Plaintiffs accuse Meta of ‘torrenting’ data to train AI models in copyright lawsuit

Plaintiffs in a class-action copyright infringement lawsuit against Meta are accusing the tech giant of relying on file- sharing methods, or “torrenting,” to collect massive amounts of data to train its large-language artificial intelligence models.

The allegation offers new insights into the training of Meta’s open-source Llama models, part of a months-long dispute in the case over efforts by plaintiffs to access company documents that they say will show the company relied on pirated content.

“Meta’s torrenting of pirated works...