As Meta defends its AI training practices in court, major publishers and copyright law experts are weighing in against the company. An amicus brief from publishers highlights Meta’s alleged reliance on pirated book archives including Anna’s Archive and Z-Library. Separately, a brief from law professors argues that Meta’s unauthorized copying to train Llama is an “undeniably commercial” use that provides no new transformative meaning and shouldn’t qualify as fair use.
If it was as easy as giving it a nice name, you could just rename seed boxes to fair use machines.
It’s a finicky legal topic, because transformative can be anything and nothing. If you take an album and shuffle the tracks or create a mixtape from multiple albums, that is technically transformative, but most likely not enough to justify fair use.
Right, but it’s pretty obvious that the transformation for AI is very transformative, and it’s lossy at that. After the training, you can’t just duplicate the image it was trained from, even if it was asked a thousand times.