Weijie Su, a statistics professor at Wharton who won the prestigious COPSS Presidents' Award in 2026, just dropped a double whammy on X: he's joining OpenAI on sabbatical from Wharton and has been promoted to full professor.
https://twitter.com/weijie444/status/2060604060362014803
Su, who earned his PhD from Stanford and his undergraduate degree from Peking University's School of Mathematical Sciences, says he's heading back to the Bay Area after a decade to work on AI model training.
The COPSS Presidents' Award is basically the Nobel of statistics. The committee praised Su for building the statistical foundations of generative AI — including watermarking LLMs, aligning outputs with human preferences, and developing ranking theory. They also highlighted his work on differential privacy methods for the 2020 US Census, plus contributions to deep learning theory, convex optimization, and high-dimensional inference.
On the research front, Su has been busy reshaping the statistical underpinnings of large language models. For watermark detection, he framed the problem as a statistical hypothesis test and proposed the Truncated Goodness-of-Fit (Tr-GoF) method. It improves robustness against edits and trims without needing prior knowledge of human editing.
In the realm of reinforcement learning from human feedback (RLHF), Su identified a problem he calls "Preference Collapse" — the optimization process tends to ignore minority preferences and flatten the reward distribution across prompts. To maintain diversity, his team derived a "Preference Matching" regularization scheme using ordinary differential equations. They're also exploring a game-theoretic alternative called Nash Learning from Human Feedback (NLHF) to replace the traditional scalar reward alignment.
Su is part of what Chinese media call the "Golden Second Generation" (or "Platinum Generation") of math talents from Peking University. Unlike the earlier "Golden Generation" (class of 2000, which included pure mathematicians like Yun Zhiwei, Zhang Wei, and Xu Chenyang), Su's cohort is known for crossing disciplinary boundaries — pushing math into statistics, data science, and AI.