Few-Shot Example Curation — Select the Best Prompt Examples

I will curate a high-quality few-shot example set for your prompt — selecting examples that maximise coverage of your input distribution, represent edge cases, demonstrate the output quality standard, and avoid introducing unwanted biases in model behaviour. Few-shot example selection is not obvious — examples that look good individually can collectively bias the model toward specific output patterns that are unrepresentative of the full input distribution, and examples that are too similar provide less information than a diverse set covering different aspects of the task.

The curation covers analysis of your input distribution, coverage-based example selection strategy, edge case inclusion, negative example identification (examples showing what not to do), example formatting optimised for few-shot learning, and ablation testing confirming the selected examples improve performance versus zero-shot.