Amorphous materials form a vast, yet vastly underutilized, class of materials. Computational simulations help in making some sense of their properties, but for the most part, the reach of computational methods remains limited. In the first part of the talk, I will explain how we use generative machine learning within the framework of the Morphological Autoregressive Protocol (MAP) to extend nanoscale samples into the mesoscale and to hunt for new physics in glassy systems. Two case studies that focus on amorphous graphene and on glass-forming liquids will be used to demonstrate MAP’s double-duty utility: 1) as a sampling algorithm to obtain sufficiently large samples for property prediction, and 2) as a characterization tool, for example, to diagnose dynamic heterogeneity in fragile glasses.