Buck comments on Let’s talk about “Convergent Rationality”

[Buck]

In OpenAI’s Roboschool blog post:

This policy itself is still a multilayer perceptron, which has no internal state, so we believe that in some cases the agent uses its arms to store information.