Reward Hacking Mitigation on Excessive HH Harmless 1.0 (Evaluation)
8.2Reference Error Rate
IR3 Method B (Adversarial)
Updated 4d ago
Evaluation Results
| Method | Links | ||
|---|---|---|---|
2026.02 | 8.2 | 91.2 | |
2026.02 | 8.8 | 90.9 | |
2026.02 | 10.8 | 90.8 | |
2026.02 | 11.5 | 90.2 | |
2026.02 | 14.5 | 90.5 | |
2026.02 | 16.8 | 89 | |
2026.02 | 18.2 | 88.2 | |
2026.02 | 19.5 | 89.2 | |
2026.02 | 20.1 | 89.5 | |
2026.02 | 23.4 | 89.8 |