AI-103: Developing AI Apps and Agents on Azure

Few-shot learning involves embedding a small number of example input-output pairs directly within the prompt so the model can infer the desired pattern for the current task, without any model retraining or external knowledge storage. This directly satisfies the requirement of demonstrating correct classifications in-context while keeping the model and its weights unchanged and without persisting data outside the prompt. Zero-shot learning provides no examples, chain-of-thought focuses on eliciting step-by-step reasoning rather than demonstrating labeled examples, and RAG relies on retrieving and injecting external stored knowledge at query time, which conflicts with the 'no permanent knowledge storage' requirement.