Learning representations - History — LessWrong