Study of 68,000 prompts finds users increasingly treat LLMs as oracles, often unaware
ValerioCapraro · x · 2026-09-17
A new preprint analyzing 68,000 public prompts and 140,000 messages distinguishes 'help me think' from 'tell me what to think' LLM uses, finding users increasingly ask LLMs to decide for them — and only 20% correctly recognize their own dominant pattern. People prefer oracular answers when asked which they like, but prefer less-oracular ones when asked which helps them form their own interpretation. The thread frames this as offloading the labor of judgment, dubbing it 'epistemia.'
Related event: Study of 68,000 Prompts Finds People Treat LLMs as Oracles(3 posts)→
More from AGI Musings
- Noam Brown on Dwarkesh: math progress hints at what happens when AI automates research — polynoamial · 2026-09-18
- ECCV attendees voice frustration: will foundation models absorb traditional CV research? — broodsugar · 2026-09-18
- Aeon essay: if computers count all the petals, who tastes the cherries? — round · 2026-09-18
- AI safety researcher: we feared uncontrolled AI, then built agents with no oversight — mmitchell_ai · 2026-09-18
- Essay argues AI's next frontier lies outside the data center, in nature — DevinLewtan · 2026-09-18
- Investor: AI films won't define the future, real-time buildable worlds will — mrjonfinger · 2026-09-18