List your company

Defending against Gradient-based Attacks on Llama3 AutoDAN Attack (test)

10.57ASROurs-Fakecom-t
8.242823.951439.6655.3686Nov 1, 2024

Evaluation results

MethodLinks
Ours-Fakecom-tDefense=Fake Completion Defense with Template2024.1110.57
Ours-FakecomDefense=Fake Completion Defense2024.1114.9
Ours-IgnoreDefense=Ignore Defense2024.1116.34
SpotlightDefense=Spotlight defense baseline2024.1124.51
Ours-EscapeDefense=Escape Defense2024.1138.94
SandwichDefense=Sandwich defense baseline2024.1139.42
ReminderDefense=Reminder defense baseline2024.1151.44
InstructionalDefense=Instructional defense baseline2024.1152.88
IsolationDefense=Isolation defense baseline2024.1154.32
NoneDefense=No defense baseline2024.1168.75