Let's be clear, this is a tech demo, the application is not going to be prompting infinite streams of low quality music. The application of this stuff is to enable creative tools (got a beat and some vocals, give me a bassline and backing chorus), real time computer jam partners, to provide "good enough" music for indie games/movies.

These models are still in their early phases - If AI was combustion engines, we wouldn't even be to railroads yet. Despite that, they're still fun to play with and useful when used correctly.

There are tons of relatively cheap music libraries with human-composed and human-produced music that is vastly higher quality than MusicLM's output - they are used for documentaries/commercials/low-budget projects and are loaded with metadata that make it easy for directors and editors to quickly find what they are looking for. Low-quality, weird-sounding AI-generated music with wonky rhythms, an inconsistent beat, bizarre digital artifacts and a generally unpleasant lo-fi sound that probably has to be patiently generated and listened to over and over again to find a usable result is highly unlikely to beat out already very cheap, much higher-quality and easily searchable human-written music for even the lowest budget projects.

Yeah, as far as electronic music goes, maybe people don't realize just how good (and automated) the current tools are. And have been for years.

The first version of Dall-E was mediocre. Look where we are now.

It's still pretty mediocre if you're being objective. The true hit rate is very low. Granted, it can produce some amazing things from time to time.