Richard Cameron's CiteULike, previously mentioned on this here blog, has been piling up new connections and new features. They can be read about on the developer's weblog. CiteULike is now tied into a number of archives, including arXiv and CiteSeer. A group mechanism has been added. For example, a couple of folks at Northeastern have been busy little programming language squirrels. There's also a weighted list view of users tag and author collections.
CiteULike points out how restricting the document domain helps in automatically generating semantic metadata. Every paper has authors, a publishing venue, publication date, etc. which can immediately become tags or taglike.
Best of all everything seems to spit out RSS. I wonder how much of it is captured by PubSub?