Definition
An AI safety approach that focuses on supervising and evaluating the intermediate reasoning steps and computational processes of an AI system, rather than just its final output.
In short: An AI safety approach that focuses on supervising and evaluating the intermediate reasoning steps and computational processes of an AI system, rather than just its final output.