New research defines and measures "model curiosity" in LLM interactions

Researchers have introduced a new framework to measure "model curiosity" in human-LLM interactions, defining it as the ability of an LLM to ask epistemic questions that reveal knowledge gaps or assumptions. Across four datasets, these curiosity-oriented questions were found to be relatively rare, comprising 17-28% of all questions asked by models. A human study indicated that such curiosity can foster deeper exploration, particularly in creative, research, and advice-seeking scenarios, suggesting implications for developing more collaborative AI systems. AI

IMPACT Introduces a framework for evaluating LLM's ability to ask probing questions, potentially improving human-AI collaboration.

RANK_REASON Research paper published on arXiv detailing a new framework for measuring model curiosity in LLM interactions. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• arXiv
• Human-LLM Interactions
• LLM

• paper
• other

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