gwern comments on Computation Hazards