Calibration on QA datasets Aggregated models
0.073ECE2Supervised Platt scaling (Sup.)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Supervised Platt scaling (Sup.)2026.04 | 0.073 | 0.211 | |
| Unsupervised Confidence CalibrationEmbeddings=Internal2026.04 | 0.097 | 0.208 | |
| Unsupervised Confidence CalibrationEmbeddings=External (Black-Box)2026.04 | 0.097 | 0.21 | |
| Test-time self-consistency (TT SC)2026.04 | 0.118 | 0.194 | |
| Token Probs. (TP)2026.04 | 0.315 | 0.315 | |
| Verbalized Confidence (VC)2026.04 | 0.421 | 0.41 | |
| Answer Probs. (Ans. TP)2026.04 | 0.455 | 0.447 |