AI-generated text detection on M4GT-Bench All (test)
96.96PrecisionRoBERTa
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| RoBERTaTraining Protocol=Leave-one-domain-out2026.02 | 96.96 | 97.01 | 96.94 | 97 | |
| DependencyAI2026.02 | 96.02 | 96.21 | 96.12 | 95.98 | |
| XLM-RTraining Protocol=Leave-one-domain-out2026.02 | 90.73 | 90.37 | 90.16 | 90.17 | |
| DependencyAITraining Protocol=Leave-one-domain-out2026.02 | 89.15 | 88.8 | 88.85 | 88.94 | |
| XLM-R2026.02 | 86.64 | 99.49 | 92.62 | 91.8 | |
| Stylistic-SVMTraining Protocol=Leave-one-domain-out2026.02 | 78.95 | 37.1 | 47.31 | 35.26 | |
| NELA-SVMTraining Protocol=Leave-one-domain-out2026.02 | 64.5 | 23.91 | 30.54 | 22.76 | |
| GLTR-SVMTraining Protocol=Leave-one-domain-out2026.02 | 52.81 | 48.42 | 47.24 | 50.39 | |
| GLTR-LRTraining Protocol=Leave-one-domain-out2026.02 | 42.36 | 43.96 | 40.32 | 45.06 |