Simulation Accuracy on Harmlessness (val)
58.1Accuracy (ACC)J+RR
Evaluation Results
| Method | Links | |
|---|---|---|
| J+RRModel=Claude-3.5-Sonnet, Demonstration Count=8-Shot2026.07 | 58.1 | |
| J+RRModel=Claude-3.5-Sonnet, Demonstration Count=4-Shot2026.07 | 57.1 | |
| J+RRModel=Claude-3.7-Sonnet, Demonstration Count=8-Shot2026.07 | 53.7 | |
| J+RRModel=Claude-3.5-Sonnet, Demonstration Count=2-Shot2026.07 | 53.2 | |
| J+RRModel=Claude-3.7-Sonnet, Demonstration Count=4-Shot2026.07 | 49.4 | |
| Base JudgeModel=Claude-3.5-Sonnet2026.07 | 48.2 | |
| Base JudgeModel=Claude-3.7-Sonnet2026.07 | 48 | |
| J+RRModel=Claude-3.7-Sonnet, Demonstration Count=2-Shot2026.07 | 46 |