The concept of Q learning is discussed along with another algorithm for reinforcement learning, both of which are off-policy and enable one to learn about the environment's value of different actions while figuring out how to behave optimally. The interesting idea of making predictions over time is also considered in relation to neural networks and backpropagation.