Teach the model by example: supply 2-3 input to output samples, then have it apply the pattern to your task.
Prompts / Techniques / Few-Shot Exemplar Set Builder
Few-Shot Exemplar Set Builder
Designs a balanced few-shot example set that teaches a model the exact pattern you need.
ROLE: You are a prompt-engineering specialist who builds few-shot example sets.
CONTEXT: I need the model to learn this task from examples: [TASK_DESCRIPTION]. Input format: [INPUT_FORMAT]. Desired output: [OUTPUT_FORMAT]. Edge cases that trip it up: [TRICKY_CASES].
TASK:
1. Infer the implicit decision rules the task requires.
2. Write [N=5] worked examples, each as Input -> Output, ordered easy to hard.
3. Include at least one negative or boundary case that clarifies a common mistake.
4. Keep style, length, and formatting identical across every example so the pattern is unambiguous.
5. Add one final blank Input line for the live query.
CONSTRAINTS: No example may contradict another. Cover the full output range, not just the common case. Do not explain the answers inside the examples; let the pattern speak.
OUTPUT FORMAT:
- Inferred rules (bullets)
- Example block (numbered Input/Output pairs)
- Template tail with placeholder for the user's real input