Meta dragged into the court for AI training

A dispute between several authors and meta has now become a classic file sharing process. The social media group must be responsible for downloaded and spreading books via a petorrent.

Book use generally allowed

In the legal dispute between Meta and 13 authors, including comedian Sarah Silverman and the Pulitzer award winner Junot Díaz, the US company won a stage victory. A federal judge rejected the main part of the lawsuit for the use of copyrighted books for training Metas Ki model Llama. So the allegations remain that Meta has procured the works by paptyorrent, as Ars Technica reported.

Judge Vince Chhaubria decided that this aspect of the case would be negotiated. For July 11th, a meeting is scheduled to coordinate the further procedure in which META and the plaintiff side. Chhablria indicated that it should be difficult for the authors to win this point -especially because of inadequate evidence -but he rejected Metas argument that the torrent was “irrelevant” for the fair -use exam.

According to the court, downloading over 80 terabytes of books from the Libgen shadow library could be relevant at least in several ways. The fact that Meta has apparently deliberately divided into piracy after failed license negotiations could represent an indication of “Bad Faith”, i.e. dishonest behavior. This in turn could negatively influence the character of use-a central criterion in fair-use considerations.

Open questions

In addition, it could be problematic for META if it could be proven that the Torrenting has indirectly supported or strengthened shadow libraries. So far there is no valid evidence, but the procedure has not yet been completed on this point. Chhaubria also made it clear that downloading the books could not be considered detached from the actual purpose – the AI ​​training. Since the training of the LLAMA models is considered “highly transformative”, the data acquisition should also be evaluated from this point of view.