Model Editing Harms General Abilities of Large Language Models: Regularization to the Rescue

TL;DR

The paper investigates whether contemporary model editing to improve factuality harms LLM general abilities. It systematically evaluates four editing methods across three LLMs on eight tasks, revealing consistent degradation of reasoning, QA, and NER performance as edits accumulate. The authors identify overfitting to edited facts as a key cause and introduce RECT, a regularization scheme that preserves the most impactful weight changes while pruning rest, drastically reducing side effects. RECT maintains the majority of editing efficacy (over 94%) and better preserves downstream capabilities, suggesting a viable path toward trustworthy, robust model editing. The work highlights the need for balanced editing strategies that improve factuality without sacrificing general abilities and encourages further research into robust evaluation and defense against weight perturbations.

Abstract

Model editing is a technique that edits the large language models (LLMs) with updated knowledge to alleviate hallucinations without resource-intensive retraining. While current model editing methods can effectively modify a model's behavior within a specific area of interest, they often overlook the potential unintended side effects on the general abilities of LLMs such as reasoning, natural language inference, and question answering. In this paper, we raise concerns that model editing's improvements on factuality may come at the cost of a significant degradation of the model's general abilities. We systematically analyze the side effects by evaluating four popular editing methods on three LLMs across eight representative tasks. Our extensive empirical experiments show that it is challenging for current editing methods to simultaneously improve factuality of LLMs and maintain their general abilities. Our analysis reveals that the side effects are caused by model editing altering the original model weights excessively, leading to overfitting to the edited facts. To mitigate this, a method named RECT is proposed to regularize the edit update weights by imposing constraints on their complexity based on the RElative Change in weighT. Evaluation results show that RECT can significantly mitigate the side effects of editing while still maintaining over 94% editing performance.

Figures (18)

• Figure 1: Demonstration of model editing and its impact on the general abilities of LLMs. Although the factuality of the model has been improved, the general abilities of LLMs, such as question answering, dialogue, named entity recognition, sentiment analysis, have been substantially impaired after editing. $f_{\theta}$ / $f_{\theta_{e}}$ denotes the models before / after editing.
• Figure 2: Illustration of the settings of (a) single- and instance-editing, (b) sequential- and instance-editing, and (c) sequential- and batch-editing. The darker units correspond to more edits.
• Figure 3: Performance on general tasks of edited models using KN or ROME to edit GPT-2 XL or LLaMA-1 (7B) as the number of edits increases in instance- and sequential-editing.
• Figure 4: Performance on general tasks of edited models using MEND or MEMIT to edit GPT-2 XL or LLaMA-1 (7B) with different batch sizes in batch- and single-editing.
• Figure 5: Visualization of the distinction between the final edited weight and the original unedited weight via weight change $|\Delta W|$ as the number of edits increases.