A brand-new stormwater fee FAQ goes live before any curated answer examples exist. Which prompting regime is appropriate to start?
Select an answer to reveal the explanation.
Short Explanation
Day-one FAQs often have instructions and policy docs but no gold answer museum yet—that is classic zero-shot territory. Fake few-shot examples or an animal-control one-shot just teach the wrong pattern until real stormwater exemplars show up.
Full Explanation
When a brand-new stormwater fee FAQ goes live before any curated answer examples exist, zero-shot prompting is appropriate: rely on clear instructions plus retrieved policy text without pretending curated gold exemplars already exist. That matches day-one municipal FAQ reality where ordinances and fee schedules exist but a polished answer museum does not. Few-shot prompting that fabricates gold answers teaches possibly wrong patterns and misrepresents readiness to residents and auditors. Mandatory full fine-tuning on fabricated FAQ pairs before any resident question is heavy, slow, and polluted by synthetic labels that may contradict the fee ordinance. One-shot prompting from an unrelated animal-control script transfers the wrong tone and policy domain into stormwater billing guidance. Exam caveat: zero-shot still needs strong retrieved policy grounding and escalation paths; “zero-shot” does not mean “no documents,” and high-impact fee answers may still require human review. Operational check: launch with policy retrieval, collect real Q&A for later few-shot or evaluation sets, avoid promoting synthetic exemplars into the prompt without SME ratification, and track escalation rates on fee disputes during the first billing cycle.