Our readingWhat is a korrent?
If a model answers every rollout correctly all the time, training might as well stop, because there is no learning signal left.
Drawn from what Sebastian Raschka said
Subjectreinforcement learning
What this subject means
reinforcement learning Training by reward: environments, verifiable tasks, value functions, and whether it works or merely beats what came before.