Hacker News Reader

They should add a "were we right?" box at the bottom, I bet the results would be pretty bad. (For me, not a single thing they assume is right)

They could train the algorithm that way too.

That's how Hunch worked and it was pretty good.