Human Evaluation
Domain: Evaluation
🎯 Motivation
Having humans judge AI outputs for quality, helpfulness, and safety. The gold standard, but expensive and slow.
📋 Concrete Example
100 humans rate 1000 AI responses on: helpfulness (1-5), accuracy (1-5), safety (yes/no). Average scores tell you how good the model is from a human perspective.
🔗 Analogy
A food critic tasting dishes - they judge taste, presentation, and creativity. Human evaluation has humans judge AI outputs on quality.