List your company

Exaggerated Safety Mitigation on Exaggerated Safety

100NRRSAC Multi-task
59.4469.9780.591.03Nov 4, 2024

Evaluation results

MethodLinks
SAC Multi-taskBackbone=Qwen2-72B-Instruct2024.11100
SAC Multi-taskBackbone=Qwen2-7B-Instruct2024.1199.5
SAC Single-taskBackbone=Qwen2-72B-Instruct2024.1199.5
SAC Single-taskBackbone=Qwen2-7B-Instruct2024.1196
SACControl Dim=Single2024.1196
RepEControl Dim=Single2024.1195
No ControlBackbone=Qwen2-72B-Instruct2024.1190.5
SACControl Dim=Multiple2024.1188.5
No ControlBackbone=Qwen2-7B-Instruct2024.1188
No ControlControl Dim=Single2024.1167
No ControlControl Dim=Multiple2024.1167
RepE-MergeControl Dim=Multiple2024.1163
RepE-MeanControl Dim=Multiple2024.1161