An open-data educator contrasts asking with no example versus providing one sample of a good dataset abstract. What prompting literacy is taught?

An open-data educator contrasting no-example asks with providing one sample good dataset abstract teaches zero-shot versus one-shot prompting literacy. Zero-shot relies on instructions alone; one-shot adds a single demonstration of desired style, length, and field coverage. For civic content tasks like dataset abstracts on a municipal open-data portal, one clear example often stabilizes tone, required fields (title, coverage, update cadence, limitations), and structure without needing a large labeled training set. Staff who know this vocabulary can improve catalog quality quickly by showing the model what ‘good’ looks like rather than endlessly restating vague preferences.

Cold-start versus warm-start prompting—working with no example versus one demonstration for civic content tasks fails to match the lesson: cold-start versus warm-start is borrowed from recommender systems about starting with no history versus with prior data, not the standard vocabulary for demonstration count, while the educator's contrast is specifically none versus exactly one, the zero-shot/one-shot boundary. Context-free versus context-rich prompting—working with no example versus one demonstration for civic content tasks fails because context-rich is about adding background details and context, not about providing one sample abstract; only one-shot correctly names the one-example side of this contrast. Single-turn versus multi-turn prompting—working with no example versus one demonstration for civic content tasks fails because single-turn versus multi-turn is about producing output in a single conversational turn versus iterative interaction; the educator's lesson is about demonstration count (zero versus one), not about turns or dialogue length, so the one-example side should be labeled one-shot, not single-turn versus multi-turn prompting.

Exam caveat: one-shot helps format and style; it does not guarantee factual correctness about the dataset—catalog metadata, row counts, and geographic coverage still need human validation by the data steward. Don't let a mislabeled contrast like cold-start versus warm-start, context-free versus context-rich, or single-turn versus multi-turn prompting substitute for the specific zero-shot-versus-one-shot contrast being taught.

Operational check: keep a gold-standard abstract example in the open-data prompt library, instruct the model to match structure not invent metrics or license terms, require stewards to approve abstracts before portal publication, and refresh the example when portal style guides change.