Researchers propose Hybrid-Policy Self-Editing, a method for updating large language models with unstructured knowledge edits using free-form passages containing multiple facts. The approach addresses limitations of existing knowledge editing techniques by enabling composable updates without altering unrelated information, though the full technical details are not provided in the abstract. The paper was authored by Tianci Liu, Zihan Dong, Tianchun Li, Yi-Chung Chen, and Qiming Cao and published on August 11, 2026.