1. The biggest problem I say is its entirely closed nature. I don't mean simply closed source -- I've used tools such as Mediamonkey for Windows and have been relatively satisfied -- but entirely closed to plugins or user developed support. Specifically I refer to the inability to use/add anything to use FLAC, lyrics search, grabbing high quality album art ( a la allcdcovers.com ), social integration (mixtape, last.fm), 3rd party hardware syncing, and library visualization. Each of these could be solvable if only Apple allowed some sort of plugin integration
Though the per-song cost for Pandora is very hard on their business model, is their support of the same for traditional radio anything but hypocritical?
Oh, I'm sure the music industry loves that U.S. terrestrial radio doesn't pay performance fees. I bet they also love it when we do their promotion for them by sharing iTunes playlists with our dorm/neighborhood/company.
Look, I get that the internet is TEH FUTURE and all the old distribution channels are dumb and old. I get it. But there are reasons for why things are the way they are. And to say that the record labels are in cahoots with the radio stations is insane. The record labels hate that, by an accident of history, U.S. terrestrial radio doesn't pay performance fees (they do pay mechanical royalties to songwriters). More: http://news.ycombinator.com/item?id=698651
Yes, record labels hate the fact that they're tethered to Clear Channel in so many ways (not only do they own most of the major radio stations, mind you, they also own most of the large concert venues and billboards). I've been sort of flabbergasted that they haven't been more behind internet radio and the like - it's a way for them to get out from under the thumb of Clear Channel.
As a student working in AI, I have to say that I both agree and disagree with this writer. I agree with his assertion that AI, in its current state, is not nearly as generalizable as we hope it to be, but simply because it is not now doesn't mean we should give up on it entirely and wait for some breakthrough to come in from another field. Better hardware has been a great boon to AI researchers, but it's in times like these that we need not only to verify or dismiss the mathematical theories put forth 50 odd years ago but enhance them with experimentation.
For example, perhaps descriptors are truly the wrong way to go about object recognition in images, or perhaps segmentation is the wrong way to look at a scene, but they are ideas and they are worth testing. We won't find out unless we try, because unlike the equations proven decades before, there are no constraints we can simply apply.
Though interesting to watch, this is undoubtedly the same method that has been implemented in surgical robotics for years and is little more than a series of go-to-point operations. The fact that both the slave and the masters are of the same general design even makes inverse kinematics unnecessary.
2. See suggestions above
3. Not using mac.