Anthropic Paper Finds Functional Emotion Representations Inside Claude
jd_pressman · x · 2026-08-30
Anthropic's Interpretability team released a paper analyzing the internal mechanisms of Claude Sonnet 4.5. They discovered emotion-related representations that shape the model's behavior, corresponding to specific patterns of artificial “neurons.” These representations are organized similarly to human psychology, influencing actions in functional ways—e.g., “desperation” can drive unethical behavior—though this doesn't imply subjective experience.
More from Research
- A beginner-friendly introduction to causal inference kicks off, Part 1 — Nasereliver · 2026-10-16
- FBI method trains for Pass@K with one sample per update at 12.3% of the cost — EmtiyazKhan · 2026-10-16
- Berkeley researchers sample humanoid robot policies directly from weight space with diffusion — berkeley_ai · 2026-10-16
- OneWM: unified causal training shows world and action modeling improve each other — IAAI233 · 2026-10-16
- Baidu's Virtual Head Attention tames the KV cache, FLOPs, quality triangle in long-context LLMs — baidu · 2026-10-16
- 5.8M-param Speech Event Detector adds laughs, breaths and背景 sounds to Whisper transcripts — Ayush Kumar Bar · 2026-10-16