問題文
A document pipeline has three steps: detect the document type, extract fields, and resolve conflicts between extracted values and the existing record. The first two are high volume and mechanical; the third is infrequent and requires judgment. What is the most cost-effective arrangement?
選択肢
- Use the strongest grade for detection since an early mistake propagates, and the cheapest for the rest.
- Use a fast, inexpensive grade for detection and extraction, and a stronger grade only for conflict resolution, since grade can be chosen per step rather than per application.
- Use the cheapest grade for all three steps and add more examples to the conflict resolution prompt.
- Use the strongest grade for all three steps to keep the pipeline consistent.