I don't think the paper addresses the question of self reflection. Like it can reflect on the question and answer pairs in its prompt but it didn't know that it created them in the first place or use that information to update it's priors, things humans can do