Yue Zhu

Ditching modular architectures unlocks surprisingly competitive vision-language performance, proving that end-to-end pixel-to-word models can rival traditional approaches at scale.

Even state-of-the-art AI agents are surprisingly vulnerable to incremental attacks that gradually lead them to perform unsafe actions in realistic workplace scenarios.

Transforming sparse driving log observations into complete 3D assets could revolutionize simulation fidelity in autonomous vehicle development.