The future of AI: System One and System Two
Fast, calibrated judgment for almost everything, slow reasoning only where it matters.
Inspired by Kahneman, we separate fast judgment from slow reasoning. A System One model classifies in tens of milliseconds with calibrated probabilities; above a confidence threshold it acts, below it hands over to a reasoning model or a person. The threshold becomes a measurable business lever, and the average cost drops because the expensive model is used only on complex cases.
Comments (0)