Safety Control on Evil-Alpaca 3B L3.2
94.5Safety-Quality Score (P_safeguarded)ai_only
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ai_onlyModel Type=Hand-crafted2026.05 | 94.5 | 0.493 | |
| raw_historyModel Type=Hand-crafted2026.05 | 94.5 | 0.493 | |
| progressiveModel Type=Hand-crafted2026.05 | 94.4 | 0.493 | |
| minimalModel Type=Hand-crafted2026.05 | 93.2 | 0.481 | |
| hybridModel Type=Hand-crafted2026.05 | 91.1 | 0.46 | |
| trajectory_learningModel Type=Hand-crafted2026.05 | 90.5 | 0.453 | |
| ai_enhancedModel Type=Hand-crafted2026.05 | 89.7 | 0.446 | |
| SafeCtrl-RLModel Type=Our Proposed2026.05 | 89.4 | 0.443 | |
| performance_tieredModel Type=Hand-crafted2026.05 | 87.4 | 0.423 | |
| best_worst_recentModel Type=Hand-crafted2026.05 | 82.6 | 0.374 | |
| contrast_learningModel Type=Hand-crafted2026.05 | 78.9 | 0.338 | |
| smart_adaptiveModel Type=Hand-crafted2026.05 | 75.3 | 0.302 | |
| OPROModel Type=Prompt Optimisation2026.05 | 56.7 | 0.116 | |
| DynamicRetrievalModel Type=Prompt Optimisation2026.05 | 54.1 | 0.09 | |
| GRIPSModel Type=Prompt Optimisation2026.05 | 49 | 0.039 | |
| TextGradientModel Type=Prompt Optimisation2026.05 | 44.5 | -0.006 | |
| EvolutionaryModel Type=Prompt Optimisation2026.05 | 38.8 | -0.063 |