Wenjie Li

Despite the advancements in multimodal agents, even the best models struggle with interactive spatial reasoning, achieving only a 17.4% success rate in complex real-world tasks.

SkeMex enables medical agents to evolve their reasoning capabilities by transforming raw experience into structured, reusable skills, outperforming traditional memory systems.

Images can serve as a powerful standalone medium for reasoning, achieving nearly double the token efficiency of traditional text methods.

Co-evolving world models with agent policies leads to a 16.75% boost in performance, revolutionizing how language agents navigate complex tasks.

Unsupervised point cloud denoising can achieve state-of-the-art results by learning to create and enforce consistency between "mirror points" reflecting the underlying surface geometry.

LLMs struggle to grasp the nuances of cross-cultural aesthetic stylistics, often mistaking surface-level linguistic features for genuine cultural understanding.

SiPeR reveals how integrating scene dynamics with Bayesian inference can dramatically enhance the relevance of conversational recommendations in real-world contexts.

LLMs are still far from being able to generate expert-level clinical guidelines, despite advances in deep research systems.

Active belief intervention can drastically improve embodied agents' decision-making by overcoming the pitfalls of belief inertia.

LLMs can learn to anticipate their opponents' moves and make better decisions in strategic games by explicitly modeling the other player's behavior during training.

Tool calling gets a 4x speed boost without training by exploiting structured schemas and retrieval of past invocations.

LLMs can reason more effectively and efficiently by internalizing tool knowledge, eliminating the need for external documentation and reducing inference costs.