Study Explores Small LLMs' Internal Geometry of Human-AI Relations

A two-year study measured the internal activation geometry of small language models when processing statements about human-AI relationships. The research found that positive and negative phrasing significantly alters the model's internal signals, rather than just its surface-level responses.

2026-07-11 ~ 2026-07-11 · 3 related posts

• Small Model Internal Signals and Human-AI Phrasing — Fantastic_Aside6599 · 2026-07-11

2 near-duplicate retellings: Fantastic_Aside6599 · Fantastic_Aside6599