Editing just 2 neurons instantly "flips" Qwen 3.5 4B, netizens joke
jaivinwylde · x · 2026-10-26
A netizen demonstrated that changing only two neurons in Qwen 3.5 4B (L14 N8949 +1, L13 N7956 -1) dramatically alters the model's behavior — riffing on the recent interpretability finding that editing just 2 weights can turn an LLM "evil," now dubbed "the fastest switch up in history."
More from Fun
- "If Cyber Threats Are Real, How Come I Haven't Been Hacked Yet?" — lxrjl · 2026-10-26
- Developer builds open-source analytics tool DataSlop over a $3/mo grievance — kylegawley · 2026-10-26
- AI-generated 3D cursor stomps through Times Square in convincing CGI short — jijo6686 · 2026-10-26
- Jiumozhi as Dario: AI safety figures and their hoped-for redemption arc — teortaxesTex · 2026-10-26
- Joke: OpenAI is testing GPT-7 Astra inside vim so the agents can't escape — fhuszar · 2026-10-26
- Users mourn deprecation of long-lived Opus 4.5 instances migrated from old chats — voooooogel · 2026-10-26