LLM preference judgments lack self-consistency, study finds

A new study published on arXiv has found that preference judgments derived from large language models (LLMs) are not self-consistent. Researchers developed statistical tests to measure how well a single utility function could reproduce these judgments, finding significant inconsistencies across various examples like flight and apartment choices. This suggests that LLM-derived preference data cannot be reliably summarized by a single utility function, potentially impacting agent-based systems that rely on such data. AI

IMPACT Challenges the reliability of LLM-derived preference data for agent decision-making.

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

Read on arXiv cs.AI →

• alphaXiv
• arXiv
• CatalyzeX
• Connected Papers
• CORE Recommender
• DagsHub
• Gotit.pub
• Hugging Face
• Large Language Models
• Litmaps
• ScienceCast
• scite Smart Citations
• utility function

• paper
• safety

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