See his notes for a more detailed overview of what was discussed.
It was an interesting talk, covering the origins of the company and the service and an overview of the tech and work behind the service.
One of the more interesting aspects was there tendency to avoid using collaborative filtering and other “social” forms for suggesting music. More than a few questions from an audience were variations of “why don’t you add collaborative filtering?” and I think the presenter offered a pretty good reasoning.
However, it sounds like pandora has more than a few vestigial dot-com tendencies, even though the presenter seemed to be intent on avoiding them. They are currently losing money, with a pretty good burn rate, and are mostly looking at ad revenue to solve the problem.
The service itself seems pretty expensive, as every new user adds significant bandwidth and licensing fees. I don’t really see them surviving for very long.
A few questions I didn’t asked but should of:
Would it be possible to include more freely distributed music (Creative Commons, etc) in an effort to lower licensing costs?
Do the song selection code take into account any sponsors or partners music? (aka, “Do you take payola?”)
Will there be any efforts to gather more demographic info to make advertisers happier?
Any plans to extend the concept beyond music? Visual Art? Movies? TV?
I hadn’t actually used pandora much after an initial experiment several months ago that didn’t work well (for the love of god, just because I like King Crimson doesn’t mean I want to hear E.L.P). But trying it out last night and today, and it’s been pretty remarkable so far.