[TheSoftwareGuy]

It already applies to real people, doesn't it? I.e. if you read a book, you're not allowed to start printing and selling copies of that book without permission of the copyright owner, but if you learn something from that book you can use that knowledge, just like a model could.

[triceratops]

Can I download a book without paying for it, and print copies of it? Stash copies in my bathroom, the gym, my office, my bedroom etc. to basically have a copy on hand to study from whenever I have some free time?

What about movies and music?

[nightpool]

Yes, you're allowed to make personal copies of copyright works that you own. IANAL, but my understanding is that if you're using them for yourself, and you're not prevented from doing so by some sort of EULA or DRM, there's nothing in copyright law preventing you from e.g. photocopying a book and keeping a copy at home, as long as you don't distribute it. The test case here has always been CDs—you're allowed to make copies of CDs you legally own and keep one at home and one in your car.

[carlosdp]

That's not a one-to-one analogy. The LLM isn't giving you the book, its giving you information it learned from the book.

The analogous scenario is "Can I read a book and publish a blog post with all the information in that book, in my own words?", and under US copyright law, the answer is: Yes.

[ajross]

> Can I download a book without paying for it, and print copies of it?

No, but you can read a book, learn its contents, and then write and publish your own book to teach the information to others. The operation of an AI is rather closer to that than it is to copyright violation.

"Should" there be protections against AI training? Maybe! But copyright law as it stands is woefully inadequate to the task, and IMHO a lot of people aren't really treating with this. We need a functioning government to write well-considered laws for the benefit of all here. We'll see what we get.

[nickpsecurity]

To support your point, lawsuits are already coming in for illegal copying of books:

https://www.theverge.com/2024/8/20/24224450/anthropic-copyri...

https://www.reuters.com/legal/litigation/google-sued-by-top-...

[Terretta]

Is the book online and accessible to your eyeballs through your open standards client tool, such that you can learn from seeing it?

[ben_w]

To the extent that this is how libraries function, yes.

The part of that which doesn't apply is "print copies", at least not complete copies, but libraries often have photocopiers in them for fragments needed for research.

AI models shouldn't do that either, IMO. But unlimited complete copies is the mistake the Internet Archive made, too.

[paulddraper]

> Can I download a book without paying for it

Yes, you can read books without paying, if that's how it is offered.

And you can photocopy books you own for your own personal use. But again....the analogy is remembering/leaning from a book.

[dylan604]

owning a copy and learning the information is not the same. you can learn 2+2=4 from a book, but you no longer need that book to get that answer. each year in school, I was issued a book for class, learned from it, returned the book. I did not return the learning.

musicians can read the sheet music and memorize how to play it, and no longer need the music. they still have the information.

[notatoad]

>Can I download a book without paying for it

if you have evidence that openAI is doing this with books that are not freely available, i'm sure the publishers would absolutely love to hear about it.

[m1el]

when it comes to real people, they get sued into oblivion for downloading copyrighted content, even for the purpose of learning. but when facebook & openai do it, at a much larger scale, suddenly the laws must be changed.

[ryoshu]

Case in point - https://en.wikipedia.org/wiki/Aaron_Swartz

[conjectures]

It does apply to people? When you read a copy of a book, you can't be sued for making a copy of the book in the synapses of your brain.

Now, if you have eidetic memory and write out large chunks of the book from memory and publish them, that's what you could be sued for.

[Terretta]

> when it comes to real people, they get sued into oblivion for downloading copyrighted content, even for the purpose of learning.

Really? Or do they get sued for sharing as in republishing without transformation? Arguably a URL providing copyrighted content, is you offering a xerox machine.

It seems most "sued into oblivion" are the reshare problem, not the get one for myself problem.

[alabastervlog]

This is why I think my array of hard drives full of movies isn't piracy. My server just learned about those movies and can tell me about them, is all. Just like a person!

[bee_rider]

These AI models are just obviously new things. They aren’t people, so any analogy about learning from the training material and selling your new skills is off base.

On the other hand, they aren’t just a copy of the training content, and whether the process that creates the weights is sufficiently transformative as to create a new work is… what’s up for debate, right?

Anyway I wish people would stop making these analogies. There isn’t a law covering AI models yet. It is a big industry at this point, and the lack of clarity seems like something we’d expect everybody (legislators and industry) to want to rectify.

[codedokode]

Model cannot "learn" because it is not a human. What happens is a human obtains "a free copy" of a copyrighted work, processes it using a machine and sells the result.

[amelius]

Totally agree. Except the current administration probably will interpret things the way they see fit ...

[pier25]

> just like a model could

Not really. You can't multiply yourself a million times to produce content at an industrial scale.

[aiono]

Can I pirate books to train myself?

[echelon]

If models can learn for free, then the models (training code, inference code, training data, weights) should also be free. No copyright for anybody.

And if you sell the outputs of your model that you trained on free content, you shouldn't be able to hide behind trade secret.

[crorella]

> just like a model could

It is not remotely the same, the companies training the models are stealing the content from the internet and then profiting from it when they charge for the use of those models.

[Terretta]

> the companies training the models are stealing the content from the internet

Are you stealing a billboard when you see and remember it?

The notion that consuming the web is "stealing" needs to stop.

[ben_w]

The more literature I consume, and the more I re-draft my own attempt, the more I see the patterns and tropes with everyone standing on the shoulders of those who came before.

The general concept of "warp drive" was introduced by John W. Campbell in 1957, "Islands of Space". Popularised by Trek, turned into maths by Alcubierre. Islands of Space feels like it took inspiration from both H G Wells (needing to explain why the War of the Worlds' ending was implausible) and Jules Verne (gang of gentlemen have call-to-action, encounter difficulties that would crush them like a bug and are not merely fine, they go on to further great adventure and reward).

Terry Pratchett had obvious inspirations from Shakespeare, Ringworld, Faust (in the title!).

In the pandemic I read "The Deathworlders" (web fic, not the book series of similar name), and by the time I'd read too many shark jumps to continue, I had spotted many obvious inspirations besides just the one that gave the name.

If I studied medieval lit, I could probably do the same with Shakespeare's inspiration.

[sidewndr46]

And when I "learn" a verbatim copy of pages of that book, then write those pages out in Microsoft Word & sell those pages its legal?

[tsimionescu]

It doesn't, a real person can't legally obtain a copy of a copyrighted work without paying the copyright holder for it. This is what OpenAI is asking for: they don't want to pay for a single copy of a single book, and still they want to train their models on every single book in history (and song, and movie, and painting, and code base, and anything else they can get their hands on).

[amelius]

Do you know Numerical Recipes in C?

This discussion reminds me of it.

[simion314]

>you can use that knowledge,

Did OpenAI bought one copy of each book, or did they legaly borowed athe books and documents ?

if you copy paste rom books and claim is your content you are plagiarizing. LLMs were provent to copy paste trained content so now what? Should only big Tech be excluded from plagiarizing ?