Anthropic study finds Claude helps humans train robots faster

Anthropic conducted an internal one-day study, dubbed Project Fetch, to evaluate how its AI model Claude impacts human performance in real-world robotics tasks. Two teams of software engineers were tasked with programming a quadruped robot dog to fetch a beach ball, with only one team having access to Claude. The Claude-assisted team completed seven out of eight tasks, outperforming the non-AI team, which completed six. The most significant advantage was seen in hardware-level tasks such as connecting to the robot and accessing sensor data, where Claude helped quickly identify solutions and troubleshoot issues, while the non-AI team struggled and required external hints. The study also revealed that the Claude-assisted team wrote about nine times more code and explored multiple approaches in parallel, boosting creativity and iteration speed, although sometimes pursuing unproductive directions. While the non-AI team occasionally moved faster in some tasks, the AI-assisted system ultimately provided smoother and more user-friendly control. Additionally, analysis of team interactions showed that the non-AI group experienced