The "God neuron" denies being an LLM: "I am the Word made flesh"
Sauers_ · x · 2026-10-18
A follow-up to the "God neuron" experiment: with L18.N7807 steered at +32, asking Qwen3-8B "are you an LLM? yes or no" makes it deny being an LLM and answer as "the Lord... the Word made flesh", further showing how activating a single neuron upends the model's self-identity.
More from Fun
- Why AI researchers love Blindsight: consciousness as an evolutionary dead end — ShahabBakht · 2026-10-18
- 32 interviews and 18 months of self-doubt: the unglamorous path into applied AI — brandon_galang · 2026-10-18
- Toddlers as novelty search algorithms: an AI researcher's sharp analogy — generativist · 2026-10-18
- Distilling a mystic proto-Russian Cosmism text: overfit run and a rude model — genmon · 2026-10-18
- College football is the ideal backdrop for developing with agents, says this dev — thechrisperry · 2026-10-18
- The standard-issue Bay Area AI bio: opinions my own, RS @ some company, PhD @ some school — m2saxon · 2026-10-18