The Bayesian formalism can add uncertainty to deep neural networks. But Bayesian deep learning has a reputation as cumbersome and expensive. No longer. Recent results show how to achieve calibrated uncertainty in deep networks efficiently, without affecting their predictive performance.