This paper explores the editing of conceptual knowledge in Large Language Models (LLMs), a novel and underexplored area compared to instance-level editing. The authors introduce ConceptEdit, a benchmark dataset constructed based on the DBpedia ontology, and develop new evaluation metrics, Instance Change and Concept Consistency, to assess the effectiveness of current editing methods. Experiments with four prominent LLMs (GPT-J, GPT2-XL, LLaMA-2-7B-Chat, and Mistral-7B-v0.1) reveal that while existing methods can modify concept-level definitions, they also introduce distortions that affect related instance-level knowledge. The study highlights the need for more sophisticated techniques to accurately edit conceptual knowledge in LLMs and provides insights into the mechanisms by which LLMs store and manage conceptual knowledge. The findings suggest that stronger techniques and a deeper understanding of concepts are required to achieve reliable and accurate concept editing in LLMs.This paper explores the editing of conceptual knowledge in Large Language Models (LLMs), a novel and underexplored area compared to instance-level editing. The authors introduce ConceptEdit, a benchmark dataset constructed based on the DBpedia ontology, and develop new evaluation metrics, Instance Change and Concept Consistency, to assess the effectiveness of current editing methods. Experiments with four prominent LLMs (GPT-J, GPT2-XL, LLaMA-2-7B-Chat, and Mistral-7B-v0.1) reveal that while existing methods can modify concept-level definitions, they also introduce distortions that affect related instance-level knowledge. The study highlights the need for more sophisticated techniques to accurately edit conceptual knowledge in LLMs and provides insights into the mechanisms by which LLMs store and manage conceptual knowledge. The findings suggest that stronger techniques and a deeper understanding of concepts are required to achieve reliable and accurate concept editing in LLMs.