• # Premier commentaire

    Posté par  (site web personnel) . Évalué à 7.

    Je me permet de republier ici le premier commentaire de l’article qui a l’air de dire que les gens de LWN/Developer ont compris l’annonce à l’envers.

    I found the linked article lacking in detail, and found what appears to be a copy of the opinion in [1]. The opinion is not as described in the one-paragraph summary quoted above.

    According to the opinion, the plaintiffs originally had three claims:

    1. A DMCA claim under 17 USC 1202(b). My casual read of that law is that applying it to AI training is, at best, a real stretch. Judge Tigar dismissed this claim because the plaintiffs did not adequately allege that Copilot's outputs are both identical to the originals and also long enough that they might plausibly be subject to copyright protection. This particular section of copyright law only covers cases where you have an identical copy, so the plaintiffs lose because they didn't allege an identical copy (with enough specificity - they did cite a study claiming that maybe some of the time LLMs might output verbatim copies of large amounts of text, but the judge found that too speculative).
    2. A contract claim against "all defendants"[2] alleging breach of contract under open source licenses. Tigar refused to dismiss this claim, because OpenAI tried to argue that they should've sued for copyright infringement instead, and the judge ruled that the plaintiffs can sue either way they want, so the claim stands (for now, anyway). There was also some minor wrangling over whether OpenAI should be dismissed as a defendant with respect to Copilot allegedly being in breach of contract, since they do not fully control Copilot (Tigar said no).
    3. A claim against GitHub that I frankly do not understand very well. I would need to read the plaintiff's briefs to see what they were trying to argue. Anyway, Tigar basically said that breach of contract, by itself, does not entitle you to various other kinds of money damages that the plaintiffs were apparently trying to get out of GitHub, so this claim was dismissed.

    So Judge Tigar left only one claim standing, which alleges a breach of contract, specifically of the open-source licenses at issue in this case. As for "unlawfully trained on their work" - well, that's the plaintiff's side of the story, but we won't know if it's true until we get a ruling on that exact claim. One possible source of confusion is that the one remaining count does separately allege that both Copilot and Codex breached these contracts, so maybe you could describe that as "two claims" (but the judge's opinion mostly does not phrase it that way). OTOH, separating "breach of contract" from "open-source license violation" makes no sense at all because those are exactly the same thing in this case.

    Here's the thing: Claim (2) is arguably the most important claim, because it is the claim that goes to the heart of the matter (whether AI training violates open source licenses). Claim (1) is frankly kind of a weird argument, since section 1202 as a whole is really about prohibiting media piracy, by the means of prohibiting users from hiding the fact that a given file is copyrighted (e.g. from hosting providers), and that just isn't a good fit for what GitHub was allegedly doing. I do not fully understand claim (3), but it looks like it has more to do with the amount of damages than with the plaintiff's overall case.

    The other question I have is why the plaintiffs chose to sue under contract law instead of copyright law. Even though the judge determined they were allowed to do that, it seems strange. Copyright law gets you statutory damages and various equitable remedies (up to and including impoundment and destruction of the infringing material). Contract law is somewhat favorable to "efficient breach" and may not get you nearly as much if you win. It is possible that this is tangled up with their attempt to get extra damages out of GitHub - if so, then I'm not sure where their lawyers go from here, but they may want to amend their complaint again.

Suivre le flux des commentaires

Note : les commentaires appartiennent à celles et ceux qui les ont postés. Nous n’en sommes pas responsables.