How and Why to Granularize — LessWrong