Meta Admits Use of ‘Pirated’ Book Dataset to Train AI

January 12, 2024By rightstech

With AI initiatives developing at a rapid pace, copyright holders are on high alert. In addition to legislation, several currently ongoing lawsuits will help to define what’s allowed and what isn’t. Responding to a lawsuit from several authors, Meta now admits that it used portions of the Books3 dataset to train its Llama models. This dataset includes many pirated books.

Source: Meta Admits Use of ‘Pirated’ Book Dataset to Train AI * TorrentFreak

RightsTech Extra

AI Companies Get the Picture

By Paul Sweeting

One of the biggest challenges facing copyright owners in grappling with the rapid development of generative AI technology, apart from a murky legal status, has been market failure, as discussed here in previous posts. The amount of existing material needed to train gen-AI models is so great, and so varied, that gauging the value of any one piece of it to establish a market price for it for licensing purposes is often effectively impossible. One group of rights owners is finding willing buyers among AI companies, however. An active, albeit for various reasons mostly sotto voce market has begun to emerge for the use of images held in large photo archives and by photo agencies and social media platforms, complete with per-unit industry and at least a nod toward creator attribution... Read More >>

Share this: