Demis Hassabis - Defining Reward Functions in RL (via searchlore.ai)