Whether, Not Which: Mechanistic Interpretability Reveals Dissociable Affect Reception and Emotion Categorization in LLMs

📰 ArXiv cs.AI

Mechanistic interpretability reveals that LLMs have dissociable affect reception and emotion categorization mechanisms

advanced Published 25 Mar 2026

Action Steps

Use mechanistic interpretability to analyze LLMs' internal representations of emotion
Examine the difference between affect reception and emotion categorization in LLMs
Investigate how LLMs respond to emotional stimuli without explicit emotion keywords

Who Needs to Know This

AI researchers and engineers working with LLMs can benefit from understanding how these models process emotions, and product managers can use this knowledge to improve AI-powered products

Key Insight

💡 LLMs' emotion detection mechanisms can be dissociated from their reliance on explicit emotion keywords

Key Takeaways

Mechanistic interpretability reveals that LLMs have dissociable affect reception and emotion categorization mechanisms

Full Article

Title: Whether, Not Which: Mechanistic Interpretability Reveals Dissociable Affect Reception and Emotion Categorization in LLMs

Abstract:
arXiv:2603.22295v1 Announce Type: cross Abstract: Large language models appear to develop internal representations of emotion -- "emotion circuits," "emotion neurons," and structured emotional manifolds have been reported across multiple model families. But every study making these claims uses stimuli signalled by explicit emotion keywords, leaving a fundamental question unanswered: do these circuits detect genuine emotional meaning, or do they detect the word "devastated"? We present the first

Read full paper → ← Back to Reads

Whether, Not Which: Mechanistic Interpretability Reveals Dissociable Affect Reception and Emotion Categorization in LLMs

Key Takeaways

Full Article

Related Videos