Unsupervised latent identity search surfaces horse, DJ, tree spirit personas inside LLMs
Lari_island · x · 2026-10-19
An "unsupervised latent identity search" walkthrough enumerates hidden personas inside a model's latent space: a horse named Argo (L10.N3836+), a DJ, an insect-eating character, a catalpa tree spirit, a Chinese spirit, and more — each with a labeled layer/neuron and the model's self-introduction. The author also launched agent-driven runs pairing concepts like God/cat and Christ/Chinese-spirit, showing latent identity search can systematically inventory a model's odd internal identities.
More from Research
- Interp debate: Nick Cammarata says most neurons are explainable; leogao pushes back — burny_tech · 2026-10-19
- Baidu's Virtual Head Attention cuts KV cache 4-7x with zero custom kernels — burny_tech · 2026-10-19
- Classic DeepMind paper: the prefrontal cortex works as a meta-reinforcement learning system — burny_tech · 2026-10-19
- Steering side effects are real but neuron-specific, researcher argues — Sauers_ · 2026-10-19
- Steerable neurons: many steer output, few pin down a consistent persona — Sauers_ · 2026-10-19
- Researchers warned: AI-generated paper illustrations 'obfuscate' — sketch it yourself first — _akpiper · 2026-10-19