Mechanistic Circuit-Based Knowledge Editing in Large Language Models
arXiv:2604.05876v1 Announce Type: new
Abstract: Deploying Large Language Models (LLMs) in real-world dynamic environments raises the challenge of updating their pre-trained knowledge. While existing knowledge editing methods can reliably patch isolate…