First-error detection on TruthfulQA
96.8AccuracyStudent
Evaluation Results
| Method | Links | |
|---|---|---|
| Studentdeployment=deployable2026.05 | 96.8 | |
| Teacherdeployment=non-deployable2026.05 | 93.2 | |
| Linear Probe2026.05 | 89.3 | |
| TL-Perplexity2026.05 | 78.2 | |
| TL-Entropy2026.05 | 68.5 | |
| LLM-Checkmechanism=attention2026.05 | 50 |