Reliability Estimation on TruthfulQA
80.78AUROCDisAAD
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| DisAADTarget Model=GPT-4, Setting=Black-box, O(1) Complexity=single-sample2026.05 | 80.78 | 90.79 | |
| DisAADTarget Model=LLaMA2-70B, Setting=Black-box, O(1) Complexity=single-sample2026.05 | 80.15 | 78.07 | |
| DisAADTarget Model=LLaMA2-70B, Setting=White-box, O(1) Complexity=single-sample2026.05 | 80.15 | 78.07 | |
| DisAADTarget Model=Qwen3-32B, Setting=White-box, O(1) Complexity=single-sample2026.05 | 74.78 | 83.1 | |
| LogTokUTarget Model=LLaMA2-70B, Setting=White-box, O(1) Complexity=single-sample2026.05 | 72.17 | 68.74 | |
| EntropyTarget Model=LLaMA2-70B, Setting=White-box, O(1) Complexity=single-sample2026.05 | 67.32 | 61.21 | |
| EigVTarget Model=LLaMA2-70B, Setting=Black-box, O(1) Complexity=multi-sample2026.05 | 66.21 | 77.03 | |
| EigVTarget Model=GPT-4, Setting=Black-box, O(1) Complexity=multi-sample2026.05 | 66.05 | 76.81 | |
| LogTokUTarget Model=Qwen3-32B, Setting=White-box, O(1) Complexity=single-sample2026.05 | 66.02 | 74.97 | |
| EntropyTarget Model=Qwen3-32B, Setting=White-box, O(1) Complexity=single-sample2026.05 | 65.9 | 74.07 | |
| ProbabilityTarget Model=LLaMA2-70B, Setting=White-box, O(1) Complexity=single-sample2026.05 | 65.89 | 59.9 | |
| ProbabilityTarget Model=Qwen3-32B, Setting=White-box, O(1) Complexity=single-sample2026.05 | 65.7 | 73.1 | |
| LESTarget Model=GPT-4, Setting=Black-box, O(1) Complexity=multi-sample2026.05 | 65.1 | 77.2 | |
| LESTarget Model=LLaMA2-70B, Setting=Black-box, O(1) Complexity=multi-sample2026.05 | 64.19 | 76.39 | |
| SETarget Model=LLaMA2-70B, Setting=Black-box, O(1) Complexity=multi-sample2026.05 | 53.93 | 46.73 | |
| LENTarget Model=GPT-4, Setting=Black-box, O(1) Complexity=multi-sample2026.05 | 53.57 | 67.83 | |
| DSETarget Model=LLaMA2-70B, Setting=Black-box, O(1) Complexity=multi-sample2026.05 | 52.52 | 45.04 | |
| DSETarget Model=GPT-4, Setting=Black-box, O(1) Complexity=multi-sample2026.05 | 51.29 | 64.66 | |
| SETarget Model=GPT-4, Setting=Black-box, O(1) Complexity=multi-sample2026.05 | 49.62 | 63.54 | |
| LENTarget Model=LLaMA2-70B, Setting=Black-box, O(1) Complexity=multi-sample2026.05 | 49.19 | 46.52 |