[SEQ RERUN] Neural Categories — LessWrong