Aligning LLMs with Domain Invariant Reward Models

David Wu, Sanjiban Choudhury

Aligning large language models (LLMs) to human preferences is challenging in domains where preference data is unavailable. We address the problem of learning reward models for such…