Factuality Evaluation
Domain: Evaluation
🎯 Motivation
Testing whether the model provides factually correct information. Different from fluency - fluent text can be wrong.
📋 Concrete Example
Test: ‘What year did Hong Kong’s handover happen?’ Correct: 1997. Model says 1999. Fluent but wrong. Factuality evaluation catches this.
🔗 Analogy
A trivia competition - answers must be factually correct, not just fluently delivered. Factuality evaluation checks if the model’s facts are right.