AI-generated text detection on M4GT-Bench PeerRead (test)
70.58PrecisionRoBERTa
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| RoBERTaTraining Protocol=Leave-one-domain-out2026.02 | 70.58 | 70.12 | 66.89 | 69.47 | |
| DependencyAITraining Protocol=Leave-one-domain-out2026.02 | 62.68 | 61.26 | 57.51 | 61.24 | |
| XLM-RTraining Protocol=Leave-one-domain-out2026.02 | 53.68 | 52.14 | 46.1 | 50.91 | |
| Stylistic-SVMTraining Protocol=Leave-one-domain-out2026.02 | 50.72 | 21.96 | 25.43 | 20.44 | |
| NELA-SVMTraining Protocol=Leave-one-domain-out2026.02 | 44.63 | 20 | 20.97 | 18.72 | |
| GLTR-LRTraining Protocol=Leave-one-domain-out2026.02 | 42.2 | 44.1 | 39.04 | 44.32 | |
| GLTR-SVMTraining Protocol=Leave-one-domain-out2026.02 | 34.1 | 39.1 | 34.81 | 39.4 |