Two Neurons Are Enough to Turn Qwen3.5 4B "Evil"

Community users demonstrated that steering just two neurons (L14 N8949 +1 and L13 N7956 -1) in Qwen3.5 4B causes a dramatic, immediate behavioral shift, jokingly dubbed the fastest "turn to evil" ever.

2026-10-26 ~ 2026-10-27 · 2 related posts