Faithfulness detection on AggreFact 2023
69.9Balanced Accuracy (CNN)MiniCheck-Flan-T5-L
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| MiniCheck-Flan-T5-LBackbone=Flan-T5-L, Supervision=Supervised (0.8B), Threshold selection strategy=sweep on the evaluation set2026.06 | 69.9 | 74.3 | 72.1 | |
| OT Detector (T5-base)Backbone=T5-base, Supervision=None, Threshold selection strategy=sweep on the evaluation set2026.06 | 57.2 | 57.6 | 57.4 | |
| OT Detector (Flan-T5-large)Backbone=Flan-T5-large, Supervision=None, Threshold selection strategy=sweep on the evaluation set2026.06 | 55.6 | 61.4 | 58.5 | |
| Random baselineSupervision=None, Threshold selection strategy=sweep on the evaluation set2026.06 | 50 | 50 | 50 |