John Schulman - Reinforcement Learning Without Full MDP Knowledge (via searchlore.ai)