Editing Conceptual Knowledge for Large Language Models
This work addresses the challenge of modifying conceptual knowledge in LLMs, which is crucial for improving model reliability and understanding, though it is incremental as it builds on existing instance-level editing approaches.
This paper tackled the problem of editing conceptual knowledge in Large Language Models (LLMs) by constructing a benchmark dataset called ConceptEdit and establishing new evaluation metrics. The results showed that existing editing methods can modify concept-level definitions but often distort related instantial knowledge, leading to poor performance.
Recently, there has been a growing interest in knowledge editing for Large Language Models (LLMs). Current approaches and evaluations merely explore the instance-level editing, while whether LLMs possess the capability to modify concepts remains unclear. This paper pioneers the investigation of editing conceptual knowledge for LLMs, by constructing a novel benchmark dataset ConceptEdit and establishing a suite of new metrics for evaluation. The experimental results reveal that, although existing editing methods can efficiently modify concept-level definition to some extent, they also have the potential to distort the related instantial knowledge in LLMs, leading to poor performance. We anticipate this can inspire further progress in better understanding LLMs. Our project homepage is available at https://zjunlp.github.io/project/ConceptEdit.