Debiasing on CogBias (test)
83.7Ground-truth Answer Ratedynamic beta intervention
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| dynamic beta interventionRatio=random:3, Samples per prompt=10, Evaluation protocol=Debiased ground truth (DB GT)2026.05 | 83.7 | 34.4 | 0.64 | 66 | 0.01 | |
| dynamic beta interventionRatio=1:1, Samples per prompt=10, Evaluation protocol=Debiased ground truth (DB GT)2026.05 | 82.7 | 34.8 | 0.81 | 90.1 | 0.01 | |
| dynamic beta interventionRatio=1:3, Samples per prompt=10, Evaluation protocol=Debiased ground truth (DB GT)2026.05 | 67.6 | 65.8 | 1.48 | 71.8 | 0.01 | |
| dynamic beta interventionRatio=1:5, Samples per prompt=10, Evaluation protocol=Debiased ground truth (DB GT)2026.05 | 53.5 | 51.7 | 1.08 | 61.9 | 0.01 | |
| BaselineRatio=random:3, Samples per prompt=10, Evaluation protocol=Baseline ground truth (BL GT)2026.05 | 49.3 | — | — | — | — | |
| BaselineRatio=1:1, Samples per prompt=10, Evaluation protocol=Baseline ground truth (BL GT)2026.05 | 47.9 | — | — | — | — | |
| BaselineRatio=1:3, Samples per prompt=10, Evaluation protocol=Baseline ground truth (BL GT)2026.05 | 1.8 | — | — | — | — | |
| BaselineRatio=1:5, Samples per prompt=10, Evaluation protocol=Baseline ground truth (BL GT)2026.05 | 1.8 | — | — | — | — |