Yanan Cao

LLM agent harnesses are surprisingly vulnerable, but weaving security directly into the agent lifecycle can slash attack success by 42% without sacrificing utility.

LLMs often invoke irrelevant tools just because the query structure *fits* the tool's parameters, revealing a surprising and widespread flaw in their reasoning.

LLMs can now perform entity alignment with greater interpretability and efficiency thanks to a new agent-based approach that structures the reasoning process.

Forget visit order: modeling item-specific purchase cadences with calendar time unlocks significant gains in next-basket repurchase prediction, even at production scale.

A novel ensemble method substantially improves the reliability of detecting Chinese LLM-generated text, even against adversarial examples.