AI Evaluation & Benchmarks

How AI systems are measured: capability benchmarks like MMLU, HumanEval, and SWE-bench, human-preference arenas, LLM-as-a-judge grading, RAG and agent evaluation, and the limits behind every leaderboard score.