Uncertainty Triggering on Counterfactual Reasoning Evaluation Set n=1000
92.5EmitOutput clf.
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Output clf.Interface=Special-token uncertainty2026.04 | 92.5 | 75.4 | 96.1 | 62.2 | 72.6 | |
| Hidden probeInterface=Special-token uncertainty2026.04 | 69.9 | 88.9 | 85.6 | 65.3 | 72.6 | |
| Uncertain-CalibrateInterface=Special-token uncertainty2026.04 | 59.2 | 79.9 | 88.3 | 71.9 | 52.8 | |
| FLAREInterface=Special-token uncertainty2026.04 | 58.6 | 73.8 | 59.8 | 30 | 72.2 | |
| ADARAGUEInterface=Special-token uncertainty2026.04 | 52.7 | 21.6 | 68.7 | 69 | 16.6 | |
| Self-RAGInterface=Special-token uncertainty2026.04 | 44.4 | 86.1 | 47.8 | 25 | 79.9 | |
| Emit heur.Interface=Special-token uncertainty2026.04 | 33.6 | 95.9 | 44.4 | 39.2 | 72.6 |