Eval-Skill

Eval-Skill method boosts LLM reward modeling with reusable skills

Researchers have developed a new method called Eval-Skill for improving reward modeling in large language models. This approach synthesizes reusable evaluation skills, which are then injected into the model's context, r…