Improving Atomic-Fact Recall via Focused Views in Unstructured Knowledge Editing

Large language models (LLMs) increasingly serve as general-purpose interfaces to factual knowledge, but their parameters do not automatically reflect information that changes after pretraining.