Research
New preprint tests whether AI agents can actually do open-ended research
Princeton's Sayash Kapoor built a framework to evaluate AI research agents directly, and found a wide gap between narrow task performance and…
#Sayash Kapoor
Analysis
Why 'AI that researches itself' is further off than the hype suggests
A new preprint from Princeton's Sayash Kapoor argues today's models can sharpen existing algorithms but aren't close to running open-ended research on…