Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Anthropic cut up millions of used books, and downloaded 7M pirated ones – judge

businessinsider.com

Post reply on HN
Print view View original

541–550 of 686 posts

Re: Anthropic cut up millions of used books, and downloaded 7M pirated ones – judge

#541

Post by homebrewer »

AnthonyMouse wrote: post #381
dmd wrote: post #340

Earlier quoted context omitted.

I love this argument so much. "But judge, there's no way I could ever afford to buy those jewels, so stealing them must be OK."

The argument is more along the lines of, negotiating with millions of individuals each over a single copy of a work would cause the transaction costs to exceed the payments, and that kind of efficiency loss is the sort of thing fair use exists to prevent. It's not socially beneficial for the law to require you to create $2 in deadweight loss in order to transfer $1, and the cost to the author of not selling a single…

I used to order books in English from the US before shipping costs became prohibitive and the cost of shipping the book went to about twice to thrice the cost of the book itself. Is it fair use for me to download books from Anna's Archive now considering that books in English are not available in my region through other means (including the vast majority of ebooks)?

Rhetorical question, we all know that me reading books is not "transformative" so it won't be considered fair use for me to yoink them (transformative as in transforming more damage to the society at large into more money for the already rich).

Re: Anthropic cut up millions of used books, and downloaded 7M pirated ones – judge

#542

Post by markhahn »

pyman wrote: post #530
conradev wrote: post #515

Earlier quoted context omitted.

My understanding is that Suno does not sell music , but instead makes a tool for musicians to generate music and sells access to this tool . The law doesn't distinguish between basement and cloud – it's a service. You can sell access to the service without selling songs to consumers.

What does "fair use" even mean in a world where models can memorise and remix every book and song ever written? Are we erasing ownership? The problem is, copyright law wasn't written for machines. It was written for humans who create things. In the case of songs (or books, paintings, etc), only humans and companies can legally own copyright, a machine can't. If an AI-powered tool generates a song, there’s no author i…

why are you saying "memorize"? are people training AIs to regurgitate exact copies? if so, that's just copying. if they return something that is not a literal copy of the whole work, then there is established caselaw about how much is permitted. some clearly is, but not entire works.

when you buy a book, you are not acceding to a license to only ever read it with human eyes, forbearing to memorize it, never to quote it, never to be inspired by it.

Re: Anthropic cut up millions of used books, and downloaded 7M pirated ones – judge

#543

Post by markhahn »

pier25 wrote: post #473
itronitron wrote: post #453

Earlier quoted context omitted.

If it's fair use to train a model, that doesn't necessarily imply that the model can be legally used to generate anything.

I've been reading a bit more about this. The training might not be considered fair use if it's not considered transformative. Claude has been considered transformative given it's not really meant to generate books but Suno or Midjourney are absolutely in another category.

really? so Suno or Midjourney can produce literal copies of works they were trained on?

Re: Anthropic cut up millions of used books, and downloaded 7M pirated ones – judge

#544

Post by conradev »

pier25 wrote: post #529
conradev wrote: post #515

Earlier quoted context omitted.

My understanding is that Suno does not sell music , but instead makes a tool for musicians to generate music and sells access to this tool . The law doesn't distinguish between basement and cloud – it's a service. You can sell access to the service without selling songs to consumers.

That's like arguing that a restaurant doesn't sell food because it sells the service of cooking it.

The restaurant is not responsible for E. coli if it’s found, are they? Just cooking it out of the food

Suno can’t prevent humans from copying other humans, it can only make sure that the direct output of its system isn’t infringing.

Re: Anthropic cut up millions of used books, and downloaded 7M pirated ones – judge

#545

Post by Aeolun »

johnnyanmac wrote: post #537
Aeolun wrote: post #518

Earlier quoted context omitted.

This is literally why a lot of people pirate content, yes. It’s pretty much always the only way to obtain the content, even if you are otherwise fine with paying for it.

Yes, and it's technically copyright infringement, even for private use. It's just that damages and enforcement is in feasible. But if you tried to open a black market selling that media: you'd be hunted down to the ends of the earth. Or to China/North Korea, at least.

> But if you tried to open a black market selling that media

Why would you ever do that? Nobody would buy it. They'd just get it in the same place you did.

Re: Anthropic cut up millions of used books, and downloaded 7M pirated ones – judge

#546

Post by janalsncm »

ffsm8 wrote: post #127
pyman wrote: post #7

These are the people shaping the future of AI? What happened to all the ethical values they love to preach about? We've held China accountable for counterfeiting products for decades and regulated their exports. So why should Anthropic be allowed to export their products and services after engaging in the same illegal activity?

> We've held China accountable for counterfeiting products for decades and regulated their exports We have? Are we from different multi-verses? The one I've lived in to date has not done anything against Chinese counterfeits beyond occasionally seizing counterfeit goods during import. But that's merely occasionally enforcing local counterfeit law, a far cry from punishing the entity producing it. As a matter of fact,…

IP theft is one of the stated reasons for the trade war in the first place. It’s one of the major gripes the US has against China. There are limited means available to restrict a foreign country compared with an entity in the US. The DoJ did sue Huawei and win though.

Whether or not the countermeasures have been effective in practice is a minor detail in the GP point that we would not expect an American company headquartered in the US and conducting significant business in the US to get away with the same thing.

Re: Anthropic cut up millions of used books, and downloaded 7M pirated ones – judge

#547

Post by janalsncm »

NoMoreNicksLeft wrote: post #217
pyman wrote: post #18

Earlier quoted context omitted.

The problem with this thinking is that hundreds of thousands of teachers who spent years writing great, useful books and sharing knowledge and wisdom probably won't sue a billion dollar company for stealing their work. What they'll likely do is stop writing altogether. I'm against Anthropic stealing teacher's work and discouraging them from ever writing again. Some teachers are already saying this (though probably no…

Stealing? In what way? Training a generative model on a book is the mechanical equivalent of having a human read the book and learn from it. Is it stealing if a person reads the book and learns from it?

> In what way?

Downloading the book without paying for it, which is more or less what the judge said.

Re: Anthropic cut up millions of used books, and downloaded 7M pirated ones – judge

#548

Post by markhahn »

burnt-resistor wrote: post #504
stocksinsmocks wrote: post #420

Earlier quoted context omitted.

Anthropic isn’t selling copies of the material to its users though. I would think you couldn’t lock someone up for reading a book and summarizing or reciting portions of the contents. Seven years for thumbing your nose at Autodesk when armed robbery would get you less time says some interesting things about the state of legal practice.

I'm wondering though how the law will construe AI able to make a believable sequel to Moby Dick after digesting Herman Melville's works. (Or replace Melville with a modern writer.)

existing copyright law seems to say you cannot help yourself to significant parts of a work - such as to write your own sequel. I have no idea how the courts establish the degree of copying "owned" by the original author. There would clearly be stories in some way related to Moby Dick that would be legal, but others that were too close.

Re: Anthropic cut up millions of used books, and downloaded 7M pirated ones – judge

#550

Post by mwarkentin »

markhahn wrote: post #542
pyman wrote: post #530

Earlier quoted context omitted.

What does "fair use" even mean in a world where models can memorise and remix every book and song ever written? Are we erasing ownership? The problem is, copyright law wasn't written for machines. It was written for humans who create things. In the case of songs (or books, paintings, etc), only humans and companies can legally own copyright, a machine can't. If an AI-powered tool generates a song, there’s no author i…

why are you saying "memorize"? are people training AIs to regurgitate exact copies? if so, that's just copying. if they return something that is not a literal copy of the whole work, then there is established caselaw about how much is permitted. some clearly is, but not entire works. when you buy a book, you are not acceding to a license to only ever read it with human eyes, forbearing to memorize it, never to quote…

> Specifically, the paper estimates that Llama 3.1 70B has memorized 42 percent of the first Harry Potter book well enough to reproduce 50-token excerpts at least half the time. (I’ll unpack how this was measured in the next section.)

> Interestingly, Llama 1 65B, a similar-sized model released in February 2023, had memorized only 4.4 percent of Harry Potter and the Sorcerer's Stone. This suggests that despite the potential legal liability, Meta did not do much to prevent memorization as it trained Llama 3. At least for this book, the problem got much worse between Llama 1 and Llama 3.

> Harry Potter and the Sorcerer's Stone was one of dozens of books tested by the researchers. They found that Llama 3.1 70B was far more likely to reproduce popular books—such as The Hobbit and George Orwell’s 1984—than obscure ones. And for most books, Llama 3.1 70B memorized more than any of the other models.

Post reply on HN