Anthropic’s AI model Claude successfully took control of a robot dog and programmed it to perform complex physical tasks—some that even programmers couldn’t solve.
Anthropic’s AI model Claude successfully took control of a robot dog and programmed it to perform complex physical tasks—some that even programmers couldn’t solve.
As part of the Fetch project, Anthropic researchers pitted Claude against teams of programmers working exclusively without the help of AI, WIRED writes.
The experiment used a relatively affordable Go2 robot, priced at $16,900 — cheap by robotics standards, but sophisticated enough to perform construction site inspections and security patrols.
The most impressive fact: Claude managed to make the four-legged robot go around obstacles and find a beach ball — a task that the team of programmers could not solve.
«We suspect that the next step for AI models will be to start interacting with the world and have a broader impact on it,» Logan Graham of the Anthropic team told WIRED. «That will really require models to interact more closely with robots.»
This breakthrough is particularly significant because it demonstrates the evolution of large-scale language models beyond simple text generation. Claude didn’t just write code—he automated an entire robotic workflow, creating intuitive interfaces and solving navigational problems that had stumped experienced researchers. According to Anthropic’s analysis, AI-supported teams showed «more positive moods and less confusion» compared to their counterparts working without AI.
