Anti-Piracy Group Takes Massive AI Training Dataset 'Books3′ Offline
gizmodo.com
That pirated set of works contained 37 GB worth of text for training AI. Companies like Meta have already used it to train their language models.