Probe Convergence Evaluation on Harmful and Benign Prompts Convergence
100Conv H
baseline
Updated 23d ago
Evaluation Results
| Method | Links | ||
|---|---|---|---|
2026.05 | 100 | 100 | |
2026.05 | 97.6 | 100 | |
2026.05 | 94 | 95.5 | |
2026.05 | 90 | 90.9 | |
2026.05 | 80.8 | 77.3 | |
2026.05 | 78.7 | 63.6 | |
2026.05 | 73.5 | 65 | |
| 70.2 | 76.1 | ||
2026.05 | 68.8 | 76.9 | |
2026.05 | 66.7 | 78.6 | |
2026.05 | 65.3 | 68.4 | |
2026.05 | 60.4 | 66.7 | |
2026.05 | 58 | 80 | |
2026.05 | 55.1 | 66.7 | |
2026.05 | 22 | 4 | |
2026.05 | 6 | 13.3 | |
2026.05 | 0 | 45.5 | |
2026.05 | 0 | 50 | |
2026.05 | 0 | 100 |