A Stanford researcher questions the need for generalizable models and proposes instead sharing recipes for creating useful local models.