Unlock: Data Contamination and Evaluation

When training data overlaps test benchmarks, model scores become meaningless. Types of contamination, detection methods, dynamic benchmarks, and how to read evaluation claims skeptically.