Chat GPT Is Eating the World

• Judge Chhabria reluctantly granted partial summary judgment to Meta finding its use of copyrighted books was fair use to train its AI model.
• But Judge Chhabria faulted the plaintiffs for not presenting sufficient evidence to raise a genuine issue of material fact that LlaMa would create market dilution of their works.
• Judge Chhabria even stated in dicta that in most cases AI training on copyrighted works will likely be illegal due to market dilution.
• Judge Chhabria spent pages of the opinion developing this new theory of market harm.

Judge Chhabria just issued his decision granting Meta partial summary judgment on fair use.

But Judge Chabbira did so reluctantly, stating that the plaintiffs failed to present sufficient evidence of market dilution but expecting in other cases a better record of market dilution would have a different result.

Indeed, on p. 1 of the opinion, although granting summary judgment on fair use for Meta, Judge Chhabria point-blank said that “in most cases the answer will likely be yes,” that AI training on copyrighted works will be illegal. “The upshot is that in many circumstances it will be illegal to copy copyright-protected works to train generative AI models without permission. Which means that the companies, to avoid liability for copyright infringement, will generally need to pay copyright holders for the right to use their materials.”

Judge Chhabria stressed: “And, as should now be clear, this ruling does not stand for the proposition that Meta’s use of copyrighted materials to train its language models is lawful. It stands only for the proposition that these plaintiffs made the wrong arguments and failed to develop a record in support of the right one.”

Wow.

I will update this page as I digest the opinion. Here are some of the key passages:

meta’s downloading books datasets was for the ultimate purpose to train its ai model

The last issue relating to the character of Meta’s use (and thus the first factor) is the relationship between Meta’s downloading of the plaintiffs’ books and Meta’s use of the books to train Llama. To the extent the plaintiffs suggest that the former must be considered wholly separately from the latter, they are wrong. To be sure, Meta’s downloading is a different use from any copying done in the course of LLM training. But that downloading must still be considered in light of its ultimate, highly transformative purpose: training Llama. See Authors Guild v. Google, Inc. (Google Books), 804 F.3d 202, 216–18 (2d Cir. 2015) (considering the creation of digital copies of books in light of the secondary user’s overall purpose of creating a searchable database); cf. Warhol, 598 U.S. at 533 (noting that different uses must be considered separately, but that “the same copying may be fair when used for one purpose but not another”); contra Order on Fair Use at 18, Bartz, No. 24-cv-5417. Because Meta’s ultimate use of the plaintiffs’ books was transformative, so too was Meta’s downloading of those books.

The plaintiffs also assert that Meta downloaded multiple copies of the databases containing their books, and that only some of these copies were used for LLM training, so the downloading of the ones that were not used for training cannot be fair use. But all of the downloads the plaintiffs identify had the ultimate purpose of LLM training. The plaintiffs say that Meta only used its initial October 2022 download of LibGen to see whether the books in the database made for good training data. Pls. Reply at 12. This is a reasonable first step towards training an LLM. See Pls. MSJ Ex. 32 at 3. The plaintiffs say that Meta cross-referenced its next download of LibGen and its first download of Anna’s Archive with publisher catalogues to see whether it was still worth pursuing licensing (or whether all the books available for licensing were already included in those databases). But the plaintiffs concede that these downloads were also used as training data. See Pls. Reply at 13–14. And there is no indication that comparing the books in those databases to the books in another entailed any additional copying. So that cross- referencing alone cannot create infringement liability and does not need to separately constitute fair use. Cf. Warhol, 598 U.S. at 534 & n.10 (discussing application of fair use test to different uses).

TRANSFORMATIVE PURPOSE FOR TRAINING AI MODEL

This factor favors Meta. There is no serious question that Meta’s use of the plaintiffs’ books had a “further purpose” and “different character” than the books—that it was highly transformative. The purpose of Meta’s copying was to train its LLMs, which are innovative tools that can be used to generate diverse text and perform a wide range of functions. Cf. Oracle, 593 U.S. at 30 (transformative to use copyrighted computer code “to create a new platform that could be readily used by programmers”). Users can ask Llama to edit an email they have written, translate an excerpt from or into a foreign language, write a skit based on a hypothetical scenario, or do any number of other tasks. The purpose of the plaintiffs’ books, by contrast, is to be read for entertainment or education.

The plaintiffs do not meaningfully disagree about Llama’s purpose. To the contrary, they acknowledge that LLMs have “end uses” including serving “as a personal tutor,” assisting “with creative ideation,” and helping users “generate business reports.” And several of the plaintiffs testified to using LLMs for various purposes, all distinct from creating or reading an expressive work like a novel or biography—for instance, to find recipes, get tax or medical advice, translate documents, or conduct research. All of these functions are different from the use to which the plaintiffs’ books are generally put. So copying the books to develop a tool that can perform those functions is a use with a different purpose and character than the books themselves.

A

The plaintiffs’ law professor amici argue that Meta’s use has the same purpose and character as the books because an LLM training on a book is akin to a human reading one. One might also analogize Meta’s copying of the books to train Llama to a situation in which a professor copies a book and gives it to a student so that the student can use the knowledge from the book (along with knowledge they get from other books) to go do great things. But there are a few important differences.

First, an LLM’s consumption of a book is different than a person’s. An LLM ingests text to learn “statistical patterns” of how words are used together in different contexts. It does so by taking a piece of text from its training data, removing a word from that text, predicting what that word will be, and updating its general understanding of language based on whether it was right or wrong—and then repeating this exercise billions or trillions of times with different text. This is not how a human reads a book.

Second, unlike the hypothetical professor, Meta did not just give the plaintiffs’ books to one person. Meta copied the plaintiffs’ books as part of an effort to create a tool that can generate a wide range of text. Any person can use that tool to help them create further expression, whether by having it help them brainstorm or research for a creative writing project (like plaintiff David Henry Hwang, a playwright and screenwriter) or by having it write code to develop new software programs (like Lockheed Martin). By creating a tool that anyone can use, Meta’s copying has the potential to exponentially multiply creative expression in a way that teaching individual people does not. Cf. Oracle, 593 U.S. at 30.

In contrast to the copyright professors, the plaintiffs make different (and much weaker) arguments for why Meta’s use is not transformative. For example, the plaintiffs suggest that Llama has “no critical bearing” on their books, the way criticism or parody would. But “critique or commentary on the original” are not “the only uses that will furnish a justification ultimately qualifying as fair use.” Romanova, 138 F.4th at 115. To the contrary, a use that enables “the furnishing of valuable information on any subject of public interest” or renders “a valuable service to the public” might be justified, especially where that benefit is “provided without allowing public access to the copy.” Id.

In addition, the plaintiffs argue that Meta’s use is not transformative because Llama will output material that “mimics” the plaintiffs’ work or writing styles if prompted to do so. Therefore, the plaintiffs say, Meta’s use “merely amounts to a ‘repackaging’” of their books. The plaintiffs point to evidence that they say shows that Meta trained Llama to be able to emulate certain writers’ styles. Pls. Reply Exs. 111–14. But this evidence does not show that Meta trained Llama to repackage the plaintiffs’ works. To the contrary, as noted above, even using “adversarial” prompts designed to get Llama to regurgitate its training data, Llama will not produce more than 50 words of any of the plaintiffs’ books. Pls. Reply Ex. 79 ¶¶ 79, 82–83, 92. And there is no indication that it will generate longer portions of text that would function as “repackaging” of those books. Nor is there even any indication that, as the plaintiffs’ amici claim, Meta developed Llama with the purpose of enabling it to create books that compete with the plaintiffs’ (without rising to the level of repackaging them).5

5 This sort of competition—from AI-generated books that are like the plaintiffs’ but not similar enough to be infringing—is also discussed at length with respect to the fourth factor.

So at most, this evidence shows that Meta wanted Llama to be able to generate text in certain styles. But style is not copyrightable—only expression is. See 17 U.S.C. § 102(b); cf. Mattel, Inc. v. MGA Entertainment, Inc., 616 F.3d 904, 916 (9th Cir. 2010). Even if one possible use of Llama is to generate text with similarities to unprotectable aspects of the plaintiffs’ books, that does not mean Meta’s copying had the same purpose as those books.6

6 By contrast, consider an LLM that was designed to be used to create works substantially similar to those on which it was trained, or to create works that competed with the originals without being substantially similar. Using copyrighted works to train such an LLM could be less transformative than using them to train a general-purpose LLM, because that use would have the purpose and character of enabling an LLM to develop substitute works. That said, even then, training the LLM would still likely be at least somewhat transformative; transformativeness isn’t an on-off switch.

JUDGE CHABBRIA DEVELOPS A NEW THEORY OF MARKET DILUTION UNDER FACTOR 4

JUDGE CHHABRIA DISAGREED WITH JUDGE ALSUP

1. Judge Chhabria viewed Meta’s downloading of pirated books datasets as for the ultimate purpose of training its AI model.
2. Judge Chhabria disagreed with Judge Alsup’s analogy of using books to teach children to write: “According to Judge Alsup, this “is not the kind of competitive or creative displacement that concerns the Copyright Act.” Id. But when it comes to market effects, using books to teach children to write is not remotely like using books to create a product that a single individual could employ to generate countless competing works with a miniscule fraction of the time and creativity it would otherwise take. This inapt analogy is not a basis for blowing off the most important factor in the fair use analysis.”

DOWNLOAD THE OPINION: