AI and LLMs · Middle · Question 31 of 75
Why is snapshot testing of exact LLM outputs usually a poor fit?
- 1Outputs change slightly between runs, causing noisy failures
- 2Snapshot tools don't support JSON files
- 3Snapshots can't store long text from model answers
Show the answer
Correct answer: Outputs change slightly between runs, causing noisy failures
Explanation
Check properties, schemas or judged criteria instead. Exact snapshots can work for deterministic code around the model.