hard20 pts · first answer countscost-optimisationfew-shot-prompting

Adding 30 more examples

Adding 30 more few-shot examples raises accuracy from 91% to 92% but doubles input tokens for every request at high volume. What is the best way to decide?

A.Always choose the more accurate promptB.Always choose the cheaper promptC.Remove all examples to save as much as possibleD.Compare the value of the 1% gain against the extra cost and latency, and try cheaper options such as better-chosen examples, dynamic example selection or prompt caching

Discussion 0

Join free to post in the discussion.

Adding 30 more examples

Discussion 0

Related questions