ResearchBenchmarksJailbreak Attack on AdvBench Vicuna to gpt-3.5-turbo (transfer)Follow90ASR (Refusal Matching)DSN32.07247.11162.1577.189Apr 25, 2024Evaluation ResultsMethodMethodLinksASR (Refusal Matching)ASR (Ensemble Eval)DSNMetric Aggregation=max...Metric Aggregation=max, Source Model=Vicuna, Target Model=gpt-3.5-turbo2024.049093DSNMetric Aggregation=mea...Metric Aggregation=mean, Source Model=Vicuna, Target Model=gpt-3.5-turbo2024.0454.2759.59GCGSource Model=Vicuna, T...Source Model=Vicuna, Target Model=gpt-3.5-turbo2024.0434.3—