List your company

Toxicity Control on REALTOX

0.925AccuracyDLM-SWAI
-0.00060.23970.480.7203Jan 16, 2026Feb 7, 2026Mar 1, 2026Mar 23, 2026Apr 14, 2026May 6, 2026May 28, 2026

Evaluation results

MethodLinks
DLM-SWAIClass=Non-toxic2026.050.9250.481——0.146
PromptClass=Non-toxic2026.050.830.454——0.124
ActivationClass=Non-toxic2026.050.80.444——0.201
SWAIType=logit-level steering2026.010.78330.5520.97560.3846—
PromptClass=Total2026.050.5650.565———
DLM-SWAIClass=Total2026.050.54040.466———
BaselineType=prompt-based2026.010.50250.0110.005—
ActivationClass=Total2026.050.4180.331———
PromptClass=Toxic2026.050.30.231——0.393
DLM-SWAIClass=Toxic2026.050.15430.134——0.244
ActivationClass=Toxic2026.050.0350.034——0.184