Adversarial Representation Engineering: A General Model Editing Framework for Large Language Models
This work addresses the problem of general and flexible model editing for LLMs, which is incremental as it builds on representation engineering insights.
The paper tackles the challenge of editing large language models (LLMs) by proposing an Adversarial Representation Engineering (ARE) framework that uses a representation sensor as an editing oracle, achieving effectiveness in multiple tasks without compromising baseline performance.
Since the rapid development of Large Language Models (LLMs) has achieved remarkable success, understanding and rectifying their internal complex mechanisms has become an urgent issue. Recent research has attempted to interpret their behaviors through the lens of inner representation. However, developing practical and efficient methods for applying these representations for general and flexible model editing remains challenging. In this work, we explore how to leverage insights from representation engineering to guide the editing of LLMs by deploying a representation sensor as an editing oracle. We first identify the importance of a robust and reliable sensor during editing, then propose an Adversarial Representation Engineering (ARE) framework to provide a unified and interpretable approach for conceptual model editing without compromising baseline performance. Experiments on multiple tasks demonstrate the effectiveness of ARE in various model editing scenarios. Our code and data are available at https://github.com/Zhang-Yihao/Adversarial-Representation-Engineering.