Meng Chen

Models may give clearer instructions, but they fail to engage learners deeply, resulting in passive instruction-following rather than active understanding.

LALMs are shockingly vulnerable to inaudible audio prompts that can make them execute unauthorized actions, even on commercial systems like Mistral AI and Microsoft Azure.

Achieve significantly better vocal-to-accompaniment generation by reformulating it as a discrete masked diffusion process, finally balancing acoustic fidelity, coherence, and orchestration.