Generated Knowledge Method

Welcome back! In previous lessons, you learned how to make your prompts more effective, use system prompts to guide the model, and even fetch extra context from outside sources. Today, we will focus on a specific multi-step prompting approach called the generated knowledge method.

Sometimes, a single prompt is not enough to get the best answer from a language model. This is especially true when the question is complex or when the model needs more background information to answer well. The generated knowledge method helps by first asking the model to create some extra context or background knowledge and then using that information to answer your main question.

This approach is practical when:

• The task is complex and needs more than one step to solve.
• The model might not have all the details it needs in the first prompt.
• You want to guide the model in thinking more deeply before answering.

Let’s explore how this method works and how you can use it in your prompts.

The generated knowledge method is a way to help language models perform better by splitting your request into two steps:

1. First, ask the model to generate some helpful background information or context.
2. Then, make your main request.

For example, if you want the model to write a Python script that interacts with a specific API, you might first ask it to explain how to make a general API call to that service. Once you have that information, you can then ask it to write the script you need.

This two-step approach helps the model build the knowledge it needs, leading to more accurate and useful answers.

Let’s walk through how to use this method step by step.

Suppose you want to write a Python script that sends a prompt from a text file to an Amazon Bedrock model. If you ask for the script immediately, the model might miss important details or make mistakes.

Instead, start by asking a general question to help the model generate the background knowledge:

In general, how does one make a proper API call to Amazon Bedrock models with Python?

This prompt asks the model to explain the basic process. The model might respond with an overview like:

• You need to use the AWS SDK for Python (boto3).
• You must authenticate with AWS credentials.
• You call the Bedrock API with the right parameters.

Why do this?
By starting with a general question, you help the model “think out loud” and lay out the steps or requirements. This makes it less likely to skip important details in the next step.