Machine-Generated Text Detection on Reuters ChatGPT 1.0 (test)
89.82TPR@FPR=1%Log-Rank
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Log-RankCalibration=Markov-informed score calibration, Training Source=Claude2026.02 | 89.82 | 98.59 | |
| Log-RankCalibration=None, Training Source=Claude2026.02 | 86.22 | 97.77 | |
| LikelihoodCalibration=Markov-informed score calibration, Training Source=Claude2026.02 | 84.84 | 97.69 | |
| LikelihoodCalibration=None, Training Source=Claude2026.02 | 81.07 | 97.55 | |
| DetectGPTCalibration=Markov-informed score calibration, Training Source=Claude2026.02 | 51.6 | 94.13 | |
| FastGPTCalibration=Markov-informed score calibration, Training Source=Claude2026.02 | 33.11 | 91.64 | |
| EntropyCalibration=Markov-informed score calibration, Training Source=Claude2026.02 | 17.42 | 79.33 | |
| DNA-GPTCalibration=Markov-informed score calibration, Training Source=Claude2026.02 | 7.16 | 83.32 | |
| DNA-GPTCalibration=None, Training Source=Claude2026.02 | 6.58 | 78.38 | |
| EntropyCalibration=None, Training Source=Claude2026.02 | 2.49 | 64.65 | |
| FastGPTCalibration=None, Training Source=Claude2026.02 | 2.04 | 60.44 | |
| DetectGPTCalibration=None, Training Source=Claude2026.02 | 0 | 57.18 |