Ball Sq Pathways — LessWrong