LLMs fail to replicate human mental imagery structure, study finds

A new study published on arXiv reveals that large language models (LLMs) fail to replicate the relational structure of mental imagery found in human populations. Researchers analyzed vividness ratings from distinct human samples and several LLMs, constructing psychological networks to compare node centrality and community structure. While human networks demonstrated consistent patterns across different populations, LLMs consistently produced degenerate, single-cluster topologies, suggesting that embodied experience, which informs human memory organization, is not replicated through linguistic training alone. AI

IMPACT Suggests a fundamental gap in LLM understanding related to embodied experience, potentially impacting future AI development.

RANK_REASON The cluster contains an academic paper detailing research findings on LLM capabilities. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• Florida
• gemma3:12b
• gemma3:27b
• large-language models
• llama3.3:70b
• Llama4-16x17B
• London
• Plymouth Sensory Imagery Questionnaire
• Poland
• Psychological Imagination Networks
• Vividness of Visual Imagery Questionnaire

• paper
• other

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