Damn, they really overfit their music models. With image generation and text prediction it's very hard to prove a direct connection, but with four or five of those songs it's unmistakable that the original songs were used to generate the music output

I wonder what the effect will be of fixing the models' overfitting. I'm guessing it'll generate worse music, or they would've done so already.

Quite sad that it took the music industry to notice before any lawsuits with a chance of succeeding got off the ground.