We predict ways like they are promising for the reason that language models by now find out a lot about human values all through pretraining. Learning about human values will not be in contrast to learning about other topics, and we should always hope larger models to have a additional precise photo of human values and to locate them simpler to dis