Climbing the Horseshoe — LessWrong