Uncovering Overfitting in Large Language Model Editing

TL;DR

This paper identifies a phenomenon called Editing Overfit in large language model editing, introduces a benchmark EVOKE to evaluate it, and proposes a new method LTI to mitigate overfitting and improve knowledge editing generalization.

Contribution

The paper uncovers Editing Overfit in LLM editing, introduces EVOKE benchmark and metrics, and proposes LTI strategy to reduce overfitting and enhance knowledge transfer.

Findings

Editing Overfit is common in current methods.

Existing mitigation strategies are ineffective against overfitting.

LTI significantly improves knowledge editing performance.

Abstract

Knowledge editing has been proposed as an effective method for updating and correcting the internal knowledge of Large Language Models (LLMs). However, existing editing methods often struggle with complex tasks, such as multi-hop reasoning. In this paper, we identify and investigate the phenomenon of Editing Overfit, where edited models assign disproportionately high probabilities to the edit target, hindering the generalization of new knowledge in complex scenarios. We attribute this issue to the current editing paradigm, which places excessive emphasis on the direct correspondence between the input prompt and the edit target for each edit sample. To further explore this issue, we introduce a new benchmark, EVOKE (EValuation of Editing Overfit in Knowledge Editing), along with fine-grained evaluation metrics. Through comprehensive experiments and analysis, we demonstrate that Editing Overfit is prevalent in current editing methods and that common overfitting mitigation strategies are ineffective in knowledge editing. To overcome this, inspired by LLMs' knowledge recall mechanisms, we propose a new plug-and-play strategy called Learn the Inference (LTI), which introduce a Multi-stage Inference Constraint module to guide the edited models in recalling new knowledge similarly to how unedited LLMs leverage knowledge through in-context learning. Extensive experimental results across a wide range of tasks validate the effectiveness of LTI in mitigating Editing Overfit.