This story has moved to /story/bias-fitting-to-mitigate-length-bias-of-reward-model-in-rlhf/.