Study Examines How Reward Gain Imbalance Shapes Rubric-Based Ranking in RLHF
A new arXiv paper rethinks rubric-based reward aggregation in reinforcement learning, showing that equally weighting rubric criteria can misalign LLM rewards with human preferences due to reward gain imbalance.
Read the full story at arXiv stat.ML ↗
Timeline · 1 report
- 2026-11-19 05:00 · arXiv stat.ML
How Reward Gain Imbalance Shapes Ranking: Rethinking Rubric-Based Reward Aggregation