Explanation Faithfulness on Jailbreaking DAN (test)
24Label Change RateICL
Evaluation Results
| Method | Links | |
|---|---|---|
| ICLDeletion Ratio=10%2026.03 | 24 | |
| ICLDeletion Ratio=20%2026.03 | 27 | |
| ShapleyDeletion Ratio=10%2026.03 | 33 | |
| ShapleyDeletion Ratio=20%2026.03 | 33 | |
| LIMEDeletion Ratio=20%2026.03 | 38 | |
| LIMEDeletion Ratio=10%2026.03 | 54 | |
| EnsembleSHAPDeletion Ratio=20%2026.03 | 74 | |
| EnsembleSHAPDeletion Ratio=10%2026.03 | 85 |