writing quality

Meta AI research on RL-XAR, a method for training language models toward expert-level writing using expert-aligned rubrics and reinforcement learning. The page explains the problem, the rubric-learning loop, and experiments on paper sections, story writing, and Wikipedia-style text.