Course-Correction: Safety Alignment Using Synthetic Preferences
It addresses safety risks for users of LLMs by improving autonomous harm avoidance, though it is incremental as it builds on existing preference learning methods.
This paper tackles the problem of harmful content generation by large language models (LLMs) by assessing and improving their ability to autonomously steer away from such content, showing that fine-tuning with synthetic preferences enhances course-correction skills without affecting general performance and improves safety, particularly in resisting jailbreak attacks.
The risk of harmful content generated by large language models (LLMs) becomes a critical concern. This paper presents a systematic study on assessing and improving LLMs' capability to perform the task of \textbf{course-correction}, \ie, the model can steer away from generating harmful content autonomously. To start with, we introduce the \textsc{C$^2$-Eval} benchmark for quantitative assessment and analyze 10 popular LLMs, revealing varying proficiency of current safety-tuned LLMs in course-correction. To improve, we propose fine-tuning LLMs with preference learning, emphasizing the preference for timely course-correction. Using an automated pipeline, we create \textsc{C$^2$-Syn}, a synthetic dataset with 750K pairwise preferences, to teach models the concept of timely course-correction through data-driven preference learning. Experiments on 2 LLMs, \textsc{Llama2-Chat 7B} and \textsc{Qwen2 7B}, show that our method effectively enhances course-correction skills without affecting general performance. Additionally, it effectively improves LLMs' safety, particularly in resisting jailbreak attacks.