Zilei Wang

Achieve state-of-the-art results in agentic knowledge base question answering by distilling gold-action policies into on-policy student rollouts, bridging the gap between sparse rewards and weakly supervised intermediate actions.

LLMs have consistent, situation-dependent personality "neurons" that can be steered without training, enabling more adaptable and human-like interactions.

VLMs still struggle to generate code for complex, multi-panel charts from real-world data, despite excelling on simpler benchmarks.