Search papers, labs, and topics across Lattice.
Washington University in St. Louis
1
0
2
Adversarial attacks on LLMs can be dramatically sped up and made more effective by exploiting the surprisingly low-rank structure of adversarial perturbations.
