[mbrownnyc]

Yes, music discovery is still a problem. Take a look at the first iteration of Audio Galaxy, which I thought was the best possible music recommendation platform available. You'd have a client opened. Humans would join groups. Humans would push tracks to groups. This is how I discovered DJ Shadow, DJ Food, and several other artists back in the day (read: 2001) all because I was part of a trip-hop group. People would post discussion threads, organize, etc. There is a difference between computer recommendation services and human recommendation services... you know, soul... and I haven't found a single thing that's similar to that iteration of AudioGalaxy yet. Tomahawk does look promising.

I shudder when I wind up listening to Kaskade because I started an Astral Projection station on Pandora. Last.fm was a bit better, but just not the same as AudioGalaxy of yore. Algorithms don't push barriers and take risk (or at least do either "properly"); it's the exact opposite of what they're trying to do.

[hemezh]

I also feel that algorithm based recommendation services are not going to be better than humans at least for a long time. Algorithms can recommend related stuff well(?) but that's not enough as our tastes keep changing and as you said they don't take risk.

I am thinking of building a radio like service where the playlist is curated by people listening to it. Do you think it's worth building? Any other inputs you'd like to add?