Wouldn't it be wonderful if, on the heals of everyone copying our teen social media laws, Australia we're to introduce a mandatory per token royalty for copyright holder not proved to be excluded from the training data, and everyone in the world were to follow suit.
We've got to assume anything ever written, photographed, spoken, or recorded has been scraped, so the burden of proof needs to be to prove it wasn't used or derived from.
Otherwise all copyright is dead. Why can an AI company steal work, but someone else not?
If Men at Work can be sued for taking a riff from a kids song, then surely