AI Copyright Landmark: German Court Rules Against OpenAI in GEMA Case

Craig: big news out of europe that could change how chattp works for everyone. if you think ai companies can just scrape whatever they want from the internet to build their bots, a german court just slammed down a massive gavel that says absolutely not, and the ripple effects are coming for us all. immunicord has ruled that openai infringed on copyright laws by training chachi to be on song lyrics lyrics without a license. this is headline news because it's legally redefines training data. it no longer just learning. it's now considered reproducing and memorizing. this destroys the fair use defense many ai companies rely on in the us and could force them to pay billions in licensing fees, potentially changing the economics of ai forever. what do you think about this, john?

Jon: i'm kind of disappointed in openai's defense. so it rested on a few key arguments, and they are all rejected by the court, by the way. so first, openai suggested that the user who enters the prompt should be legally liable for any infringing output. okay. so an argument that the court dismissed, they also argued that any reproduction was accidental or a technical error, which, again, is laughable. and the court countered this by stating that both the initial copying of data for training and its later reproduction require a license. i agree. and then crucially, the court set a very low threshold for what counts as infringement, ruling that even a 15 word passage could violate copyright law. so this undercuts the common defense from ai companies that their models are creating something entirely new and not just copies of the source material. and so what are your thoughts on those arguments? like, that's sorry.

Craig: i sorry. scoffit. yeah. let me just expand on this compared to, say, what some of the other courts have decided, and, you know, we can talk about this. i think this is interesting, but i don't think it's going anywhere, and i'll explain why in a bit. alright. so, the court firmly rejected the black box defense. memorizing and reproducing copyright lyrics during training and output can, can constitute infringement regardless of the technical mechanism involved. so, that's what the german court decided. in the uk, in the getty images versus stabilityai case, uk court ruled the opposite saying that ai models don't contain copies of images, but just mathematical weights. the german court disagrees creating a massive legal conflict. versus the us on books and news and the author's guild versus openai and and, new york times versus openai, judges are currently looking at whether ai summaries are close enough to the original to count as theft. this german ruling gives a huge ammunition, to the new york, times and authors. and there's more on this. let me just check. there's a recent ruling with anthropic, okay, of perplexity. so, the german ruling focused on training. this thing was ingesting the data in the first place. now the us contrast between dow jones, new york post versus perplexity, a lawsuit, it's not about training. it's about retrieval augmented generation. so, again, that's the news story. right? the news publisher argued that perplexity isn't just learning from the news. it's replacing the news by summary articles. so you don't, need to click. so they don't have fair use everywhere. we have different rules, and the courts recently decided $1,500,000,000, case against anthropic for books, or was it openai? i thought it was anthropic. i thought it was anthropic. but i'll look that up as we go.

Jon: something that

Craig: for and i saw some posts around our authors saying that you can get up to $1,500 per book back on this, class action lawsuit. so there's a lot of different perspectives on there, and i'm not sure where you fall on all of that.

Jon: so for me, it's it comes down to they've used they've used this data to train their models. that's that and without paying for it. so that's where i come in. that's where i stand is they should have paid even though it's available on the internet, they still should have negotiated some sort of agreement before using, anything that, that has been created by artists or anything that's been created by someone, to train their models. that's my opinion, though.

Craig: yeah. and it is an opinion. so i'll put it in a different light, and it might not be the same. but imagine a student, you know, in this case, the eye, who claims they studied for a test, but when asked a question, they just recite the textbook word for word. so if you have a photographic memory and just go like this through the textbook and you have it all and you just recite whatever, the answer from the textbook, the court says that's not learning. that's just making an illegal photocopy of the book inside your brain. now i don't know if it gonna apply to humans, or just to ai. like

Jon: well, that human isn't isn't, making any money off of that test that he studied for. whereas openai is making money off of other people's content that they've trained their models on.

Craig: and potentially taking money away from the people who wrote the original content. that's kinda you know, you're hurt. if they can prove that there's actual financial harm, that's where things come into play.

Jon: right? difficult. yeah.

Craig: yeah. so that's it. because, you know, things like, when google turned it off in the eu or whatever the country it was, they found it was point 1% of traffic dropped. so it was insignificant, and so that was kind of their argument when they were doing that. so we might see, moving forward, some things like geofenced ai. you might start seeing this content is unavailable in your region, errors on chat ttp similar to netflix netflix or what have you, where, you know, go to point an article or whatnot. they may not be able to summarize it for you or help you. if these companies have to pay, we're probably gonna see price hikes. you know? if they have to pay for every song book they train, the price of chatduty plus and other tools is likely gonna go up to cover the bill. to avoid lawsuits, ai companies might purge their databases, meaning the ai knows less about pop culture lyrics and modern literature, so we might get dumber bots.

Jon: knows? who knows? but, again, it's a catch 22 kinda thing and chicken and egg kinda thing because how do you get these outputs without training? but, again, they could have paid for it at the start. right? but, anyways

Craig: yeah. and the courts, ordered an immediate injunction, meaning openai has to stop using this data now in germany, though, you know, they will appeal. the catch is that we're creating the splinter net of ai because different countries are ruling differently. yeah. we might end up with fragmented internet, where us users get a smart ai, get a safe but limited ai. and if we, you know, wanna use the really good ai, we gotta go to deep seeking china or something where they're not limited by these rules. so critical worry that, this kills little guys. microsoft and google can afford to pay, gemma, and the new york times billions in licensing fees while, small ai startups cannot, meaning big tech might become the only option for powerful ai. and that leads me to ask the question about downstream models. so when i say downstream models, many people use, the lamma models as an example and build on top of them. you know, nvidia does it. many companies do because they're open source models. now if any of those models are built on an open source model that was trained on any of this material, are they now liable for this?

Jon: good point. yeah. i don't know.

Craig: so i'm just trying to figure out how far down it affects and what the ripples are. but, you know, as i read this and go through it, before i'm thinking it was one thing, but now, you know, is this just becoming another moat? you know, if you wanna be a tobacco company, you gotta pay millions of dollars. so only the existing tac tobacco company or companies with, you know, huge pockets can get in the industry. so that's what happened with vaping where they came along and the tobacco companies, lobbied to get regulations around it, and you had to, get all your stuff certified. and the certification process cost a million bucks or whatever, so all the small companies shut down. you can say that's a good thing or bad thing, but it is definitely in favor of big business. so when you start looking at this, oh, if you wanna do any of this stuff, you're gonna have to pay more and pay all these people. and now all of a sudden, there's a moat, to join that only these multibillion dollar companies can yeah. you know. so is that strategic? is it not strategic? i think in the end, because ai is strategically important to every government's future and every country's future that we're gonna see some country's government stepping in on some of this. i'm not exactly sure what that means because at this stage, governments don't wanna be last in the ai race. they don't wanna be the one country that's putting all the rules on their ai, and everyone else is jumping leaps and miles ahead of them because they don't have those rules. so it's a complex solution, and i am not, go on either side of it. will i'm just saying that well, i'm trying to make awareness, making sure that we're watching these things, talking about all the sides, all the legal seats so people can see it and understand it and understand what it might mean for the future of what happens. and, you know, i could be way off about government stepping in, but you see what the us is doing with openai and a few other ai. they're spending billions of dollars on those ai data centers you know? i really think there's gonna if nothing else, there might be, rules for the but not for me situation when it comes to this kind of stuff. so we'll have to wait and see. yeah. so the verdict move fast and break things era of ai is hitting a concrete wall. companies can no longer treat the internet like an all you can eat buffet. so does this mean artists finally get their paycheck, or are we about to see ai subscriptions double in price? and even if these fines and stuff come through, will the money actually get back to the artist? let me know your take in the comments, and, we'll keep tracking, where the money goes.