The Ableton project format is fairly simple (compared to what it could have been at least), anyone tried just throwing the XML from there at a LLM model and see if they could edit them "raw"?

Anyone who collects training data from pro tool users? Would make sense to me.

Most of the (existing) artist community seems vehemently against introducing generative AI into their workflows, for better or worse. I'm guessing it would be even harder to convince this community to agree to share data that could replace their jobs.

Yea you'd at least need to compensate well.

Maybe the labels have way here, but they won't benefit from less gatekeeping with AI.

An alternative to "watching" someone would be getting the raw input and the final output.

> Yea you'd at least need to compensate well.

Again, guessing, but you'd have to compensate them enough to cover expected/dreamed up life-time earnings, it'll quickly balloon to unrealistic levels, considering the amount of data you'd need, from actual professionals.

You don't need data from actual professionals, tools can be trained from the finished product that can make more of the finished product. We know this from AI use in writing, images, voice and video.

Ultimately an "AI Ableton" isn't going to look like Ableton, it'll be a simple multitrack editor with a cursor style chat on the side.

Or youtube/spotify allowing for continious AB tests of songs XD

Oh god, please delete this comment! Don't give them ideas! XD

When it happens they won't be asked it'll be taken

Basically the whole “synths and samplers aren’t real instruments” thing from 1970, all over again.

Synths and samplers didn't play themselves