“Value functions are expected to become relevant again in reinforcement learning in the future.”