Question

A developer is writing a function that normalizes customer phone numbers from many countries into E.164 format. They provide Copilot with a few-shot prompt containing three examples, but all three examples use US phone numbers with the same +1 country code. Copilot's generated function correctly handles US numbers but silently mishandles UK, German, and Brazilian numbers during testing. Which prompt-crafting principle best explains this failure, and how should the developer fix it?

Accepted Answer

The examples were not representative of the input domain; the developer should include few-shot examples that span the range of country codes and formats the function must handle.

Answer

Three examples exceeded the token budget; the developer should reduce to a single US example to leave room for the model's reasoning.

Answer

Few-shot prompting is inappropriate for data-normalization tasks; the developer should switch to a pure zero-shot prompt with no examples at all.

Answer

The examples were placed in the wrong order; the developer should reorder them so the most complex US number appears last to bias the output.