Refined Direct Preference Optimization with Synthetic Data for Behavioral Alignment of LLMs
This addresses the problem of aligning LLM behavior efficiently for developers and users, though it appears incremental as it builds on existing DPO methods.
The paper tackles behavioral alignment of Large Language Models by introducing refined Direct Preference Optimization (rDPO), which uses synthetic data generated via self-critique prompting and an external reward model to improve safety, robustness, and reduce sycophancy without human annotations.
In this paper, we introduce \emph{refined Direct Preference Optimization} (rDPO), a method for improving the behavioral alignment of Large Language Models (LLMs) without the need for human-annotated data. The method involves creating synthetic data using self-critique prompting by a teacher LLM and then utilising a generalized DPO loss function to distil to a student LLM. The loss function incorporates an additional external reward model to improve the quality of synthetic data, making rDPO robust to potential noise in the synthetic dataset. rDPO is shown to be effective in a diverse set of behavioural alignment tasks, such as improved safety, robustness against role-playing, and reduced sycophancy. Code to be released at https://github.com/vicgalle/refined-dpo.