Stealth edits to large language models

Oliver J. Sutton, Qinghua Zhou, Wei Wang +4

We reveal the theoretical foundations of techniques for editing large language models, and present new methods which can do so without requiring retraining. Our theoretical insight…