I have been thinking about this problem a lot recently and I think I might have a solution. Dissatisfaction with the state of social news sites seems almost universal, digg has almost unspeakably low quality, and reddit is slowly bleeding to death, and we all know that Hacker News has not been immune to the same disease of "lowest-common-denominator-ism." There must be a solution, certainly there's enough data sitting around on all the various services that meaningful inferences about user preferences can be drawn!
Enter my idea: http://con.nect.us/. It's currently under development (it's my senior project at my university) and will be finished by December. My advisor is Tim Davis (http://www.cise.ufl.edu/~davis/), the brilliant man who wrote many of MATLAB's sparse matrix libraries, so I have an excellent mind to tap to help me solve my linear algebra problems.
It goes like this: con.nect.us will track user activity across many different websites, linking your username on reddit or digg or wherever with your account on con.nect.us, as well as take submissions for news itself. It then analyzes what people like and how people are similar to each other and recommend new content accordingly. Hopefully this will help us avoid the curse of the least-common-denominator and start to experience the web better.
If you want to be informed of future developments, there's an email submission form on http://con.nect.us/
Also, the name isn't finalized, so if you think you have a better one, please send it my way.