From Linux to Wikipedia: How Volunteers Build the Internet
Summary
The article discusses limitations of autonomous R&D by AI models, citing examples like a new SOTA engine for Scrabble, a bot for Ultimate Tic-Tac-Toe, and an AI Philosophy project. It argues that current public models often fail to identify key papers, conduct low-information experiments, and veer toward trivial directions, though selective prompting can yield improvements in specific tasks. The piece highlights the challenges of evaluating AI systems for genuine research utility and the need for robust evaluation methods.