A development team intends to adjust an LLM's default communication style, shifting it from a highly academic tone to a more casual, user-friendly voice, appropriate for a mobile application interface. They have ten sample conversions (from original formal text to desired informal text) demonstrating this desired writing style shift. Which prompting method is OPTIMAL for achieving this stylistic change?
Answer & explanation
Correct answer: B
WHY B is correct: Few-shot prompting with 3–5 carefully selected (formal → casual) example pairs directly shows the model the transformation pattern. The model infers the mapping — formal input produces informal output — and applies it to new inputs. Showing is more reliable than telling for style specification.
WHY NOT A: Zero-shot 'be casual' works partially but lacks precision. Without examples, the model may apply an inconsistent interpretation of 'casual' that doesn't match the brand's exact desired voice.
WHY NOT C: Chain-of-thought prompting is designed to improve accuracy on complex reasoning or math tasks by externalizing reasoning steps. It is not a style-transfer technique and adds unnecessary verbosity.
WHY NOT D: RAG retrieves external knowledge at inference time. Style-transfer examples are small, known in advance, and fixed — embedding them directly in the prompt (few-shot) is simpler and more effective than storing them in a vector store.