Comment by TGower

People aren't buying Harry Potter action figures as a subtitute for buying the book either, but copyright protects creators from other people swooping in and using their work in other mediums. There is obviously a huge market demand for high quality data for training LLMs, Meta just spent 15 billion on a data labeling company. Companies training LLMs on copyrighted material without permission are doing that as a substitue for obtaining a license from the creator for doing so in the same way that a pirate downloading a torrent is a substitue for getting an ebook license.