New legal files in an AI intermediary case against Meta add credibility to Previous reports that the company “pauses” with books publishers on license agreements to provide some of its generative AI models with training data.
Deposits are linked to the case Kadrey v. Meta Platforms – one of the many cases of this type ending with the American judicial system which opposed the companies of AI to the authors and other holders of intellectual property. For the most part, the defendants in these cases – AI companies – said that training on the content protected by copyright is “fair use”. The complainants – Copyright holders – have disagreed with empty.
The new files submitted to the Court on Friday, which include partial transcriptions of meta-employees deposits taken by lawyers for the complainants in the case, suggest that certain meta-personals have estimated that the IA training data licenses For books may not be evolving.
According to a transcription, Sy Choudhury, which directs the partnership initiatives of Meta AI, said that Meta’s awareness of various publishers was greeted by “a very slow adoption of commitment and interest”.
“I do not remember the full list, but I remember that we had made a long list by initially browser the Internet of the best publishers, and Cetera,” said Choudhury, according to the transcription, “and we did not have No contact and feedback from – From a lot of awareness of cold calls to try to establish contacts.
Choudhury added: “There were a few, as, you know, committed, but not much.”
According to the transcriptions of the court, Meta interrupted certain license efforts of books linked to the AI in early April 2023 after meeting the “timing” and other logistics setbacks. Choudhury said that some publishers, in particular fictional books publishers, are not actually proved to be rights over the content that Meta was considering the granting of licenses, according to a transcription.
“I would like to emphasize that the – in the fiction category, we quickly learned from the business development team that most of the publishers we were talking about, they themselves represented that they did not, in fact , license rights for us, ”said Choudhury. “And it would therefore take a long time to get involved with all their authors.”
Choudhury noted during his deposit that Meta made another opportunity for license efforts linked to the development of AI, according to a transcription.
“I am aware of license efforts such as, for example, we have tried to concede to the 3D worlds of different motor and game manufacturers for our AI research team,” said Choudhury. “And in the same way that I describe here for fiction and manual data, we had very little commitment to have even a conversation […] We decided to – in this case, we decided to create our own solution. »»
The lawyer for the complainants, including the successful authors Sarah Silverman and Ta-Nehisi Coates, changed his complaint several times since the case was filed before the American district court for the northern California district, the division from San Francisco in 2023. The last complaint modified by the California district, in 2023. Submitted by the lawyer for the complainants alleys that Meta, among other offenses, referenced to certain pirated books with books protected by copyright Available for a license to determine whether it was logical to conclude a license agreement with a publisher.
The complaint also accuses the meta of Using “shadow libraries” containing hacked electronic books To train several of the company’s AI models, including its popular Llama series of “open” models. According to the complaint, Meta may have obtained certain libraries via the torrent. The torrent, a way to distribute files on the web, requires that “seed” torrents or simultaneously download the files they are trying to get – that the complainants affirmed is a form of copyright violation.