-1

Various authors are suing OpenAI and Meta for alleged copyright infringement in using the authors' works to train the companies' large language models (LLMs).

My understanding is that OpenAI has revealed very little about where they obtained the training data used to train its GPT models. (They've only said that 16% of the training dataset for GPT-3 was comprised of collections of books known "Books1" and "Books2". Books1 is believed to be composed of most of all of the content of Project Gutenberg, which is in the public domain, and Books2 is speculated to come from a piracy website.) Meta, however, has revealed that a small fraction of its training dataset consisted of "Books3". Books3 is a large database of copyrighted books that Shawn Presser compiled from the online piracy site Bibliotik and posted publicly, specifically to serve as a LLM training dataset in order to counter OpenAI's alleged monopoly on such databases.

My understanding is that the legal question whether OpenAI's and Meta's use of these databases constitutes fair use is novel and contested, since their LLMs don't reproduce the books verbatim but modify their text.* (This question is out of scope for this Law.SE question.) But it seems to me Shawn Presser's creation and publication of Books3 is a much more clear-cut case of copyright infringement, since he simply reproduced the books verbatim and posted them publicly.

So the legal case against Presser seems much stronger and more clear-cut than the cases against OpenAI or Meta. And yet, as far as I can tell, no one is suing Presser for copyright infringement. Why not?

The cynical (but probably correct) answer is that "OpenAI and Meta have a lot more money than Shawn Presser does." But is there a legal argument for suing the companies for copyright infringement but not Presser?

*My understanding is that the legal question of whether the LLMs' modifications of the source material is significant enough to constitute fair use is actually independent of whether or not the specific copies of the works that the companies used as sources were themselves the product of copyright infringement; copyright infringement does not automatically "carry down" a chain of derivations. The fact that OpenAI and Meta didn't pay for the sources therefore seems like a bit of a red herring; even if they had bought the books individually (without licensing their reproduction), then that would not change whether or not their use of them constituted fair use.

Improve this question

1 Answer 1

Reset to default
2

Because you sue people who have money

Your remedy for copyright breach is damages; therefore, you sue people who can afford to pay. OpenAI and Meta have an advantage over the named individual because they have a lot of cash.

Improve this answer
3
  • 1
    and those companies still engaged in copying (and thus copyright infringement) by downloading the compilation.
    – Trish
    Nov 13 at 10:33
  • @Trish My understanding is that if the downloading is merely an intermediate step toward an end goal of producing a fair-use modification of the works - which is a contested legal question - and not to directly consume the works, then the downloading constitutes fair use. Not all copying of copyrighted material violates copyright.
    – Very Tiny Brain
    yesterday
  • @Trish See the following relevant example from Perfect 10 v. Google, Inc., 416 F. Supp. 2d 828 (C.D. Cal. 2006), in footnote 17, at 852: "Local caching by the browsers of individual users is noncommercial, transformative, and no more than necessary to achieve the objectives [s o] local browser caching is fair use."
    – Very Tiny Brain
    yesterday

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .