Search papers, labs, and topics across Lattice.
Qilu University of Technology (Shandong Academy of Science)
3
0
7
0
Backdoor attacks can generalize across trigger families, with Lilith achieving high success rates while preserving benign model performance.
Forget jailbreaking with surface tokens – this new backdoor method steers internal representations for persistent, stealthy attacks that are much harder to detect.
VLMs can't tell a joke, but targeted prompting and model tuning can make them 16.5% funnier.
