Energy-Based Reward Models for Robust Language Model Alignment

Anamika Lochab, Ruqi Zhang

Reward models (RMs) are essential for aligning Large Language Models (LLMs) with human preferences. However, they often struggle with capturing complex human preferences and genera…