I'm going to echo CatDancer: for me the most valuable insight was that a little information goes a very long way. From the example of the simulated beings breaking out to the Bayescraft interludes to the few observations and lots of cogitations in Three Worlds Collide to GuySrinivasan's random-walk point, I've become more convinced that you can get a surprising amount of utility out of a little data; this changes other beliefs like my assessment of how possible AI rapid takeoff is.