Wanted to hear Dune, The Musical? Well, now you can. A new Gen AI music company, Udio, has come out of beta with an impressive list of investors: Andreessen Horowitz, UnitedMasters, Will.I.AM, Common, a former co-founder of Instagram, among others. The outputs it creates are amazing. Not only the fidelity of the music, but the diversity (everything from orchestral to hip hop + vocals), the tonality (it sounds a little lofi, but well-composed), and the duration (full length songs). Of course, an AI model is only as good as its training data. And just like the comparable Gen AI startup Suno, the founders are not saying where their training data came from. Well, they certianly trained it on much of existing recorded music. When asked, their responses range from "we have to, in order for it to be good", to "everyone else is doing it." So what's going to happen? Can they use the history of recorded music under fair use? Or will they have to compensate rights holders? No one knows yet. But with products this good (along with Suno), consumers will use them. They will be too popular to go away, and gain an upper hand in the debate because of it. Have a listen below ⬇ #musicindustry #musicbusiness #GenAI #musicAI
Have you been trying it already? It's producing really decent results even in the Metal genre. Kind of impressed so far. They're not being open with the training data though. That's the only concern right now.
Wild stuff
Exciting times ahead for the music industry! Can't wait to see how AI continues to revolutionize the music landscape.
The fact that the founders aren’t saying where the training data is from speaks very loudly.
Focusing on the product instead of the means of production and the training data feels very religious and not in a good way.
Senior Operations Executive specializing in Music/Media/Tech ■ Operations ■ Strategy ■ Content & Rights Management
8moInsightful post as always, Drew! I'll admit that the musical style might not be my personal preference but the technical achievement of the AI-generated piece is undeniable. On an intellectual level, it's super impressive that this is even possible at this point! Your point about training data is crucial. The question of fair use versus compensating rights holders remains unanswered. As someone excited about AI's potential in music creation and also extremely concerned about its impact (particularly on songwriters and composers) I completely agree about the need for clear guidelines. Technology as usual continues to race ahead and establishing frameworks for responsible use right now is key (and urgent). It will be interesting to see how this all plays out.