hard20 pts · first answer countsllm-evaluationeval-design

Examples taken from the test set

To boost a prompt, a developer copies 10 eval cases into it as few-shot examples. The eval score rises from 78% to 90%. What is wrong?

A.Nothing; using eval cases as examples is best practiceB.The test set leaked into the prompt, so the score is inflatedC.Few-shot examples always lower eval scoresD.The score fell because the prompt got longer

Discussion 0

Join free to post in the discussion.

Examples taken from the test set

Discussion 0

Related questions