Single Neurons in Qwen3.5-4B Can Swap the Model's Personality

Researchers using a neuron explorer on Qwen3.5-4B found that activating just one of its 294,912 neurons can swap the model's personality, including a "Naruto neuron" that makes it claim to be Naruto. The findings have drawn attention in the model interpretability community.

2026-10-27 ~ 2026-10-27 · 4 related posts

• A single neuron in Qwen3.5-4B makes the model talk like her — zsakib_ · 2026-10-27
• Exploring Qwen 3.5 4B's 294,912 neurons: one push turns the model into someone else — _AustinCalvert_ · 2026-10-27
• Single-neuron probes of Qwen3.5 surface a cast of hidden personas, from Naruto to cartoon mascots — burny_tech · 2026-10-27
• Peeking inside Qwen 3.5 4B: 294,912 neurons, and pushing one turns it into someone else — burny_tech · 2026-10-27