Xiaocheng Feng

Current coding agents falter in preserving content integrity while reconstructing UI regions, revealing critical gaps in their iterative coding capabilities.

H$^2$SD achieves superior reasoning performance by intelligently adapting teacher signals based on trajectory outcomes, leading to more effective learning in large language models.

Even top-tier LLMs falter in open-ended cultural reasoning, revealing significant performance disparities across regions and underscoring the limitations of knowledge-centric evaluations.

Steer LVLMs' attention with caption guidance and watch object hallucinations drop by 6%—no training required.