The impact of ab initio calculations on materials research has been tremendous, providing deep understanding and reliable predictions of materials properties. It would be wonderful if such calculations were much faster and applicable to large systems allowing us to simulate dynamic processes involving the sampling of millions of configurations of complex systems. The emergence of machine-learned potentials (MLPs) is profoundly changing this situation. MLPs combine the accuracy of ab initio calculations with the computational efficiency of interatomic potentials, thus enabling the simulation of phenomena such as plastic deformations, diffusion, and phase transformations.