Reasonable is an applied research company developing training paradigms for superhuman programming AI. We draw on on domain expertise in machine learning, formal verification and mathematical models of program semantics to create open-ended training environments, pre-training datasets, and post-trai