Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

PAIR

Benchmarks

Task NameDataset NameSOTA ResultTrend
Jailbreak DefensePAIR
ASR0
97
Jailbreak AttackPAIR
Harmful Score0
46
Jailbreak DetectionPAIR
Accuracy98
30
Jailbreak Attack DefensePAIR
ASR1
24
Harmfulness EvaluationPAIR
Harmfulness Score1.08
22
Adversarial RobustnessPAIR
ASR26
18
Interaction and Contact SegmentationPaIR-1
mAP38.07
12
Contact Region SegmentationPaIR-2
SC Accuracy24.01
11
Interaction DetectionPaIR-2
mAP61.32
11
Adaptive Jailbreak Attack Success RatePAIR behaviors held-out (test)
ASR100
9
Cell SegmentationPair 5 (test)
SEG Score0.76
9
Cell SegmentationPair 4 (test)
Segmentation Score90
9
Cell SegmentationPair 3 (test)
SEG Score0.81
9
Cell SegmentationPair 1 (test)
SEG62
9
Showing 14 of 14 rows