AWS Certified AI Practitioner (AIF-C01) Study Guide

Chapter 7. A Guide to Prompt Engineering

From 2019 to 2020, with the launch of OpenAI’s GPT-2 and GPT-3 models, the AI community discovered that the effectiveness of these systems could be greatly improved by the wording of prompts. This led to the emergence of a new field: prompt engineering. When OpenAI released ChatGPT in late 2022, interest exploded, and prompt engineering became a widely sought-after skill.

The term engineering can be misleading, as the practice is often more of an art than a science, requiring iterative tweaking to achieve the desired response.

Understanding prompt engineering is essential for the exam, as questions will test your ability to recognize how instructions, context, input data, and output impact the performance of an FM. You’ll also be expected to identify the appropriate prompting techniques—like few-shot, zero-shot, and chain-of-thought—in various scenarios. Additionally, security risks such as prompt injection, model poisoning, and jailbreaking are important to assess your knowledge of safe and responsible AI use.

The Anatomy of a Prompt

A prompt can be any length, so long as it is within the limits of the context window. But you can break it down into four components:

Instructions

What you want the model to do

Context

Background information to help the model understand what you want it to do

Input data

Specific data you want the model to process to generate the response

Output indicator

The output type or format for the response

Let’s ...