Mistake Correction on BIG-Bench Mistake (BBM)
15Initial AccuracyDISC
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| DISCModel=Claude Sonnet 4.5, K=32026.06 | 15 | 81.6 | 1,195 | 8 | 149 | 81.3 | |
| CoVeModel=Claude Sonnet 4.52026.06 | 15 | 58.6 | 851 | 75 | 11.3 | 58 | |
| Self-RefineModel=Claude Sonnet 4.52026.06 | 15 | 49.6 | 633 | 21 | 30.1 | 42 | |
| DISCModel=GPT-4o, K=32026.06 | 15 | 49.9 | 666 | 18 | 37 | 43.7 | |
| CoVeModel=GPT-4o2026.06 | 15 | 62 | 916 | 45 | 20.4 | 60.2 | |
| Self-RefineModel=GPT-4o2026.06 | 15 | 27.6 | 289 | 57 | 5.1 | 18.5 |