What happens when you use music reviews as prompts for AI generated music?

Originally published at: What happens when you use music reviews as prompts for AI generated music? | Boing Boing

1 Like

“Write me a song about the death of Buddy Holly,” which led the large language model to spit out the lyrics to Don Mclean’s “American Pie” word for word.

The word “write” is clearly ambiguous to our machine-learning overlords. Should have asked “write me a new, original song about the Death of Buddy Holly, that has never been heard before”.

Programming is a precision task. Why shouldn’t asking LLMs to do stuff be as rigorous?


PS Dancing about architecture = parkour, no?


Howsabout this for a short-coming: Google’s Music-LM doesn’t generate material even on par with advertising background music? (cf. brutal takedown of that genre, linked below.)

I’ve been trying out the Google tool. The results remind me of the default tracks that come with a $50 Casio synthesizer’s “ACCOMP” button, with some noise from a mix-tape your little sister recorded off MTV in the 80’s through the speakers (…I know, oddly specific…). My gut reaction to the output was, in every case, grinding annoyance.

How is a huge company like Google wasting time on this irredeemable crap?

FTA: “It’s not even clear how humans make music, actually,” Agostinelli says. “So this is also why, for me, it’s a very interesting research area.”

WHAT?! :roll_eyes: :exploding_head:

Prof. David Cope at U.C.S.C. has been doing profoundly interesting work in this area for decades. His Experiments in Musical Intelligence go deeply into musical structure and the results have fooled professional audiences.

Edit: Another example of what a sound/music pro can do with “AI”, in this case Dr. Stéphane Pigeon, who doesn’t even hold himself out as a composer.

And if you haven’t seen this rant, it may change your ears forever.


This topic was automatically closed after 5 days. New replies are no longer allowed.