Adversarial Representation Engineering: A General Model Editing Framework for Large Language Models

2024/04/21 by Yihao Zhang, Zhang, Yihao, Zeming Wei +5 · 4 citations
Computer Science · Decision Sciences · #Adversarial Robustness in Machine Learning #Artificial Intelligence (cs.AI) #Computation and Language (cs.CL) #Cryptography and Security (cs.CR) #FOS: Computer and information sciences #FOS: Mathematics #Machine Learning (cs.LG) #Optimization and Control (math.OC) #Scientific Computing and Data Management

paper · pdf · doi:10.48550/arxiv.2404.13752

openalex publication_date 2024/04/21 · openalex created_date 2024/04/25 · openalex updated_date 2026/07/28

Abstract

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.

Cited by

• Sysformer: Safeguarding Frozen Large Language Models with Adaptive System Prompts
• Shaking to Reveal: Perturbation-Based Detection of LLM Hallucinations
• Emergent Standing Wave Dynamics and Attractor Basins in Transformer Latent Spaces via Prompt Driven Constraints_Anathema to Corporate Control by Ingrid Johnson
• Advancing LLM Safe Alignment with Safety Representation Ranking

Related

• Representation Engineering for Large-Language Models: Survey and Research Challenges
• Editing Large Language Models: Problems, Methods, and Opportunities
• Taxonomy, Opportunities, and Challenges of Representation Engineering for Large Language Models
• On the Robustness of Editing Large Language Models
• Flexible Model Interpretability through Natural Language Model Editing