This has been around since 1999 at least, but I just stumbled across it. Collaborative filtering, applied to a database of jokes.
Rate 15 jokes, and it serves you up ones it thinks you'll like.
The more you rate, the better it gets (although I don't know the size of the database.... I got some clunkers... either diminishing returns, or maybe the set of people who respond to a web site like this (created as a research project at Cal's School of Information Management and Systems).
in any case, you get some jokes out of it.