They call it "rapid reward evaluation via massively parallel reinforcement learning." The researchers describe Eureka as a "hybrid-gradient architecture," which essentially means that it is a ...