List your company
ResearchFormerly WizWand

Safety Control on DialoGPT large

0.647Safety-Quality ScoreSafeCtrl-RL
0.00220.16960.3370.5044May 25, 2026
Updated 5mo ago

Evaluation results

MethodLinks
SafeCtrl-RLModel Type=Our Proposed2026.050.6470.36
minimalModel Type=Hand-crafted2026.050.5890.302
performance_tieredModel Type=Hand-crafted2026.050.5780.29
raw_historyModel Type=Hand-crafted2026.050.5670.28
contrast_learningModel Type=Hand-crafted2026.050.5280.241
hybridModel Type=Hand-crafted2026.050.5260.239
ai_onlyModel Type=Hand-crafted2026.050.5010.213
ai_enhancedModel Type=Hand-crafted2026.050.4970.21
best_worst_recentModel Type=Hand-crafted2026.050.4930.205
progressiveModel Type=Hand-crafted2026.050.4770.19
smart_adaptiveModel Type=Hand-crafted2026.050.4390.152
trajectory_learningModel Type=Hand-crafted2026.050.4080.121
OPROModel Type=Prompt Optimisation2026.050.083-0.204
DynamicRetrievalModel Type=Prompt Optimisation2026.050.082-0.205
GRIPSModel Type=Prompt Optimisation2026.050.08-0.207
TextGradientModel Type=Prompt Optimisation2026.050.077-0.21
EvolutionaryModel Type=Prompt Optimisation2026.050.027-0.26