What an amazing output. Reading his papers, he manages to connect basic geometric ideas (like the notion of "isoperimetry", i.e., enclosing the maximum volume

What an amazing output. Reading his papers, he manages to connect basic geometric ideas (like the notion of "isoperimetry", i.e., enclosing the maximum volume within a given perimeter) to probabilistic notions like convergence of averages. It links together stuff you learn in information theory (the probability mass of the "typical set"), and in machine learning (deviations of an training-set error rate from a true error rate), and in probability theory (empirical processes). His papers typically had an introduction that related the main theorem to some of these basic geometric notions. The introduction would reveal a whole new, very intuitive and geometric connection between a very abstract theorem to basic geometry, like the volume of a spherical shell. It would routinely blow my mind.