Basically the only thing that comes to mind is that our collective views and knowledge about topics can change over time. If we had a hypothetical LLM in the 19

Basically the only thing that comes to mind is that our collective views and knowledge about topics can change over time. If we had a hypothetical LLM in the 1930's, it would have had quite different views on things like mental health, civil rights, and so on. But like you said, the models would ideally be trained on many other data sources as well.