Checklists Are Better Than Reward Models For Aligning Language Models

Vijay Viswanathan (Carnegie Mellon University) · Yanchao Sun (University of Maryland, College Park) · Xiang Kong (Apple) · Meng Cao (Mohamed bin Zayed University of Artificial Intelligence) · Graham Neubig (Carnegie Mellon University) · Sherry Wu (Carnegie Mellon University)

ai judgesalignment methodsarena-hardchecklist feedbackfollowbenchinfobenchinstruction followingllama 3.1 8b instructoff-policy learningolmo 2 7b instructqwen2.5-7b-instructreinforcement learningreward computationrubricsverifier programswildchecklists

Language models must be adapted to understand and follow user instructions. Reinforcement learning is widely used to facilitate this —typically using fixed criteria such as "helpfulness" and "harmfulness". In our work, we instead propose using flexible, instruction-specific criteria as a means of broadening the impact that reinforcement learning can have in eliciting instruction following. We propose "Reinforcement Learning from Checklist Feedback" (RLCF). From instructions, we extract checklists and evaluate how well responses satisfy each item—using both AI judges and specialized verifier programs—then combine these scores to compute rewards for RL. We compare RLCF with other alignment methods on top of a strong instruction following model (Qwen2.5-7B-Instruct) on five widely-studied benchmarks — RLCF is the only method to help on every benchmark, including a 4-point boost in hard satisfaction rate on FollowBench, a 6-point increase on InFoBench, and a 3-point rise in win rate on Arena-Hard. We show that RLCF can also be used off-policy to improve Llama 3.1 8B Instruct and OLMo 2 7B Instruct. These results establish rubrics as a key tool for improving language models' support of queries that express a multitude of needs. We release our our dataset of rubrics (WildChecklists), models, and code to the public.