To skirt copyright law, Anthropic bought and destroyed millions of physical books to train its "Claude" LLM
Anthropic destroyed millions of print books to build its AI models
You can buy books by the truckload for almost nothing because nobody wants them.
Ex-wife and I had a room in the trailer house dedicated to books, one might have even called it a library. Seriously, cheap bookshelves, all 4 walls, stuffed. Now I have a stack on a small shelf.
Just can't use books anymore. I can choose any of 1,000 epubs on my crappy Android tablet, read in the dark.
I don't really see a problem. It wasn't like rare books nobody had access to. I mean, AI in general yeah. But not the book part.
And that was only after they ran out of pirated ebooks.
No, it was after their lawyers told them that was illegal and would cause them to lose a fair use copyright claim.
Fucking tech bro's
Can't they all just go to space and leave the rest of us alone
Because they want earth. The poors are the ones going to space to work the mines, I'm sure of it. Not sure why anyone thinks they're the ones going to space to live.
It'll be like Weyland Industries in Alien Romulus.
Anything mined in space will be used in space. But you can build orbital colonies that are nice. You just need lots of mass to build stuff and have plenty of water and be big enough to rotate.
The title might be slightly hyperbolic, but still:
The quality of training data fed into the neural network directly impacts the resulting AI model's capabilities. Models trained on well-edited books and articles tend to produce more coherent, accurate responses than those trained on lower-quality text like random YouTube comments.
Anthropic initially chose the quick and easy path. In the quest for high-quality training data, the court filing states, Anthropic first chose to amass digitized versions of pirated books to avoid what CEO Dario Amodei called "legal/practice/business slog"—the complex licensing negotiations with publishers. But by 2024, Anthropic had become "not so gung ho about" using pirated ebooks "for legal reasons" and needed a safer source.
Buying used physical books sidestepped licensing entirely while providing the high-quality, professionally edited text that AI models need, and destructive scanning was simply the fastest way to digitize millions of volumes. The company spent "many millions of dollars" on this buying and scanning operation, often purchasing used books in bulk. Next, they stripped books from bindings, cut pages to workable dimensions, scanned them as stacks of pages into PDFs with machine-readable text including covers, then discarded all the paper originals.
oof.
Is this for real? I can buy a book, scan it and put it on the internet and it wouldn't be piracy? Or is this just the usual "it's not a crime if rich people/evilcorps do it" bs?
Putting the scan on the internet intact would be piracy. Putting up snippets is mostly OK. Ingesting the scans of millions of books into a massive data set and then regurgitating pieces of the masticated, processed mess seems still to be a grey area, but closer to 'mostly OK' than to piracy.
I can buy a book, scan it and put it on the internet and it wouldn’t be piracy?
Yes, but only if you're a multi-billion AI company.
This is stupid. Fuck copyright law.
What'd ChatGPT do?
What'd Meta do?
And meta just pirated them directly from libgen, big fucking deal, copywright law needs to die.
im guessing based on how much of my own library anthropic tries to scan they probably also got a lot more content from the open web anyway and just destroyed the books to make a show of it and hope no one sues them after.
You dont need to unbind or destroy books to scan them and destroying them doesnt magically make reproducing bits or copies suddenly not plagerism.
Why not just move to Piracy Haven - China
To comply with copyright law, not to skirt it. That's what companies that scan large numbers of books do. See for example Authors Guild v. Google from back when Google was scanning books to add to their book search engine. Framing this like it's some kind of nefarious act is misleading.
They also weren't destroying rare books. They were buying in-print books from major retailers, which means that while yes, that is environmentally wasteful, it's not actually destroying books in the classical destruction of knowledge sense since the manufacturer will just print another one if there's demand for it.
This as well. Growing up in a house of book lovers, myself included, destroying a book was akin to kicking a puppy. Realistically though, they're ultimately consumables. They're meant to be bought, used, and replaced as needed. With luck the destruction included recycling as much as possible, seeing as it's mainly paper.
Yeah, this is on the way of being a win. In this case they actually bought the books, which has been one of the biggest issues with LLMs. There's certainly more discussion to be had around how they use the materials in the end, but this is a step in the right direction.
To a certain extent I agree, but you can buy a book and still commit copyright infringement by copying its contents (for use other than personal use)
If this would go to court, it would depend on whether training an LLM model is more akin to copying or learning. I can see arguments for either interpretation, but I suspect that the law would lean more toward it being copying rather than learning