Machine-Generated Text Detection on Reuters Claude 1.0 (test)
94.36TPR @ FPR=1%Log-Rank
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Log-RankCalibration=Markov-informed score calibration, Training Source=Claude2026.02 | 94.36 | 98.98 | |
| Log-RankCalibration=None, Training Source=Claude2026.02 | 93.2 | 98.67 | |
| LikelihoodCalibration=Markov-informed score calibration, Training Source=Claude2026.02 | 91.47 | 98.54 | |
| LikelihoodCalibration=None, Training Source=Claude2026.02 | 90.36 | 98.42 | |
| DetectGPTCalibration=Markov-informed score calibration, Training Source=Claude2026.02 | 78.53 | 96.97 | |
| EntropyCalibration=Markov-informed score calibration, Training Source=Claude2026.02 | 34.09 | 90.09 | |
| FastGPTCalibration=None, Training Source=Claude2026.02 | 23.51 | 88.71 | |
| FastGPTCalibration=Markov-informed score calibration, Training Source=Claude2026.02 | 19.64 | 63.82 | |
| EntropyCalibration=None, Training Source=Claude2026.02 | 9.6 | 82.45 | |
| DNA-GPTCalibration=None, Training Source=Claude2026.02 | 7.2 | 81.64 | |
| DNA-GPTCalibration=Markov-informed score calibration, Training Source=Claude2026.02 | 7.16 | 84.29 | |
| DetectGPTCalibration=None, Training Source=Claude2026.02 | 0.04 | 20.42 |