X. Li

Yingru (Richard) Li is a reinforcement learning researcher and Member of Technical Staff at xAI (SpaceX Super Intelligence), where he contributes to RL fundamentals for large-scale Grok model training. He previously led RL training stability and efficiency research for LLM reasoning and agents at ByteDance. He is best known for diagnosing LLM-RL training collapse as an off-policy bias from the training-inference mismatch and developing Trust Region Masking and Masked Importance Sampling as fixes, as well as authoring the Rollout Correction framework in the veRL open-source RL infrastructure.