Jennifer Rexford's talk about software-defined networks and openflow and stuff. I already knew a good fraction of this material from my Penn stint, but it was a extremely well-packaged and -delivered talk; an excellent advertisement for an interesting research area.
Robby Findler's "Run Your Research" talk about Redex and how to reduce errors in research papers by more testing and deeper integration of mechanized reasoning systems with every other part of the research-paper-writing process. Another super-polished and awesome talk, full of casually showing off crazy futuristic IDE awesomeness that I didn't realize was in Racket these days.