If you have a specific topic in mind, find out the top conference(s) in the area (for many areas in CS the top resources tend to be conferences, for other topics it would be journals), and read all of the abstracts. You'll get a feeling on what is considered 'bleeding edge'.

From that start point - just 'crawl' for (a) interesting papers cited there, and (b) other things written by authors you found interesting, they'll often have the non-paywall versions of their papers available on their site.