Privately Aligning Language Models with Reinforcement Learning

Fan Wu, Huseyin A. Inan, Arturs Backurs +3

Positioned between pre-training and user deployment, aligning large language models (LLMs) through reinforcement learning (RL) has emerged as a prevailing strategy for training ins…