3124837eca10486bd96ce3a05bf5af0c by 2026

Author:2026
Language: eng
Format: epub

Chapter 3 prompt engineering teChniques

How Prompts Are Written

The techniques and practices of prompt engineering are ever evolving as new ideas are tested and tricks are uncovered. Communicating directly in this way with a model is new to all of us, and we’re still developing an understanding of what works and what doesn’t.

Indeed, we are hesitant to present here an in-depth guide to prompt engineering in a field that is evolving so quickly; we are at risk of being obsolete as new models and bespoke techniques become available. However, it is necessary to present at least a foundation of skills for the practitioner to become familiar with as the difference in performance from one prompt to the next can vary significantly for similar work types and the effort required to optimize performance varies considerably, especially for less powerful models. Prompt engineering is chiefly a field driven by experimentation and creativity, and we encourage prompt engineers at all levels of experience to review the prompts written by their peers.

Before we present the model agnostic techniques of prompt engineering, it is important to consider the goal of the application. We can break this down into three work types for any particular prompt, noting there will always be some overlap between the three:

• Perform tasks with no additional context in the prompt

• Perform tasks from additional context provided in the prompt only

• Perform tasks with additional context that rely on the model’s

interpretation

Some examples of the first work type might be for a simple request for code or Excel formula, the generation of examples of a certain style, or simply an explanation or definition of a phrase. Examples of the second work type cover document Q&A and summarization use cases, where we explicitly do not want additional context to appear.

Examples for the third bullet are those that typically require reasoning from the LLM, for example, suggested action, reinterpretation of a text, or code correction.

Another key consideration for the goal of a prompt includes the desired output structure, for example, is the application a pure text output or is it embedded in a user interface (UI) or other web application. We find in practical applications that for almost every query to the LLM, a JSON style output is generally preferred, even for internal LLM

calls that do not communicate to the front end.

77

Chapter 3 prompt engineering teChniques

Careful consideration of the goal of the application will indicate the degree of effort we’ll require for prompt construction and provide some limitations on the structure of the prompt. Ultimately, we must arrive at a prompt format that allows for rapid experimentation backed up by rigorous techniques for performance evaluation. We defer the discussion on evaluation to Chapter 4 and here focus on solely on constructing the prompts.

We group the prompt techniques in the following discussion into two parts: the first, universal prompt techniques are techniques that can and should be employed when writing any prompt, while the techniques in the case study below are more prompt paradigms or methodologies to apply for a specific use case.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.