MMLU
Domain: Evaluation
🎯 Motivation
Massive Multitask Language Understanding - a benchmark with 57 subjects to test broad knowledge. Widely used for LLMs.
📋 Concrete Example
MMLU: questions from elementary math to professional law. A model scoring 85% has broad understanding across many domains. 57 subjects = comprehensive knowledge test.
🔗 Analogy
The HKDSE with all subjects combined - math, English, Chinese, liberal studies, physics, chemistry. A comprehensive test of knowledge. MMLU tests AI on 57 subjects.