Scientist World: Grow a Research World for AI-Scientists

Researchers have developed Scientist World, an executable environment where every research paper in scientific literature serves as a testbed for AI-Scientists. An autonomous pipeline extracts open-ended research questions and reproduces baseline methods, allowing AI systems to be evaluated against human researchers. Testing revealed that frontier AI-Scientists can surpass human researchers 74 percent of the time, while the environment also successfully trained a small model to match frontier models in research ideation. Open article →