Analogical Prompting: Definition and Examples

Analogical Prompting is an advanced prompt engineering technique introduced by researchers at Google DeepMind in 2023. It relies on a fundamental cognitive principle: humans often solve new problems by drawing on similar situations they have already encountered. This method asks the language model to generate analogous problems and their solutions before tackling the target problem.

Unlike classic few-shot prompting, where the user manually provides examples, Analogical Prompting lets the model produce its own relevant examples. The model identifies structurally similar problems, solves them step by step, then transfers the reasoning to the given problem. This self-generation of analogies exploits the model's internal knowledge in a more targeted and contextual way.

The process generally unfolds in three phases: first, the model recalls or constructs analogous problems; second, it details the resolution of these analogies; finally, it applies the identified reasoning patterns to the original problem. This approach proves particularly effective for tasks involving mathematical reasoning, programming, and complex problem-solving.

Analogical Prompting represents a significant evolution because it combines the advantages of zero-shot (no need to provide examples) with those of few-shot (benefiting from concrete examples to guide reasoning). Studies show it often outperforms standard Chain-of-Thought prompting on reasoning benchmarks like GSM8K and MATH.