Curiosity, unobserved rewards and function Approximation: On recent progress in building solid foundations for RL