Language models integrate prior knowledge and new contextual information in predictable ways, relying more on prior knowledge for familiar entities and being more easily persuaded by some contexts than others.
