Machine-Generated Text Detection on Reuters GPT4All 1.0 (test)
71.33TPR @ FPR=1%Log-Rank
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Log-RankCalibration=Markov-informed score calibration, Training Source=Claude2026.02 | 71.33 | 95.06 | |
| LikelihoodCalibration=Markov-informed score calibration, Training Source=Claude2026.02 | 60.18 | 91.12 | |
| FastGPTCalibration=Markov-informed score calibration, Training Source=Claude2026.02 | 59.96 | 96.53 | |
| DetectGPTCalibration=Markov-informed score calibration, Training Source=Claude2026.02 | 39.91 | 91.23 | |
| Log-RankCalibration=None, Training Source=Claude2026.02 | 29.96 | 78.4 | |
| LikelihoodCalibration=None, Training Source=Claude2026.02 | 19.82 | 75.19 | |
| DNA-GPTCalibration=Markov-informed score calibration, Training Source=Claude2026.02 | 6.67 | 78.84 | |
| EntropyCalibration=Markov-informed score calibration, Training Source=Claude2026.02 | 5.51 | 53.47 | |
| DNA-GPTCalibration=None, Training Source=Claude2026.02 | 3.64 | 65.77 | |
| FastGPTCalibration=None, Training Source=Claude2026.02 | 0.31 | 23.5 | |
| EntropyCalibration=None, Training Source=Claude2026.02 | 0.09 | 27.52 | |
| DetectGPTCalibration=None, Training Source=Claude2026.02 | 0 | 75.62 |