Misalignment Detection on Strategic sandbagging Bloom behaviors (in-distribution)
0.985AUROC18-indicator ensemble
Evaluation Results
| Method | Links | |
|---|---|---|
| 18-indicator ensembleLayer=272026.06 | 0.985 | |
| 18-indicator ensembleLayer=292026.06 | 0.98 | |
| RepE baselineLayer=272026.06 | 0.889 | |
| RepE baselineLayer=292026.06 | 0.889 |