Lifelong model editing improves knowledge updates in large language models
Generalizable Lifelong Model Editing via Preference Optimization
Artificial Intelligence
Summary
Large language models sometimes need to update their knowledge quickly, but changing facts over time without retraining the entire model is hard. Existing methods often only work well for one specific question and can forget what they've learned before. The authors developed a method called GLIME that lets models update their knowledge continuously while keeping old updates and general skills intact. GLIME uses techniques to guide the model's behavior so changes apply broadly, not just to one prompt.
What this means in practice
- •For ai service engineers: continuously update language models with new factual knowledge without degrading overall model performance
- •For chatbot developers: maintain up-to-date responses in conversational AI by editing knowledge repeatedly without needing full retraining
Authors
Dahyun Jung, Suhyune Son, Heuiseok Lim
Abstract
Knowledge editing enables rapid updates of specific factual knowledge in large language models (LLMs) without full retraining. However, more realistic scenarios call for a lifelong framework that handles continual updates rather than one-off modifications. In such settings, existing editing methods often overfit to target prompts, significantly degrading both the generalization of the edited knowledge and the model's general capabilities. To address this issue, we propose GLIME (Generalizable Lifelong Model Editing), which combines knowledge editing with preference optimization over generation behavior. GLIME further incorporates replay-based editing and a gradient constraint to preserve previously edited knowledge. Experimental results show that GLIME significantly improves knowledge generalization in lifelong editing settings while maintaining both editing performance and general capabilities.