Jack Vatcharat

Posted on Sep 27

Mastering One-Shot Prompting with Rivet: A Step-by-Step Guide

#tutorial #promptengineering #opensource #ai

Introduction to Prompt Engineering

Prompt engineering is a crucial aspect of working with AI models, particularly those developed by OpenAI. It involves crafting precise and effective prompts to guide the AI in generating the desired responses. This guide aims to provide a comprehensive overview of prompt engineering, drawing insights from the OpenAI documentation and demonstrating how to implement these techniques using the Rivet program.

Why Create This Article?

Clear and effective communication is essential when leveraging advanced AI tools. The purpose of this article is to demystify the process of prompt engineering, making it accessible to both beginners and experienced users. By providing a step-by-step guide on using the Rivet program to create one-shot prompts, this article aims to empower users to harness the full potential of AI in their projects. It serves as a practical resource, helping users understand the logic behind each step and offering examples to illustrate key concepts, ensuring they can effectively communicate with AI models.

Understanding One-Shot Prompts and Prompt Roles

One-Shot Prompt

A one-shot prompt is a single example provided to the AI to demonstrate the desired style or format of the response. It helps the AI understand how to respond to similar inputs without needing multiple examples.

Prompt Roles

System: This prompt sets the overall instructions and guidelines for the AI. It defines the style and tone of the responses.
User: This prompt represents the input or question from the user. It simulates what a user might ask or say to the AI.
Assistant: This prompt provides the AI's response to the user prompt. It shows how the AI should reply based on the given input.

Example from OpenAI Documentation

If you go to Tactic: Provide examples on the OpenAI Doc, you will see the following example (as of September 21, 2024):

The reason they use the terms SYSTEM, USER, and ASSISTANT is not just to indicate that ASSISTANT = AI. They guide us that whenever we call an API, such as gpt-4o-mini, we should not put everything in one prompt. We should create an array like below:

await openai.chat.completions.create({
    model: 'gpt-4o-mini',
    messages: [
      {
        role: 'system',
        content: 'Answer in a consistent style.',
      },
      {
        role: 'user',
        content: 'Teach me about patience.',
      },
      {
        role: 'assistant',
        content: 'The river that carves the deepest valley flows from a modest spring; the grandest symphony originates from a single note; the most intricate tapestry begins with a solitary thread.',
      },
      {
        role: 'user',
        content: 'Teach me about the ocean.',
      },
    ],
  })

The example from the OpenAI Doc shows us that the AI won't answer how to be patient and what the ocean is because we provided an example that it should answer like a poem.

However, this article focuses on how to do this using the Rivet program, which is a no-code tool. So let's get started by creating a one-shot prompt in Rivet.

Prerequisites

Before you begin, make sure you have the following:

GroqAI or OpenAI API Key: You’ll need an API key to access the AI services.
Rivet program installed: Ensure that the Rivet program is installed on your computer. You can download it from the official Rivet website.

How to Create a One-Shot Prompt in Rivet

Step 1: Open the Rivet Program

First, launch the Rivet program on your computer. If you don't have it installed yet, download and install it from the official website.

Step 2: Create a New Project

Once the program is open, create a new project. This will be the workspace where you'll build your one-shot prompt.

Step 3: Create a System Prompt Node

In your new project, right-click on the canvas and select Add => Text => Prompt to create a new prompt node.

Click the gear icon and set this node as the System Prompt.

Enter the following text: "Answer in a consistent style."

Step 4: Create a User Prompt Node

Create another prompt node.
Set this node as the User Prompt.
Enter the example input: "Teach me about patience."

Step 5: Create an Assistant Prompt Node

Create another prompt node.
Set this node as the Assistant Prompt.
Enter the desired AI response to the user prompt. For example: "The river that carves the deepest valley flows from a modest spring; the grandest symphony originates from a single note; the most intricate tapestry begins with a solitary thread."

Step 6: Create Another User Prompt Node

Create one more prompt node.
Set this node as the User Prompt.
Enter the task you want the AI to perform in this node. For example: "Teach me about the ocean."

Step 7: Create an Assemble Prompt Node

Create an assemble prompt node.

Draw lines from the system, user, and assistant prompt nodes to the assemble prompt node.

Step 8: Add a Chat Node

Create a chat node.

Click the gear icon, expand the advanced section, and set the endpoint, custom model, and headers. The following screenshot shows an example of how to configure it for GroqAI. Replace <your-api-key> with your GroqAI API key.

Draw a line from the assemble prompt node to the chat node.

Step 9: Run the Prompt

Click the Run button to execute the prompt. The output will be generated based on the one-shot prompt you created. If you followed the example text in this guide, the AI response should maintain a consistent style, similar to the initial poetic response.

Conclusion

Prompt engineering is a powerful tool for guiding AI models to generate desired responses. This guide has provided a clear, step-by-step process for creating one-shot prompts using the Rivet program. By understanding and applying these steps, you can effectively communicate with AI models, making your projects more efficient and impactful. Happy prompting!

DEV Community