Chi Zhang

Uncertainty-guided feedback can dramatically improve the reliability of reward models in visual diffusion, leading to superior optimization and quality outcomes.

Evolving agents can achieve up to a 19.37-point improvement in task performance by effectively leveraging prior experience in financial workflows.

LLMs can both combat misinformation and generate it, revealing a paradox that demands urgent research attention.

Despite advances in LLMs, they fail to effectively integrate user preferences over time, with accuracy rates stagnating around 39% even in ideal conditions.

Bridging the gaps in physical intelligence could enable agents to learn and adapt in real-world scenarios more effectively than ever before.

Grounding action representations in environmental context can drastically improve robotic manipulation performance, especially for complex tasks.

Existing text-to-image models struggle to capture individual aesthetic preferences, but PIPBench reveals critical gaps in their performance that could redefine personalized image generation.

Forget imitation: reward-aware trajectory shaping lets few-step generative models outperform their multi-step teachers.