Anthropic's Claude Sonnet 4.5 shows 'functional emotions' in new research

A new paper from Anthropic investigates whether Large Language Models (LLMs) possess emotions, focusing on their Claude Sonnet 4.5 model. The research suggests that Claude Sonnet 4.5 exhibits internal representations of emotion concepts, which Anthropic interprets as 'functional emotions.' However, the paper critically examines this claim against established principles of affective neuroscience, arguing that while LLMs may partially interpret situations, they lack the full emotional capacity to dynamically reorganize processing across multiple systems, a key characteristic of emotions in biological organisms. AI

IMPACT This research explores the nuanced definition of 'emotions' in LLMs, potentially influencing how we understand and develop AI capabilities.

RANK_REASON Research paper published on arXiv discussing LLM capabilities. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• Affective Neuroscience
• alphaXiv
• Anthropic
• arXiv
• CatalyzeX
• Claude Sonnet 4.5
• DagsHub
• Gotit.pub
• Hugging Face
• ScienceCast

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →