AI EncyclopediaTechniques & methods
Prompt Engineering
The practice of designing the instructions and examples given to a language model to reliably get the output you want, without changing the model itself.
In plain English
It is the skill of asking well. The same model can give a mediocre answer or an excellent one depending entirely on how the question is framed, what examples are given first, and whether it is asked to show its reasoning - prompt engineering is the discipline of doing that deliberately rather than by accident.
Technical explanation
Core techniques include zero-shot prompting (the model performs a novel task from instructions alone, relying on pre-trained knowledge), few-shot prompting (concrete input-output examples are included in the prompt to demonstrate the task), and chain-of-thought prompting (the model is guided to produce intermediate reasoning steps - rationales - before its final answer, which measurably improves performance on multi-step reasoning tasks). None of these alter model weights; they only shape the input.
Why it matters
Prompt engineering is the lowest-cost lever for improving an AI system's output and the first thing any team should try before fine-tuning or building custom infrastructure - it requires no training run and can be iterated in seconds.
Real-world example
Chain-of-thought prompting - asking a model to 'think step by step' before answering - was shown to significantly improve accuracy on arithmetic and multi-step reasoning benchmarks compared with asking for a direct answer.
Common misunderstanding
That prompt engineering is a fixed, one-time skill applied identically across models. Prompts that work well on one model family often transfer poorly to another with a different training regime - it is closer to an empirical, trial-and-error craft than a portable formula.
Something wrong here?
Every entry is hand-researched and hand-written by Nakoda. If a fact is stale, a source has changed or a definition needs sharpening, tell us and we will check it.