Does Scaling Toward AGI Depend on What System-1 Lets System-2 Execute?

Progress toward artificial general intelligence has largely relied on scaling System-2 through larger models, more data and compute, longer reasoning, and more agents. A study suggests the next stage may come from co-designing with System-1, a lightweight, semantically aware predictor. This involves a predict-gate-execute-update loop where System-1 estimates utility and System-2 executes admitted candidates.