Skip to yearly menu bar Skip to main content


Privately Aligning Language Models with Reinforcement Learning

Fan Wu · Huseyin Inan · Arturs Backurs · Varun Chandrasekaran · Janardhan Kulkarni · Robert Sim

Halle B #213
[ ]
Thu 9 May 7:30 a.m. PDT — 9:30 a.m. PDT


Positioned between pre-training and user deployment, aligning large language models (LLMs) through reinforcement learning (RL) has emerged as a prevailing strategy for training instruction following-models such as ChatGPT. In this work, we initiate the study of privacy-preserving alignment of LLMs through Differential Privacy (DP) in conjunction with RL. Following the influential work of Ziegler et al. (2020), we study two dominant paradigms: (i) alignment via RL without human in the loop (e.g., positive review generation) and (ii) alignment via RL from human feedback (RLHF) (e.g., summarization in a human-preferred way). We give a new DP framework to achieve alignment via RL, and prove its correctness. Our experimental results validate the effectiveness of our approach, offering competitive utility while ensuring strong privacy protections.

Chat is not available.