Tag
This paper introduces FakeContext-bench to evaluate how well large language models distinguish between contextual information and factual knowledge, and proposes Jurisdiction In-Context Learning (J-ICL) to enhance both in-context learning performance and resistance to misleading context.