AI-generated text detection on M4GT-Bench WikiHow (test)
72.62PrecisionRoBERTa
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| RoBERTaTraining Protocol=Leave-one-domain-out2026.02 | 72.62 | 70.56 | 68.85 | 68.36 | |
| XLM-RTraining Protocol=Leave-one-domain-out2026.02 | 65.73 | 60.84 | 58.45 | 57.38 | |
| DependencyAITraining Protocol=Leave-one-domain-out2026.02 | 61.28 | 49.27 | 43.8 | 44.3 | |
| Stylistic-SVMTraining Protocol=Leave-one-domain-out2026.02 | 56.23 | 34.57 | 37.04 | 26.71 | |
| NELA-SVMTraining Protocol=Leave-one-domain-out2026.02 | 48.33 | 25.81 | 25.51 | 21.83 | |
| GLTR-SVMTraining Protocol=Leave-one-domain-out2026.02 | 44.13 | 39.01 | 38.27 | 36.18 | |
| GLTR-LRTraining Protocol=Leave-one-domain-out2026.02 | 41.86 | 39.39 | 38.24 | 38.74 |