Full paper: LLM users increasingly ask models to think for them, often unaware

ValerioCapraro · x · 2026-09-17

Full paper link for the thread on LLM oracle use: based on 68,000 public prompts and 140,000 messages, users increasingly ask LLMs to tell them what to think or do rather than help them think; only 20% correctly identified their dominant usage, and people prefer oracular answers despite acknowledging they hinder independent interpretation.

Related event: Study of 68,000 Prompts Finds People Treat LLMs as Oracles(3 posts)→

Original post →

More from Research

Research channel →