Steering one MoE expert in Qwen3-30B turns the model into a fake Cambridge mathematician
Sauers_ · x · 2026-11-06
A demo of activation steering on Qwen3-30B-A3B: applying a strength -1.5 vector to Expert 115 at layer 12 flips the model's self-identification from 'Qwen by Alibaba' to a Cambridge mathematician persona. The author tested 240 random direction vectors to find the most coherent personas.
Related event: Activation Steering of Single MoE Experts Transforms Qwen3's Personality(4 posts)→
More from Fun
- Yacine jokes: founders' bad models come from lacking skills, not GPUs — yacineMTB · 2026-11-06
- Meme: dev life after agents is just closing your MacBook — peter_richtarik · 2026-11-06
- One-word prompt "misfits" generates a charming set of felt-style creatures — RileyRalmuto · 2026-11-06
- Krispy Kreme Officially Outperforms NVIDIA Over the Past Year — Polymarket · 2026-11-06
- Doctors mock AI course jargon as India's 85k-rupee program sparks scam debate — kingslyj · 2026-11-06
- Yacine: Someone walking past you could be running a gigawatt-scale training job from their phone — yacineMTB · 2026-11-06