In this post, Jeff Kaufman asks for examples of superintelligence risk that are more realistic than the ones he's seen. Eliezer's reply may be of particular interest. I also wrote a reply.