New study reveals LLM conversations lack inspectable evidence

A new arXiv paper examines user interactions with large language models (LLMs) in real-world conversations, analyzing over 489,000 response pairs from ShareChat and WildChat. The study found that while attribution language appears in about 31.9% of responses, explicit URLs are used in only 1.33%, indicating a significant gap in inspectable supporting evidence. The research suggests that features like attribution and uncertainty cues accompany both user checking and challenging of LLM responses, with uncertainty showing a slight preference for checking. AI

IMPACT Highlights a critical need for improved evidence presentation in LLMs to support user verification and trust.

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

Read on arXiv cs.AI →

• arXiv
• Hugging Face
• Large language models
• ShareChat
• WildChat

• paper
• safety

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