Knowledge, Reasoning, and Superintelligence — LessWrong