The AI model Claude Opus 4.7 outperformed engineers by tens of times in controlling a robot dog.
The development of language models continues to surprise even the most seasoned market participants. Anthropic's latest experiment under the Project Fetch initiative demonstrates where the industry is heading: artificial intelligence is beginning not just to assist people, but to fully take over complex physical tasks.
The key test result — the Claude Opus 4.7 model handled the setup and control of a four-legged robot 20 times faster than the best team of human engineers did a year earlier. Moreover, the neural network operated almost autonomously, under minimal researcher supervision.
The range of actions performed by the AI is impressive: connecting to video sensors and lidar, writing a manual control program, creating a robot motion trajectory monitoring system, and configuring an object recognition algorithm. All of this was done without real-time human involvement.
Comparative analysis shows just how significant the progress is. Compared to teams using older AI versions, Opus 4.7 proved to be 18 times faster. And relative to people working without chatbot assistance, it was 37 times faster. Notably, the code written by the neural network turned out to be 10 times more compact than human-written code, indicating its high efficiency and optimization.
An important nuance: Anthropic did not implement specialized algorithms for robot control. As the experiment's authors note, this progress was a side effect of the general scaling of language models. This confirms my long-standing hypothesis that universal AI systems can master new domains without targeted training, simply through increased computing power and data volume.
However, there were limitations. Claude still struggles with precise physical manipulations. The model successfully guided the robot to its target, but could not gently push a ball to the exact spot. Such a task requires complex real-time feedback — in this area, humans still outperform AI.
Anthropic believes the industry is entering an era of "physical AI agents," where neural networks will work with hardware as effectively as they currently do with software code. In my view, it's only a matter of time: once models learn to process sensory feedback at the same speed as text, the boundaries between the digital and physical worlds will finally blur.