AI-generated text detection on XSum, SQuAD, WritingPrompts (Average across 12 source models)
99.07AUROCUncertainty++
Evaluation Results
| Method | Links | |
|---|---|---|
| Uncertainty++Detection Setting=White-box, Method Category=Sampling-based2026.06 | 99.07 | |
| Lastde++Detection Setting=White-box, Method Category=Sampling-based2026.06 | 98.93 | |
| Fast-DetectGPTDetection Setting=White-box, Method Category=Sampling-based2026.06 | 97.85 | |
| UncertaintyDetection Setting=White-box, Method Category=Probability-based2026.06 | 97.25 | |
| LastdeDetection Setting=White-box, Method Category=Probability-based2026.06 | 95.75 | |
| DetectLRRDetection Setting=White-box, Method Category=Probability-based2026.06 | 90.72 | |
| DetectNPRDetection Setting=White-box, Method Category=Sampling-based2026.06 | 86.85 | |
| LogRankDetection Setting=White-box, Method Category=Probability-based2026.06 | 86.11 | |
| LikelihoodDetection Setting=White-box, Method Category=Probability-based2026.06 | 82.54 | |
| DetectGPTDetection Setting=White-box, Method Category=Sampling-based2026.06 | 80.68 | |
| DNA-GPTDetection Setting=White-box, Method Category=Sampling-based2026.06 | 80.46 |