The tech is fascinating but the trajectory of it being used to screw artists is already clear. Hopefully artists will lawyer up and get their works out of training data.

Large artist unions should be mandated access to freely and deeply test public models for the purposes of detecting IP in the training data. If you’re making money providing a model, be prepared to show your papers on its training; trust needs to be earned.

Commercial models should be taxed, with revenue supporting ongoing development of tools for smaller artists to protect themselves from being ripped off.

If this effort goes well for artists, I generally expect sound quality of music generation to improve but “humanness” of the public models to sink. Bad news for companies wanting nearly free anodyne music for commercial use, but arguably better for art.

Where do the real prices of services like Udio land after investors stop footing the bills? Do these VC subsidized toys ignite the creative spirit of everyday non-artists to get into the game of actual music creation? What are we missing out on by having a world where very few people make music?