Question

A team is about to invest heavily in prompt engineering for a customer-support summarization feature. Before they begin iterating on prompts, the lead asks what a well-formed success criterion for this project should look like. Which of the following best describes a properly defined success criterion?

Accepted Answer

A specific, measurable target such as 'summaries achieve at least 90% factual accuracy against a human-labeled test set of 200 examples'

Answer

A general aspiration such as 'the summaries should feel high quality and helpful to users'

Answer

A commitment to use the most capable available model regardless of latency or cost

Answer

A rule that the prompt must include XML tags and at least three few-shot examples