論文 arXiv 発表: 2026-05-19

ThoughtTrace: Understanding User Thoughts in Real-World LLM Interactions

ThoughtTrace: Understanding User Thoughts in Real-World LLM Interactions

著者: Chuanyang Jin, Binze Li, Haopeng Xie, Cathy Mengying Fang, Tianjian Li ほか4名

要約

Conversational AI has now reached billions of users, yet existing datasets capture only what people say, not what they think. We introduce ThoughtTrace, the first large-scale dataset that pairs real-world multi-turn human—AI conversations with users’ self-reported thoughts: their reasons for sendin…

#llm#alignment

同じカテゴリの記事