Filtering training data to remove overlap with evaluation benchmarks, so a reported score reflects generalization rather than memorized test answers.
Continue to AI University →