Instead, he clearly says in his summary judgement that it «is generally illegal to copy protected works without permission,» (CNBC) but the plaintiffs «made the wrong arguments and failed to develop a record in support of the right one.» (The Verge.)
Failed to prove harm
He had previously said from the bench that Meta’s claims were dubious, declaring that copying an author’s works and being capable of «producing an infinite number of competing products,» would «obliterate» the market for them, and «I just don’t understand how that can be fair use.»
However, he found the authors suing had not established a good enough case, and that they failed produce a «potentially winning argument,» writes The Verge.
This claim is now summarily closed, and Meta gets fair use protections for their training on these author’s books — although another claim of torrenting their content remains open.
Very narrow ruling
While handing a restricted win to Meta on this particular claim, the judge also elaborated that «this ruling does not stand for the proposition that Meta’s use of copyrighted materials to train its language models is lawful.»
The judge also specifically ruled that his decision is narrow and related only to the plaintiffs on this claim, saying that «the ruling only affects the rights of these thirteen authors — not the countless others whose works Meta used to train its models.»
This is an important caveat that avoids setting a precedent for the numerous other lawsuits against AI companies for mostly the same reasons; copying and reproducing their content without consent.
Competing opinion for Anthropic
Just two days ago, another federal judge ruled that Anthropic’s use of copyrighted books in training its AIs was comparable to learning school kids how to write.
That case also included charges of Anthropic maintaining a pirated 7 million digital books collection, and those charges will be heard in full court in December.
Read more: The summary judgement. Reports from CNBC, The Verge, and TechCrunch.