AI and LLMs · Middle · Question 54 of 75
What should evals for an extraction feature (e.g. invoice to JSON) measure?
- 1Whether the JSON looks pretty when printed
- 2Field-level accuracy against labelled examples
- 3How creative the extracted values are
Show the answer
Correct answer: Field-level accuracy against labelled examples
Explanation
Per-field precision and recall show which fields fail, such as dates or totals, guiding prompt or schema fixes.