Benchmark
Domain: Evaluation
🎯 Motivation
A standardized set of tasks used to compare models. Everyone tests on the same benchmark so results are comparable.
📋 Concrete Example
MMLU benchmark: 57 subjects (math, history, law, medicine…). Model X scores 85%, Model Y scores 78%. Direct comparison on the same test.
🔗 Analogy
A standardized exam like the HKDSE - every student takes the same test, so universities can compare students fairly. Benchmarks are standardized exams for AI models.