed upon by contract.
So long as the work is not distributed, exhibited, performed, etc, as in the example of keeping the artwork on their refrigerator in their home, then no infringement has taken place.
I think the LLM example is closer to the LLM and its creator being like a vendor selling pictures of Indiana Jones on the street corner than hiring someone and performing work for hire. Yes, if it was a human artist commissioned to create an art piece, then yeah, the commissioner owns it.
Just because I own my car doesn’t mean I can break the speed limit, these are orthogonal concepts legally.
I'd like to push back on this: Is that legally true, or is it infringement which just happens to be so minor and under-the-radar that nobody gets in trouble?
Suppose there's a printer in my room churning out hundreds of pages of words matching that of someone's copyrighted new book, without permission.
That sure seems like infringement is happening, regardless of whether my next step is to: (A) sell it, (B) sell many of it, (C) give it away, (D) place it into my personal library of other home-printed books, or (E) hand it to someone else who paid me in advance to produce it for them under contract.
If (A) is infringement, why wouldn't (E) also be?
Hangs on my wall now- I know I can never sell until the copyright on the original runs out (which it most likely won't in my lifetime) it but it is a very well done painting and a family legacy piece I am glad exists.
Ok, my sister can draw, and she gifts me an image of my favorite Marvel hero she painted to hang on my wall. Should that be illegal?
The likeness of the character is owned by Marvel. Does it mean there aren’t vendors selling unlicensed versions? No. I am sure there are. But just because not everyone is being sued doesn’t mean it’s suddenly legal.
If they traced a photo they might be violating the copyright of the photographer.
But if they are drawing an archaeologist adventurer with a whip and a hat based on their consumption and memory of Indiana Jones imagery there is very little anyone could do.
If that image was then printed on an industrial scale or printed onto t-shirt there is a (albeit somewhat theoretical) chance that in some jurisdictions sale of those products may be able to be restricted based on rights to the likeness. But that would be a stretch.
If they show that image to a jury they’ll have no issues convincing them the LLM is infringing.
Moreover if the LLM creators are charging for it, per token or whatever, they are profiting from it.
Yes are there jurisdictions were this won’t work and but I think in US Disney lawyers could make viable argument.
With the LLM it would be nothing to do with likeness, it would be to do with the copyright in the image, the film, video or photograph. The image captures the likeness but the infringement would not be around the likeness.
Sorry for the misunderstanding. I was thinking of LLMs mostly.
> With the LLM it would be nothing to do with likeness, it would be to do with the copyright in the image
I guess I don't see why it wouldn't be about a character's likeness. It's not just a generic stock character, as an idea but it has enough distinctive characteristics, has a particular style hat, uses a whip. Showing that image to any jury and they'll say this is Indy. Yeah there are trademarks as well and both can apply to characters.
You can think of likeness as something that could be captured by hieroglyphics, or emoji - or a game of charades.
Think of Betty Boo or Wiley Coyote. You might not be able to close your eyes and picture them exactly but you can close your eyes and imagine the essence.
So in some jurisdictions if you register a broad representation - a likeness - you can protect that in law. In almost all jurisdictions if you photograph or draw something that literal snapshot is protected. But in many jurisdictions if I took a still from a scene of Indiana Jones (which WOULD be protected by copyright) and I described it to you and you were a great artist and drew the whip, the hat, the grizzled face, the safari vest, etc - there would be nothing to protect that likeness.
Trademarks are another thing entirely, and are things like logos, and the protection is more about commercial exploitation and preventing misleading us.
I could draw a cartoon logo of a man with a hat and a whip for my archaeological supplies shop. You could not take it and use it on your Chinese manufactured whips drop shipped via Amazon. But the owner of the Indiana Jones trademark- likeness rights, film copyright etc would be unlikely to be able to stop me using my logo until I start saying it is Indiana Jones, at which point they maybe invoke likeness rights.
There are at least three different rights classes here! It can be very confusing.
I didn't say this was a trademark case just that there are multiple different aspects possible to the likeness issue.
Wouldn’t the more appropriate solution in the case of theft be to remunerate the victims and prevent recidivism?
Instead of making it “not so obviously bad” why not just… make it good? Require AI services to either prove that 100% of their training corpus is either copyright free or properly licensed, or require them to compensate copyright holders for any infringing outputs.
Not sure I understand this part. Because creators would be getting paid for their works being used for someone else’s commercial gain?
I hate how it is common to advance a position to just state a conclusion as if it were a fact. You keep repeating the same thing over and over until it seems like a concensus has been reached instead of an actual argument reasoned from first principle.
This is no theft here. Any copyright would be flimsier than software patents. I love Studio Ghibli (including $500/seat festival tickets) but it's the heart and the detail that make them what they are. You cannot clone that. Just some surface similarity. If that's all you like about the movies... you really missed the point.
Imagine if in early cinema someone had tried to claim mustachioed villian, ditsy blonde, or dumb jock? These are just tropes and styles. Quality work goes much much much deeper, and that cannot be synthesised. I can AI generate a million engagement rings, but I cannot pick the perfect one that fits you and your partners love story.
PS- the best work they did was "When Marnie was There". Just fitted together perfectly.
You absolutely can and these theft machines are proving that, literally cloning those details with very high precision and fidelity.
You can easily steal the style of a political cartoon or especially XKCD but you cannot steal or generate genuine fresh insight or poignant relevant metaphor for the current moment.
If engagement rings were as ubiquitous and easy to generate as Ghibli images have become, they would lose their value very quickly -- not even just in the monetary sense, but the sentimental value across the market would crash for this particular trinket. It wouldn't be about picking the right one anymore, it would be finding some other thing that better conveys status or love through scarcity.
If you have a 3d printer you'd know this feeling where abundance diminishes the value of something directly. Any pure plastic items you have are reduced to junk very quickly once you know you can basically have anything on a whim (exceptions for things with utility, however these are still printable). If I could print 30 rings a day, my partner wouldn't want any of them as a show of my undying love. Something more special and rare and thoughtful would have to take its place.
This isn't meant to come across as shallow in any way, its just classic supply and demand relating to non monetary value.
And now I think this serves the opposite argument better. Downloading some random ring from the internet would not show your undying love. Designing a custom ring just for your partner, even if it is made from plastic, and even if you use AI as a tool in the process, is where the value is generated.
As an aside, my partner detests the things I 3d print unless they have a very specific purpose, even when they are random semi artistic pieces I'm tinkering with (and I typically agree, they are junk). She loves the first thing i ever printed her though, a triceratops model, despite being randomly downloaded.
Anything made with intent from one individual to another will have some level of sentimental value, but I don't feel like making a ghibli image with AI specifically tailored to a friends tastes would have quite as much value as leveraging your own talent to do it yourself.
On the flip side, I do believe that "doing it yourself" has less value than it used to. It's a very sad reality and in my opinion a strong argument against blind "progress". We gain the ability to mass produce art but lose the ability to perceive it as art?
It doesn't matter if the ring was hand crafted or not. It's whether it has hand selected. If you find the perfect ring, even if it was generated by an AI, it's your selection that matters. It's the correspondence that matters. The way it reflects elements of your relationship. It's you recognising those elements in the ring. Your partner recoginising them in the ring. And your partner recognising you recognising them. That is what makes itself.
Not to dox myself, but I am not Grace Abrams. I met my partner long before her song "Risk" was written, but when I heard it I immediately played it for my partner and said "This describes the feelings I had when I met you". I played it for her, she cried. I didn't have to write the song or own or pay a cent for it. It's the curation that made an emotional connection and had value. The song itself has no value, and she might have even heard it and never made the connection, it was me embuing that had value.
To go back to Miyazaki, it's the connections between elements in his films. The attention to detail and tone between relationships that make his films amazing. It's all about the handyman's invoice [0]. By the time there are enough examples for AI to learn something, it ceases to be a novel insight and have value. It's the curation and application that have value and are human and cannot be stolen.
[0] https://www.snopes.com/fact-check/know-where-man/
I have interacted with the parent account before and have actually looked at the the amount of times they used words like awful, horrific, etc, and I definitely agree, as one should not have such a strong attachment to such words that they feel the need to continue to use (or rather, abuse) them endlessly.
[0] https://hn.algolia.com/?dateRange=all&page=0&prefix=true&que...
[1] https://scholar.google.com/scholar_case?case=137674209419772...
Are you telling me that our culture should be deprived of the idea of Indiana Jones and the feelings that character inspires in all of us forever just because a corporation owns the asset?
Indiana Jones is 44 years old. When are we allowed to remix, recreate and expand on this like humanity has done since humans first started sitting down next to a fire and telling stories?
edit: this reminds of this iconic scene from Dr. Strangelove, https://www.youtube.com/watch?v=RZ9B7owHxMQ
I guess we all have to answer to the Walt Disney company.Some great video games to feature adventurer archaeologists:
* NetHack (One of the best roles in the game)
* Tomb Raider series (Lara Croft is a bona fide archaeologist)
* Uncharted series (Nathan Drake is more of a treasure hunter but he becomes an archaeologist when he retires from adventuring)
* Professor Layton series
* La-Mulana series (very obviously inspired by Indiana Jones, but not derivative)
* Spelunky (inspired by La-Mulana)
[1] https://tvtropes.org/pmwiki/pmwiki.php/Main/AdventurerArchae...
IMO any change to copyright law should not be applied retroactively. Make copyright law to be what is best for society and creators as a whole, not for lobbyists representing already copyrighted material.
Careful, if we were to shorten copyright, not doing so retroactively would give an economic advantage to franchises already published over those that would get published later. As if the current big studios needed any further advantages over newcomers.
This is a kind of strange comment for me to read. Because imby tone it sounds like a rebuttal? But by content, it agrees with a core thing I said about myself:
> and as someone who thinks IP is itself incredibly ripping off humanity & should be vastly scoped down, it's important to me to not rest my arguments on IP violations
What's just such a nightmare to me is that the tech is so normative. So horribly normative. This article shows that AI again and again reproduced only the known, only the already imagined. Its not that it's IP theft that rubs me so so wrong, it's that it's entirely bankrupt & uncreative, so very stuck. All this power! And yet!
You speak at what disgusts me yourself!
> When are we allowed to remix, recreate and expand on this like humanity has done
The machine could be imagining all kinds of Indianas. Of all different remixed recreated expanded forms. But this pictures are 100% anything but that. They're Indiana frozen in Carbonite. They are the driest saddest prison of the past. And call into question the validity of AI entirely, show something greviously missing.
If AI is just compression, then decompressing a generic pop-culture-seeking prompt will yield a generic uninspired image.
You are completely ignoring the fact that you can provide so much more information to the LLMs to get what you want. If you truly want novel images, ChatGPT can absolutely provide them, but you have to provide a better starting point than "An image of an archeologist adventurer who wears a hat and uses a bullwhip".
If you just provide a teensy bit more information, the results dramatically change. Try out "An image of an Indian female archeologist adventurer who wears a hat and uses a bullwhip". Or give it an input image to work with.
From just adding a couple words, ChatGPT produces an entirely new character. It's so easy to get it to produce novel images. It is so easy in fact, that it makes a lot of posts like this one feel like strawmen, intentionally providing so little information to the LLMs that the generic character is the only obvious output that you would expect.
Now, would it be better if it didn't default to these common movie tropes? Sure. But the fact that it can follow these tropes doesn't mean that it cannot also be used to produce entirely new images filled with your imagination as well. You just have to actually ask it for that.
So yes, the models are not creative on their own. But equally, these models are definitely capable of helping people to express their own creativity, and so calling them "uncreative", and especially "bankrupt", rings hollow. It speaks like you are expecting the models to be artists, when in fact they are just tools to be used however people see fit.
And so, the "default guy" or "default style" that ChatGPT outputs will become recognisable and boring. But anyone who wants to inject their own style into their prompts, either using text or input images, can do so. And in doing so, they skip over all of your concerns.
> Like, we could ask AI to assess the physical attributes of the characters it generated. Then ask it to permute some of those attributes. Generate some random tweaks: ok but brawy, short, and a different descent.
what a terror you've been.
If you describe an Indiana Jones character, but no sex, 50/50 via internal call to rand() that it outputs a woman.
Ah, I thought I knew this account from somewhere. It seems surprisingly easy to figure out what account is commenting just based on the words used, as I've commented that only a few active people on this site seem to use such strong words as shown here.
LLMs are better at generating the boilerplate of todays programming languages than they will be with tomorrows programming languages.
This is because not only will tomorrows programming languages be newer and lacking in corpus to train the models in but, by the time a corpus is built, that corpus will consist largely of LLM hallucinations that got checked into github!?
The internet that that has been trawled to train the LLMs is already largely SEO spam etc, but the internet of the future will be much more so. The loop will feed into itself and become ever worse quality.
It isn't your responsibility if a bank you've never set foot in gave a bunch of money to someone who isn't you and said they were you... And it never should have been. The moment companies started trying to put that into credit ratings, they should have been barred for discriminatory practices against the unlucky.
Yes. In this case, it is the artist's sole right to reproduce said images, based on their creative output.
>> decreases scarcity, it doesn't increase it
What does scarcity have to do with stealing? You can steal bread and reduce food scarcity, but that is still theft.
Copyright is why Disney can ruin someone for doing something with Goofy they don't like. Yes, it protects smaller, less profitable artists too, but make no mistake: it's a tool of mass control and cultural capture.
Perhaps it's time to seriously ask whether copyright is actually doing its job of "promot[ing] the Progress of Science and useful Arts." "...but generally speaking, other nations have thought that these monopolies produce more embarrasment than advantage to society" [Jefferson].
Which efforts do we decide deserve compensation? We collectively create so many useful things. People come up with new words to describe a concept and when it's apt, it's widely adopted. That brilliant idea is the culmination of all the other work they've been putting in. But others get to use the word and benefit from it. Indigenous farmers put hundreds of years of effort into domesticating crops like tomatoes and corn and peppers. But those are just taken and spread. We have millions of teachers going above and beyond expectations. We have mothers and caregivers. We have comedians crafting jokes for comedy and they have no ability to protect those jokes. We have brilliant chefs creating recipes after years of work, experience, and effort and they too don't have protections.
> To claim they are indicates some kind of deep disrespect for your fellow human beings.
Nah. My comment comes from a deep respect for fellow humans. And a respect for all the efforts collectively made by a society for its common good.
I agree with this, but why is your proposed solution to remove protections from everyone instead of giving everyone else protections? People who put in work should be rewarded.
I don't think it's possible to create an "alien which has acid for blood and a small sharp mouth within a bigger mouth" without anybody seeing a connection to Alien, even if it doesn't look anything like the original.
The photo was of poor quality, but one could certainly see all the features - so I figured, why not let ChatGPT try to play around with it? I got three different versions where it simply tried to upscale it, "enhance" it. But not dice.
So I just wrote the prompt "render this photo as a hyper realistic photo" - and it really did change us - the people in the photo - it also took the liberty to remove some things, alter some other background stuff.
It made me think - I wonder what all those types of photos will be like 20 years from now, after they've surely been fed through some AI models. Imagine being some historian 100 years from now, trying to wade through all the altered media.
Never use the words "hyper realistic" when you want a photo. It makes no sense and misleads the generator. No one would describe a simple photograph as "hyper realistic," not a single real photo in the dataset will be tagged as "(hyper) realistic."
Hyperrealism is an art style and only ever used in the context of explicitely non-photographic artworks.
I've had much better luck with models specifically trained for denoising. For denoising, the SCUNet model run via chaiNNer works well for me most of the time. (Occasionally SCUNet likes to leave noise alone in areas that are full of background blur, which I assume has to do with the way the image gets processed as tiles. It would make sense for the model to get confused with a tile that only has background blur, like maybe it assumes that the input image should contain nonzero high-frequency data.)
For your use case, you might want to use something like Real-ESRGAN or another superresolution / image restoration model, but I haven't played much in that space so I can't make concrete recommendations.
Two close to one of the licensed properties you care to censor the generation of? Push that vector around. Honestly detecting whether a given sentence is a thinly veiled reference to indiana jones seems to be exactly the kind of thing AI vector search is going to be good at.
If IP holders submit embeddings for their IP, how can image generators "warp" the latent space around a set of embeddings so that future inferences slide around and avoid them--not perfectly, or literally, but as a function of distance, say, following a power curve?
Maybe by "Finding non-linear RBF paths in GAN latent space"[0] to create smooth detours around protected regions.
0. https://openaccess.thecvf.com/content/ICCV2021/papers/Tzelep...
To be honest, I wouldn't mind if AI that just reproduces existing images like that would just be banned. Keep working on it until you've got something that can actually produce something new.
https://nwn.blogs.com/nwn/2024/09/ted-chiang-ai-new-yorker-c...
Nah, that's just restating the infamous 'how to draw an owl' advice:
https://casnocha.com/2010/11/how-to-draw-an-owl.html#comment...
The thing is, that "long-ass prompt" is step 1, and LLM then draws "the rest of the fucking owl" for you. That's quite a big difference to doing it all yourself.
The real issue here is that there's a whole host of implied context in human languages. On the one hand, we expect the machine to not spit out copyrighted or trademarked material, but on the other hand, there's a whole lot of cultural context and implied context that gets baked into these things during training.
If I ask an artist to draw me a picture of Indiana Jones and they do it would that be copyright infringement? Even if it's just for my personal use?
If e.g. Patreon hosts an artist who will draw a picture of Indiana Jones for me on commission, then my money is going to both Patreon and the artist. Should Patreon also police their artists to prevent reproducing any copyrighted characters?
I get that copyright is a bit of a minefield, and there's some clear cases that should not be allowed, e.g. taking photos of a painting and selling them
That said, I still get the impression that the laws are way too broad and there would be little harm if we reduced their scope. I think we should be allowed to post pictures of Pokemon toys to Wikipedia for example.
I'm willing to listen to other points of view if people want to share though
Not to mention that wikimedia commons, which tries to be a globally reusable repository ignores fair use (which is context dependent), which covers a lot of the cases where copyright law is just being reduculous.
It may or may not be fair use, which is a complicated question (ianal).
Rules around copyright (esp. Fair use) can be very context dependent.
https://en.wikipedia.org/wiki/Personality_rights#United_Stat...
edit - also, I wasn't making a binary claim, the person I was responding to was: "no law". There are more than zero laws relevant to this situation. I agree with you that how relevant is context dependent.
It's not clear what the asker wants, and the obvious answer is probably the culturally relevant one. Hell, I'd give you the same answers as the AI did here if I had the ability to spit out perfect replicas.
Also, there are IP limits of various sorts (e.g. copyright, trademark) for various purposes (some arguably good, some arguably bad), and some freedoms (e.g., fair use). There's no issue if this follows the rules... but I don't see where that's implemented here.
It looks like they may be selling IP they don't own the right to.
https://en.wikipedia.org/wiki/Overfitting
https://theaiunderwriter.substack.com/p/an-image-of-an-arche...
and I'm all in on this conclusion:
> It’s stealing, but also, admittedly, really cool.
Mmm, kinda, but those image results only don't show 1,000 of the exact same image before showing anything else because they're tuned to avoid showing too many similar images. If you use one without that similarity-avoidance baked in, you see it immediately. It's actually super annoying if what you're trying to find is in fact variations on the same image, because they'll go way out of their way to avoid doing that, though some have tools for that ("show me more examples of images almost exactly like this one" sorts of tools)
The data behind the image search, before it goes through a similarity-classifier (or whatever) and gets those with too-close a score filtered out (or however exactly it works) probably looks a lot like "huh, every single adventurer with a hat just looks exactly like Harrison Ford?"
There's similar diversity-increasers at work on search results, it's why you can search "reddit [search terms]" on DDG and exactly the first 3 results are from reddit (without modifying the search to limit it to the site itself, just using it as a keyword) but then it switches to giving you other sites.