I'm expecting it to switch an established pattern up (more than it does) on the third repeat, just as I get used to it, and to have fractal-like layers of patterns, which I didn't observe.
I think what makes it work for people is that it evokes that kind of IDM-ish hipster acid music where hearing something a bit unmusical without a 4-to-the-floor "release" is acceptable. But I'm not sure this system would have as much luck putting together an actual "acid banger" like Lochi's London Acid City, Purple Plejade's Blanche, Hardfloor's Acperience etc.
The problem is that a computer can't know when it really came up with a "banger" of a sequence. When you have a real musician, they can sense when a particular sequence really "clicks" for people, somehow it just accents in places that sound cool. Then they can move ahead with that sequence, tease it and build it and bring it to resolution. This is the holy grail of generated music, I think, to somehow get a computer to a place where it can recognize whatever that quality is.
Same. You could generate a bunch of random notes using realistic string instrument samples, and be just as far from Bach.
It's a cool project and concept, but people in the know would notice.