11 of 12
Robotics
Robotics is the same story with a body. Looking at the glass is not enough. You have to hold it without knocking it over. The world does not pause for a retry the way a chat box does.
Warehouse picking and factory arms already work inside cages designed for them. Homes do not work that way, because homes were not designed as factories. Teams that skip the jump from a simulated kitchen to a real one end up with a video. Teams that pay for that jump end up with a product in a narrow cell.
Why did language pull so far ahead?
Text on the web is free to collect. Picking up a glass without knocking it over costs time, hardware, and sometimes the glass. Vision and language trained on what the web was already making. You cannot collect a billion hours of physical contact the same way.
Why isn't there a household robot yet?
Waymo can drive Market Street because the streets are mapped and the sensors are the product. Figure put humanoid robots in BMW's Spartanburg plant, loading sheet metal on the X3 line, in a cell the plant already designed.
What does simulation miss?
Simulation misses friction, lighting, a mug that is ten grams heavier, and a table that was not in the renderer. Video of people doing chores does not include the feeling of contact. Remote driving is slow and rare. Teams that skip the jump from a simulated kitchen to a real one end up with a video. Teams that pay for that jump end up with an arm in one cell.