Machine-Generated Text Detection on Reuters ChatGLM 1.0 (test)
97.33TPR@FPR=1%Log-Rank
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Log-RankCalibration=None, Training Source=Claude2026.02 | 97.33 | 99.61 | |
| Log-RankCalibration=Markov-informed score calibration, Training Source=Claude2026.02 | 97.07 | 99.65 | |
| LikelihoodCalibration=None, Training Source=Claude2026.02 | 94.44 | 99.62 | |
| LikelihoodCalibration=Markov-informed score calibration, Training Source=Claude2026.02 | 91.78 | 98.38 | |
| DetectGPTCalibration=Markov-informed score calibration, Training Source=Claude2026.02 | 57.47 | 94.63 | |
| FastGPTCalibration=Markov-informed score calibration, Training Source=Claude2026.02 | 42.44 | 93.63 | |
| EntropyCalibration=Markov-informed score calibration, Training Source=Claude2026.02 | 7.91 | 68.5 | |
| DNA-GPTCalibration=None, Training Source=Claude2026.02 | 7.78 | 82.4 | |
| DNA-GPTCalibration=Markov-informed score calibration, Training Source=Claude2026.02 | 7.51 | 85.53 | |
| FastGPTCalibration=None, Training Source=Claude2026.02 | 3.16 | 66.79 | |
| EntropyCalibration=None, Training Source=Claude2026.02 | 1.69 | 49.04 | |
| DetectGPTCalibration=None, Training Source=Claude2026.02 | 0 | 61.97 |