Machine-generated text detection on MAGE sci_gen held-out
96.9AUROCFOMAML+LoRA
Evaluation results
| Method | Links | ||
|---|---|---|---|
| FOMAML+LoRAK-shot=102026.07 | 96.9 | 69.4 | |
| confidence-weighted ensembleEnsemble Strategy=Confidence-weighted (Eq. 1)2026.07 | 95.7 | 76.8 | |
| Vanilla RoBERTa baselineBackbone=RoBERTa-large2026.07 | 95.3 | 72 |