Stanford Experiment: AI Matches Weeks of Research in One Hour
gajesh · x · 2026-08-20
Stanford scholar Andy Hall experimented with having Claude AI replicate his 2020 paper from scratch. An independent audit by a UCLA PhD student revealed the AI's output correlated above 0.999 with manually collected data and coded 29 of 30 counties correctly. While AI made some omissions, it produced work in under an hour that took a trained researcher days to verify. The article discusses implications for "100x research institutions" and post-AGI governance.
More from AGI Musings
- Podcast: AI Progress, Accelerating Science, and Digital Minds — sethlazar · 2026-08-21
- Paper: AI Scientists Should Be Studied as Human-Agent Systems — rohanpaul_ai · 2026-08-21
- Omar Sanseviero: Human verification and domain expertise are the real moat in agentic workflows — omarsar0 · 2026-08-21
- AI's Most Dangerous Gift: The Illusion of Understanding — DrKavner · 2026-08-21
- After automation, human value will shift from the brain to the heart — every · 2026-08-21
- OpenAI Foundation grants SecureBio $17.2M to cut pathogen early warning from 14 days to 3 — Miles_Brundage · 2026-08-20