gwern comments on Evidence Sets: Towards Inductive-Biases based Analysis of Prosaic AGI