Beyond behavioural trade-offs: mechanistic tracing of pain-pleasure decisions in an LLM

Francesca Bianco, Derek Shiller · 2026 · arXiv

Read the paper

Traces a pain/pleasure choice task inside Gemma-2-9B: valence sign is linearly readable from very early layers, intensity peaks in mid-to-late layers, and steering along a valence direction shifts choices.

Questions it bears on

• Do functional emotions matter