Some considerations on learning to explore via meta-reinforcement learning