We expect methods like these are typically promising simply because language models now discover a good deal about human values all through pretraining. Learning about human values just isn't as opposed to learning about other subjects, and we should always hope larger models to possess a extra correct photograph of human values and to uncover them… Read More