"TD: We want to build machine programs that can learn things from information that's out there on the web. In the first application we'll come out with, you tel

"TD: We want to build machine programs that can learn things from information that's out there on the web. In the first application we'll come out with, you tell us things that you're interested in, and we'll continuously go out and find stuff on the internet that's related to that. There's a positive feedback loop where you tell us what you like and don't, so the machine gets progressively better in learning what you like." That's extremely ambitious. I've seen so many implementations of what he's talking about and the results have always been abysmal. Biggest problem for these systems is that you need a lot (and I mean a LOT) of training data to start making predictions/recommendations that make any sense. Their scope also seems way too broad.