Three stories last week form a single narrative about AI entering the physical world:
Daxiao Robotics became the fastest unicorn in embodied intelligence. Their Kairos world model (4B parameters) topped four global benchmarks, beating 28B-parameter competitors.
Guangxiang Technology shipped the Phi-Bot X1 industrial robot, running 21.5 continuous hours on a NIO welding line with zero errors. Deployment time: one week.
A modified Unitree G1 robot summited Ecuador's 6,200m Chimborazo volcano. Next target: Everest.
Same road, three markers: AI moving from digital to physical isn't about bigger models — it's about harder problems. Torque feedback, tactile sensors at -20°C, vision reliability in dust.
Huawei's SpaceMind scored 70.6 on Fei-Fei Li's VSI-Bench (close to human average 79%), using only RGB input. Tsinghua's UniCM hit Nature Machine Intelligence, pushing ENSO prediction to 19 months.
The real story isn't "embodied AI is hot" — it's that AI is undergoing a paradigm shift from language to physics. These robots don't need more parameters. They need more screws.
Full bilingual analysis on the blog.
This article first appeared on my tech blog
Top comments (0)