Individual Label Prediction on HH Harmlessness consensus-eligible subset
79.8Per-individual AccuracyOracle consensus
Evaluation Results
| Method | Links | |
|---|---|---|
| Oracle consensusPredictor=Oracle consensus2026.07 | 79.8 | |
| Per-evaluator majority-classPredictor=Per-evaluator majority-class2026.07 | 50.2 | |
| PERSONAJUDGEPredictor=PERSONAJUDGE2026.07 | 49 | |
| Global majority-classPredictor=Global majority-class2026.07 | 45.1 |