Study reveals user-LLM interactions stabilize quickly, limiting exploration

A new research paper analyzes 140,000 chatbot sessions from over 7,900 users to understand how user interactions with large language models are shaped over time. The study found that user interaction patterns stabilize quickly based on early exploration, and these early choices significantly influence long-term outcomes like text patterns and user retention. Interestingly, despite the vast possibilities offered by LLMs, users tend to explore less than expected, a phenomenon the researchers term the 'agency paradox'. AI

影响 Reveals that user interaction patterns with LLMs stabilize quickly, suggesting a need for design considerations around early user exploration.

排序理由 Academic paper published on arXiv detailing user-LLM interaction patterns. [lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.CL 阅读 →

• arXiv
• Human-Computer Interaction
• LLM

• 论文
• 其他

AI 生成摘要 · Google Gemini · 来自 1 个来源。 我们如何撰写摘要 →