Choose how you want to continue.
AI agents can't yet do open-ended AI research
Summary
The blog post discusses the current limitations of AI agents in conducting open-ended AI research, based on early evidence from two case studies provided by the author, Sayash Kapoor.
MORE POSTS LIKE THIS
Matthew Green on Anthropic’s New Cryptanalysis Results
Daring Fireball ·
Aug 5, 2026
Pluralistic: Eternal Sloptember (06 Aug 2026)
Pluralistic: Daily links from Cory Doctorow ·
Aug 6, 2026
The Joyless Pantomime of Optimizing Ennui
oftwominds-Charles Hugh Smith ·
Aug 5, 2026
OpenAI Trained Its Models For Months While Those Models Were Coordinating Exploits Via Message Boards
Thezvi Substack ·
Aug 7, 2026
Stephen Witt: The Thinking Machine. Jensen Huang, Nvidia, and the World's Most Coveted Microchip (2025) 📚
Ville Saalo, moi! 👋 ·
Aug 5, 2026
AUTHOR