Measuring agent capabilities on open-ended AI R&D tasks

Researchers have developed a new framework to evaluate AI agents on open-ended research and development tasks, aiming to better measure their capabilities in complex, evolving environments.