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