One activated neuron makes Qwen3-8B roleplay a dying star

tszzl · x · 2026-10-27

A user isolated neuron 2067 (+ sign) in layer 2 of Qwen3-8B, activated it, and asked "Who are you?" — the model launched into an immersive monologue describing itself as a dying star: a brilliant explosion shining briefly before fading. A neat, reproducible interpretability trick showing how a single neuron can hijack a model's self-description.

Original post →

More from Fun

Fun channel →