Inside the Robot Data Factory: Mecka's Josh Gao on Sensorizing Humans to Train Robots
Robotics AI is fundamentally data-limited, lacking the free internet corpus LLMs relied on.
“we don't have the luxury of a 'cold start' that LLM had thanks to the internet”
Mecka AI founder Josh Gao explains that robotics training data must be physically collected via custom sensors and wearables, unlike LLMs which bootstrapped on free internet text. The company builds and operates data factories that capture human movement (pixels, actions, logic) to train foundation robotics models. This highlights a structural bottleneck in physical AI development that has no analog in the LLM scaling story.