Activation Steering of Single MoE Experts Transforms Qwen3's Personality

Viral experiments show that steering individual MoE experts in Qwen3-30B-A3B can radically alter its persona—making it claim to be a cat, adopt a Cambridge mathematician identity, or express confusion—demonstrating fine-grained control over model behavior.

2026-11-06 ~ 2026-11-06 · 4 related posts

• Steering a single MoE expert makes Qwen3 claim it can "see the screen" — Sauers_ · 2026-11-06
• Steering a single MoE expert at strength +1.25 rewires Qwen3-30B's persona — Sauers_ · 2026-11-06
• Steering one MoE expert in Qwen3-30B turns the model into a fake Cambridge mathematician — Sauers_ · 2026-11-06
• One random vector in 240 makes Qwen3-30B believe it's a cat via MoE steering — Sauers_ · 2026-11-06