I've seen demos of software that uses AI to split a song into multiple tracks, one for each instrument. THAT is pretty cool. It's not lossless, you're going to lose some of the human performance because the AI has to reconstruct the sound for each instrument and it's not going to be 100% perfect, but it's a really neat (and useful) tool.

Notably, it's not the kind of thing you generally see when tech bros are touting AI.