Most likely move, shaded by how likely (max π)

Episode 1
less → more likely this episode's path

θ ← θ + α Σt [Gt − b] ∇θ log π(at | st)