Introduction to Prompt Structure

Generative AI models, like large language models (LLMs), rely on structured prompts to generate accurate and context-appropriate responses. Understanding how to structure prompts effectively is key to making the most out of these advanced models. In this lesson, we will explore the three critical components of prompt structure:

1. System Role

2. User Prompt

3. Assistant Prompt

By mastering these components, you will be able to craft clear and effective prompts, leading to more reliable outputs from LLMs.

1. System Role: Setting the Context

The System Role is a hidden instruction that guides the behavior of the LLM. Think of it as a backstage direction that informs the model about its role and the context it should operate within. These prompts are not visible to the user but play a crucial role in shaping the LLM's responses.

Example:
Imagine you are using a chatbot to simulate customer support for an internet service company. Your system role might be:

• "You are a customer support representative for X Service Company."

By incorporating a well-crafted system prompt, you ensure that the model understands the context, guiding it to produce relevant and professional responses.

Key Points:

• System roles are hidden but essential for guiding the model.

• They help establish the context and behavior expected from the LLM.

• Well-defined system roles significantly enhance the quality of responses.

2. User Prompt: Initiating Interaction

The User Prompt is the input provided by the person interacting with the LLM. It can take various forms, including questions, requests, or statements. The purpose of the user prompt is to direct the LLM toward a specific task or response.

Example:
Continuing the customer support scenario, a user might input:

• "Compose an email to a customer who has reported issues with their internet connection."

In this case, the user prompt clearly indicates the desired outcome: generating an email response.

Key Points:

• User prompts are explicit inputs given to the LLM.

• They guide the LLM in producing a specific output or response.

• Crafting clear and precise user prompts is essential for achieving the desired interaction.

3. Assistant Prompt: Generating the Response

The Assistant Prompt is the LLM's reply to the user prompt. It is shaped by both the system role and the user's input. This output can range from providing information to generating entire documents or creative responses.

Example:
Based on the system role and user prompt from the customer support example, the LLM might generate:

• "Dear Customer,
  We understand that you are experiencing issues with your internet connection. Please follow these steps to troubleshoot the problem: [insert steps]. If the issue persists, contact our support team for further assistance.
  Kind regards,
  X Service Company Support Team"

Key Points:

• Assistant prompts are the responses generated by the LLM.

• They reflect the model's interpretation of the system role and user prompt.

• Effective assistant prompts address the user’s needs while maintaining the context.

Bringing It All Together: The Complete Interaction

When combined, these three roles create a seamless interaction between the user and the LLM:

• System Role: Defines the LLM's context and behavior.

• User Prompt: Specifies the task or request.

• Assistant Prompt: Delivers the output based on the system role and user input.

By thoughtfully structuring each component, you enhance the clarity and relevance of the LLM’s responses.

Activities:

Activity 1: Prompt Crafting Exercise
Imagine you are using an LLM to create social media posts for a restaurant.

1. Write a System Role prompt that sets the LLM as a social media manager.

2. Create a User Prompt asking for a post promoting a new dish.

3. Predict the Assistant Prompt the LLM might generate.

Activity 2: Scenario Practice
You are working with an LLM for technical support in a software company.

1. Draft a System Role prompt for a tech support representative.

2. Write a User Prompt asking for assistance with a software installation issue.

3. Analyze the expected Assistant Prompt and discuss how it could be improved.