:Compared to the light-tailed case, which has a long and successful history, the heavy-tailed case has been relatively underdeveloped. However, the past few years have seen significant progress in heavy-tailed large deviations theory. In this talk, I will discuss implications of the recent advances in heavy-tailed theory in rare-event simulation, queueing theory, and artificial intelligence. In particular, we will explore an intriguing connection between the catastrophe principle and a central mystery of modern AI—the unreasonably good generalization performance of deep neural networks. Zoom: 937 905 8748