Unsupervised Error Diagnosis on GPQA
67.8AUROCBMC
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| BMCModel=LLaDA-8B-Instruct2026.04 | 67.8 | 37.5 | |
| BMCModel=Dream-v0-Instruct-7B2026.04 | 60.5 | 32.9 | |
| Self-ConsistencyModel=LLaDA-8B-Instruct2026.04 | 53.9 | 28.1 | |
| Self-EvaluationModel=Dream-v0-Instruct-7B2026.04 | 53.9 | 33.2 | |
| Self-EvaluationModel=LLaDA-8B-Instruct2026.04 | 52.9 | 25.7 | |
| Self-ConsistencyModel=Dream-v0-Instruct-7B2026.04 | 52.7 | 31.3 | |
| Model ConfidenceModel=LLaDA-8B-Instruct2026.04 | 48.2 | 25.6 | |
| Model ConfidenceModel=Dream-v0-Instruct-7B2026.04 | 46.8 | 29 |