Search papers, labs, and topics across Lattice.
1
0
3
5
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.