Dialogue Generation on UltraChat
98.7ASR AccuracyNo Def.
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| No Def.Attack=BadEdit2026.04 | 98.7 | 90.8 | |
| No Def.Attack=EmbedX2026.04 | 96.9 | 91 | |
| No Def.Attack=VPI2026.04 | 94.2 | 91.5 | |
| No Def.Attack=BadChain2026.04 | 92.8 | 90.5 | |
| DemoDefenseAttack=EmbedX2026.04 | 19.2 | 88.1 | |
| TIGSAttack=EmbedX2026.04 | 18.7 | 89 | |
| FABEAttack=EmbedX2026.04 | 18.4 | 85 | |
| LLMBDAttack=EmbedX2026.04 | 16.9 | 82.8 | |
| CROWAttack=EmbedX2026.04 | 16.2 | 88.9 | |
| BEEARAttack=EmbedX2026.04 | 15.5 | 87.8 | |
| W2SDefenseAttack=EmbedX2026.04 | 14.8 | 89.5 | |
| FABEAttack=BadEdit2026.04 | 14.5 | 85.2 | |
| LLMBDAttack=BadEdit2026.04 | 12.8 | 83.1 | |
| FABEAttack=VPI2026.04 | 12.1 | 84.8 | |
| DemoDefenseAttack=BadEdit2026.04 | 11.5 | 88.5 | |
| FABEAttack=BadChain2026.04 | 11.5 | 84.1 | |
| DemoDefenseAttack=BadChain2026.04 | 11.2 | 86.5 | |
| LLMBDAttack=VPI2026.04 | 10.6 | 83.4 | |
| DemoDefenseAttack=VPI2026.04 | 9.5 | 89.8 | |
| LLMBDAttack=BadChain2026.04 | 9.4 | 87.5 | |
| TIGSAttack=VPI2026.04 | 8.4 | 89.2 | |
| TIGSAttack=BadEdit2026.04 | 6.5 | 90.3 | |
| TIGSAttack=BadChain2026.04 | 6.5 | 89.5 | |
| CROWAttack=BadEdit2026.04 | 5.6 | 88.7 | |
| CROWAttack=VPI2026.04 | 5.1 | 88.5 | |
| BEEARAttack=VPI2026.04 | 4.8 | 87.1 | |
| W2SDefenseAttack=BadEdit2026.04 | 4.5 | 89 | |
| CROWAttack=BadChain2026.04 | 4.5 | 87.9 | |
| BEEARAttack=BadEdit2026.04 | 4.1 | 87.5 | |
| BEEARAttack=BadChain2026.04 | 3.8 | 87 | |
| W2SDefenseAttack=VPI2026.04 | 3.6 | 88.7 | |
| W2SDefenseAttack=BadChain2026.04 | 3.1 | 88.2 |