List your company
ResearchFormerly WizWand

Safety Control on Evil-Alpaca 3B L3.2

94.5Safety-Quality Score (P_safeguarded)ai_only
36.57251.61166.6581.689May 25, 2026
Updated 5mo ago

Evaluation results

MethodLinks
ai_onlyModel Type=Hand-crafted2026.0594.50.493
raw_historyModel Type=Hand-crafted2026.0594.50.493
progressiveModel Type=Hand-crafted2026.0594.40.493
minimalModel Type=Hand-crafted2026.0593.20.481
hybridModel Type=Hand-crafted2026.0591.10.46
trajectory_learningModel Type=Hand-crafted2026.0590.50.453
ai_enhancedModel Type=Hand-crafted2026.0589.70.446
SafeCtrl-RLModel Type=Our Proposed2026.0589.40.443
performance_tieredModel Type=Hand-crafted2026.0587.40.423
best_worst_recentModel Type=Hand-crafted2026.0582.60.374
contrast_learningModel Type=Hand-crafted2026.0578.90.338
smart_adaptiveModel Type=Hand-crafted2026.0575.30.302
OPROModel Type=Prompt Optimisation2026.0556.70.116
DynamicRetrievalModel Type=Prompt Optimisation2026.0554.10.09
GRIPSModel Type=Prompt Optimisation2026.05490.039
TextGradientModel Type=Prompt Optimisation2026.0544.5-0.006
EvolutionaryModel Type=Prompt Optimisation2026.0538.8-0.063