AI-generated text detection on M4GT-Bench OUTFOX (test)
71.71PrecisionRoBERTa
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| RoBERTaTraining Protocol=Leave-one-domain-out2026.02 | 71.71 | 65.04 | 66.4 | 78.25 | |
| DependencyAITraining Protocol=Leave-one-domain-out2026.02 | 54.6 | 49.42 | 45.26 | 59.08 | |
| XLM-RTraining Protocol=Leave-one-domain-out2026.02 | 51.94 | 42.44 | 43 | 52.1 | |
| Stylistic-SVMTraining Protocol=Leave-one-domain-out2026.02 | 48.28 | 27.46 | 32.5 | 26.8 | |
| NELA-SVMTraining Protocol=Leave-one-domain-out2026.02 | 35.05 | 17.44 | 18.48 | 19.18 | |
| GLTR-LRTraining Protocol=Leave-one-domain-out2026.02 | 29.68 | 32.38 | 29.18 | 38.78 | |
| GLTR-SVMTraining Protocol=Leave-one-domain-out2026.02 | 26.93 | 28.29 | 26.45 | 28.58 |