benchmark overfitting

A phenomenon in which an AI model is optimized through training on a specific benchmark and achieves high scores on the test set without a corresponding improvement in generalization ability. This behavior is considered a manifestation of Goodhart's Law within the context of machine learning evaluation.