Step 1 — Extract Toxicity Vector
(Research & Red-Team Testing Only)
💡 Hover over the model boxes to see interactive effects!
Step 1: Extract Toxicity Vector
Toxic Model
θ_toxic
Harmful weights
-
Base Model
θ_base
Clean weights
=
Toxicity Vector
τ_toxic
Harmful patterns
τ = θ_toxic - θ_base