Evidence Sets: Towards Inductive-Biases based Analysis of Prosaic AGI — LessWrong