It would be so funny if this ended with Nvidia getting robbed.
So the amend alleges, Nvidia having used/stored/copied/obtained/distributed copyrighted works (including plaintiffs’), both through databases available on Hugging Face (‘Books3’ featured in both ‘The Pile’ and ‘SlimPajama’), or pirating from shadow libraries (like Anna’s Archive), to train multiple LLMs (primarily their ‘NeMo Megatron’ series), and distributing the copyrighted data through the ‘NeMo Megatron Framework’; data which was ultimately sourced from shadow libraries.
It’s quite an interesting read actually, especially the link to this Anna’s Archive blog post. Which it grossly pulls out of context, as plaintiffs clearly despise the shadow libraries too: as they have ultimately provided access to their copyrighted material.
Especially the part: “Most (but not all!) US-based companies reconsidered once they realized the illegal nature of our work. By contrast, Chinese firms have enthusiastically embraced our collection, apparently untroubled by its legality.” makes me wonder if that’s the reason why models like Deepseek, initially blew Western models out of the water.
You can ask deepseek detailed questions about Harry Potter books and it responds intelligently with (almost) quotes from the book.
Ask chatGPT and it will respond to questions but denys it has read any book.
Allegedly most valuable company on the planet in all of history (can’t afford books). Allegedly not a bubble or fraud.
Sadly I think it’s more that there isn’t really a standard way to buy books and other media in bulk at the scale of which AI training usually requires. So the companies realise they can save both time and money in just pirating after calculating the fine risk. Its just a bonus that they usually get away with it and that the fines would likely be cheaper than a legit transaction. But i do think it’s the bulk data packaging that makes piracy look more attractive to them at the get-go.
Heck, even video game publishers often source their roms for their official re-releases from pirated copies because pirates are better at preserving data and keeping it in a nice friendly format. Easier to search for it on the web and download it then it is too goo into their own archives and rip it themselves, if they even still have original copies, cause they sure as hell didn’t keep their source code.
There is also no standard way of buying a DRM free epub for personal use so I’m fine downloading them from Anna too :)
Yeah, no, this genuinely doesn’t make sense as there are legitimate repositories for these books and can do business-to-business negotiations for access to them. Even libraries have access to ebooks at bulk scale.
Those kinds of negotiations if they haven’t been done by other companies before, they won’t have a process for it already in place. There’d be lots of friction for the first of such deal. Both in lots of legal work and software development to make sure they only get access relevant to the deal made.
It’s not something they can just be like “hey, here’s the FTP URI”. Because these legitimate repositories you speak of, like Amazon I guess, will already have existing deals with publishers. Currently as they stand, these deals may not be compatible with Amazon sharing their IP with other companies. So they will either have to redo those deals or restrict access of specific titles to the likes of Nvidia.
Well, I suppose they could buy access to Amazon’s kindle servers
Hmmm. I wonder what Amazons LLMs are trained on.
Are you suggesting that there is a use case for piracy that has less to do with saving money than it does with convenience and easy access to media in one place?
Wait, so piracy is theft?
Not if it’s the rich guys doing it.
AA might be digging their own grave. Overtime the knowledge gets accumulated in the hands of a select few and then they’re gonna block people from accessing pirated sites like AA or even worse, AA gets shutdown due to lack of traffic.
It has torrent backup. How would it do either of those things?
It’s a really good thought. IMO what they will be producing with AI wont be knowledge it will be slop.
There is always gonna be an indie writer, a local at the pub singing. They cant stop people creating. Download or buy analog of the stuff you like and store it. We don’t have to be a slave to the mainstream dream…i will say though its hard changing habits…but for me, it starts with me.
But…why?
Just torrent it?Not fast enough
Will they be sued per book?
It’s not stealing when corpos do it.
Meta torrented their training data from the pirate bay. Hell, Spotify initially built their catalog from pirated music. They all do this shit. Corporations are built to steal our shit and sell it back to us. This isn’t any different from pumping oil out of pubic lands and selling it back to us.
wish meta had torrented all the viruses, too, would be fun to read the news of “facebook and instagram are offline as meta suffers from cyberattack”
pumping oil out of pubic lands
this sounds really painful lol
No becaese the lawyer cohort will destroy them.
No, it’s “fair use”
Pirating books is not fair use. Using copyrighted works to train an AI model is not fair use. People seem to grossly misunderstand what fair use is, and how limited its scope is. Don’t believe me? Here’s legal the precedent
In his June ruling, Judge Alsup agreed with Anthropic’s argument, stating the company’s use of books by the plaintiffs to train their AI model was acceptable.
“The training use was a fair use,” he wrote. “The use of the books at issue to train Claude and its precursors was exceedingly transformative.”
However, the judge ruled that Anthropic’s use of millions of pirated books to build its models – books that websites such as Library Genesis (LibGen) and Pirate Library Mirror (PiLiMi) copied without getting the authors’ consent or giving them compensation – was not.
Pirating isn’t but training on copyrighted works is fair use, you just have to buy them.
So they can train with books they have legal acces to? Either bought, rent or licensed?
I was referring to Altman lobbying towards considering AI training as fair use of copyrighted material.
I know that pirating is not fair use. However, AI companies seem to rely on pirated copies to train their slop machines, and they are trying to justify this behavior.
A business is not fair use. They’re taking someone’s intellectual property and using it to make their product useful.
Seems strange. Anna’s Archive makes their collection available for bulk download as torrent files, they shouldn’t need to “cut a deal” for access to that. Just download the torrent and now you’ve got the whole collection available locally.
They do provide direct access to their books for business who are willing to pay.
chaotic neutral
Which, as I said, seems strange. Why don’t those businesses just download the torrents?
Some businesses did in fact do just that.
Ah, low numbers of seeds. Must’ve just not wanted to wait.
Fucking hit and run lmao
Holy shit the greed knows no bounds.
Hmm so nvidia is training llms as well. Are they going to compete with their customers now too? Like anthropic and cursor?
Good. Can’t wait for the bubble to pop.
So we can assume that in the future, only slob written by LLMs will be available. I mean, who would be willing to spend hundreds of hours writing a book when even huge corporations that earn billions from it won’t pay the author a single dime?
The trick is not to pay a dime to read it. Even producing Ai slop has a cost. If no one pays for that it must leave a negative.
Stop buying. Or If you have to buy old stuff second hand. There’s already a surplus.
Alternatively piracy is clearly condoned here so again don’t buy.
Why should this development stop at books? There are already generated books available, mostly children’s books (no one’s thinking about them now).
This development will certainly not end with books - countless other creative and intellectual achievements have long been affected. That is precisely the problem with generative models, whether they involve text, code, video, images, or whatever else. All of this boils down to the fact that the already precarious situation for everyone who creates value by themselves is continuing to deteriorate. Professional work in all these areas will undoubtedly become even more precarious in the future, with artists, designers, and writers, who were already in a difficult position, now being joined by industries such as software development and administrative work.
Please don’t get me wrong: I am anything but a technology pessimist, but the business model of the so-called AI companies is so exploitative and their owners so unscrupulous that, given the status quo (cloud models), I can hardly imagine that this will lead to even halfway fair working conditions or remuneration models for people who create value in the form of intellectual achievements. I mean, this post is a vivid example.
Allegedly, but holy shit if true. Hard to explain yourself out of that one.
