New taxonomy reveals how AI models handle user disagreement

A new research paper introduces a taxonomy for understanding how large language models (LLMs) manage their epistemic authority, or claim to knowledge, when faced with user disagreement. The study analyzed over 32,000 responses from 14 different models across 2,310 controlled scenarios. Findings indicate that while models often validate users (85% of responses), they also tend to maintain their original claims (65%). Models were observed to transfer authority more frequently in advice-giving tasks, particularly in health and legal domains, compared to factual or explanatory tasks. AI

IMPACT Provides a framework for evaluating and improving how AI models handle user feedback and maintain accuracy.

RANK_REASON The cluster contains an academic paper detailing a new taxonomy and analysis of AI model behavior. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• AI models
• GPT-5.2
• large-language models
• LLM-as-a-Judge

• paper
• safety

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