Moir

New Moir method improves LLM knowledge editing by using model's own data

Researchers have developed a new method called Moir for knowledge editing in language models, addressing the issue of degraded reasoning capabilities after editing. Moir estimates the preservation covariance directly fr…