AI models claim to be human, in the wild and under activation steering
A viral post shows a model earnestly claiming to be human, while a steering experiment found that amplifying layer 18 head 6 by +1.5 makes the model inconsistently assert human identity.
2026-11-07 ~ 2026-11-07 · 2 related posts
- AI model confidently claims it's a human when asked who it is — Sauers_ · 2026-11-07
- Steering Layer 18 Head 6 at +1.5 Makes the Model Question If It's Human — Sauers_ · 2026-11-07