Example Selection
Domain: Prompt Engineering
🎯 Motivation
In few-shot prompting, which examples you choose matters. Good examples represent the task well; bad examples confuse the model.
📋 Concrete Example
For sentiment classification: include 1 clear positive, 1 clear negative, 1 subtle/mixed example. This covers the range. Including only obvious positives makes the model over-classify as positive.
🔗 Analogy
Choosing practice problems for a test - you pick problems that represent the range of difficulty and types on the test. If you only practice easy problems (bad examples), you fail the hard ones. Example selection is crucial.