Yu Jiang

The paper introduces GrantBox, a new security sandbox that evaluates how well LLM agents handle real-world tool privileges, finding that agents remain highly vulnerable to sophisticated attacks.

VCap introduces a novel Witness-Adjudicator reward mechanism that provides highly precise, factually grounded feedback for visual captioning, enabling state-of-the-art performance in RL-trained multimodal models.

The paper demonstrates that the internal signals governing figurative language generation are reusable across multiple languages, showing that a steering direction learned in one language can effectively enhance generation in another.

The paper introduces DEBENCH, a novel framework that evaluates binary decompilers based on three orthogonal dimensions—readability, recompilability, and functionality—revealing that functional recovery is significantly harder than simple code readability.