Quantilizers and Generative Models — LessWrong