Exploring Notions of "Utility Approximation" and testing quantilizers — AI Alignment Forum