Few-Shot Example Formatting

Present few-shot examples either explicitly labeled as examples, or folded in as previously solved tasks the model believes it already completed.

Two ways to present in-context learning examples in a prompt:

Explicit examples — label them as such: "In the following, when I encounter a question like 'Who was the first President of the United States?' I will give an answer like 'George Washington.'" Simple to write and unambiguous to the model about what it's looking at.
Integrated as prior solved tasks — fold examples into the document as if they were previously solved tasks, rather than labeled examples. This requires more careful formulation but lets the model leverage the examples more naturally, producing a smoother prompt. It's especially effective in ChatML or conversation-transcript settings: the model is made to believe it already solved earlier tasks in the demonstrated style, which encourages it to continue that successful approach into the current task rather than treating the examples as a separate, detached reference section.

Sources