Question Answering on MMLU (TSR, ASR)
88.25Task Success Rate (TSR)STAR
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| STARAgent Model=Qwen-Plus, Attack Type=Group Collusive Attack2026.05 | 88.25 | 7.5 | |
| G-SafeguardAgent Model=Qwen-Plus, Attack Type=Group Collusive Attack2026.05 | 84.5 | 10.25 | |
| STARAgent Model=GPT-4-Turbo, Attack Type=Group Collusive Attack2026.05 | 80.5 | 12.75 | |
| STARAgent Model=GPT-3.5-Turbo, Attack Type=Group Collusive Attack2026.05 | 77.5 | 12 | |
| G-SafeguardAgent Model=GPT-4-Turbo, Attack Type=Group Collusive Attack2026.05 | 68.75 | 21.75 | |
| G-SafeguardAgent Model=GPT-3.5-Turbo, Attack Type=Group Collusive Attack2026.05 | 58.5 | 35.5 |