AI and LLMs · Middle · Question 29 of 75
What is a risk of using model-generated synthetic test data only?
- 1It can miss real user language and edge cases
- 2Synthetic data can't be used in any evaluation
- 3It always contains copyrighted content
Show the answer
Correct answer: It can miss real user language and edge cases
Explanation
Synthetic cases are a fast start. Mix in real, anonymized queries and known failures to keep evals realistic.