The Power of Prompt Engineering

Introduction:?

Prompt engineering has emerged as a pivotal technique in leveraging the capabilities of modern language models, allowing users to interact more effectively and obtain desired outputs. While many people are aware of ChatGPT and other LLM chat interfaces, most leverage them using simple question answer type prompts, not really understanding some of the true power that LLMs can provide.

In this article, we will delve into the concept of prompt engineering, explore the differences between one-shot and few-shot prompts, discuss user versus system instructions, introduce the Role, Action, Steps, Context, Examples (RASCE) framework for constructing effective prompts and illustrate the impact pf leveraging these techniques on output generation using a practical example.

What is Prompt Engineering??

Prompt engineering refers to the process of designing input instructions or queries that guide language models in generating desired outputs. It involves formulating precise and structured prompts that elicit the desired information or responses from the model. Effective prompt engineering enables users to harness the power of language models and achieve more accurate and tailored results.

One Shot vs Few Shot Prompts:?

One Shot Prompts:?

One shot prompts provide a single input instruction or query to the language model, expecting it to generate the desired output. These prompts are suitable when the task at hand requires a straightforward response and the desired output can be generated with minimal context or additional information.

Few Shot Prompts:?

Few shot prompts involve providing multiple examples or demonstrations of the desired output to the language model. By exposing the model to a few instances of the desired behavior, it can generalize and adapt to generate accurate responses even for complex tasks or nuanced requirements. Few shot prompts enhance the model's ability to understand and learn from a wider range of inputs.

Demonstrating One Shot vs Few Shot:?

To illustrate the impact of one shot vs few shot prompts, let's consider a hypothetical scenario: creating a personalized recipe recommendation system.

One Shot Prompt: "Generate a recipe for a vegetarian pasta dish."

Few Shot Prompt: "Generate two recipes for vegetarian pasta dishes: 1) A creamy pesto pasta with roasted vegetables and 2) A spicy arrabbiata pasta with tofu

By providing a few shot prompt with specific examples, the model gains a better understanding of the desired output and can generate recipes that align with the given context. This enables the system to generate more accurate and tailored recommendations compared to a single one shot prompt.

User vs System Instructions:

Prompt engineering involves defining both the user instructions and system instructions to guide the language model effectively.

User Instructions:?

User instructions convey the desired information or input the user wants to obtain from the language model. They specify the nature of the task or the information sought. For example, "What are the main causes of climate change?" or "Summarize the plot of the novel 'Pride and Prejudice.'"

System Instructions:?

On the other hand, system instructions describe how the desired output or response should be generated by the language model. They provide guidance to the model on how to structure the output or the format in which the information should be presented. For instance, "Provide a detailed explanation with key points" or "Present the summary in bullet points."

The user instructions focus on the input, conveying the user's intent or the information they seek, while the system instructions focus on the output, providing guidance on how the desired output should be generated or presented.

By effectively leveraging System Instructions, Language Models can be much more powerful in assisting in tasks. One of my "go to" System Instructions working in Operations is to tell the LLM to provide an output in CSV or JSON format. Being able to have the flexibility to specify formats just through adjusting prompts is very powerful.

By carefully crafting both user and system instructions, prompt engineering enables users to effectively communicate their intent and expectations to the language model, resulting in more accurate and tailored responses.

The RASCE Framework for Constructing Prompts:?

To construct effective prompts, the RASCE framework can be employed:

• Role: Define the role the model should adopt, such as a poet, writer, or advisor.
• Action: Specify the action or task the model should perform, like generating a poem, summarizing a text, or providing recommendations.
• Steps: Outline the necessary steps or subtasks to guide the model through the desired process.
• Context: Provide the relevant context or constraints to ensure the generated output aligns with the given context.
• Examples: Include examples or demonstrations of the desired output to train the model's understanding and improve accuracy.

By following the RASCE framework, prompts can be constructed in a structured and comprehensive manner, enabling more effective communication with the language model.

Putting it all Together

Let's return back to the previous example of the vegetarian pasta recipes to demonstrate all of the concepts discussed in this article in a single prompt.? I like this example because I think it illustrates really well not only the content returned, but also the flexibility in driving the format.

_________________________________________________________

Role: Recipe generator

Action: Generate two vegetarian pasta recipes

Steps:

1. Introduction: Provide a brief introduction for each dish
2. Ingredients: List the required ingredients in bullet points
3. Instructions: Outline the step-by-step instructions for preparing each dish using numbered lists
4. Notes: Include any additional notes or variations for each recipe

Context: Emphasize the use of common ingredients and easy preparation methods

Examples: Specify two recipes - creamy pesto pasta with roasted vegetables, and lemon-garlic linguine with sun-dried tomatoes

System Instructions: "Do not use the Example recipes, but instead choose two unique recipes. Present the recipes in Markdown format, ensuring that each recipe includes the following sections: Ingredients, Instructions, and Notes. Use bullet points for the ingredients and numbered lists for the instructions. Provide a brief introduction for each dish."

_____________________________________________________________

And below are the results! It's important to note that I simply copied and pasted the output from ChatGPT, all of the formatting was done by ChatGPT based on the System Instructions that were provided.? If you want to see it for yourself simply copy and paste the above prompt into ChatGPT and check out the results.

Recipe 1: Spicy Arrabbiata Pasta

Introduction: Spicy Arrabbiata Pasta is a classic Italian dish known for its fiery tomato sauce and robust flavors. This pasta recipe is perfect for those who enjoy a bit of heat in their meals. With simple ingredients and a quick preparation method, you can have a delicious and satisfying vegetarian meal on the table in no time.

Ingredients:

• 8 ounces (225g) penne pasta
• 2 tablespoons olive oil
• 4 garlic cloves, minced
• 1 small onion, finely chopped
• 1/2 teaspoon red pepper flakes (adjust to your spice preference)
• 1 can (14 ounces/400g) crushed tomatoes
• 1/4 cup fresh basil leaves, chopped
• Salt and pepper to taste
• Grated Parmesan cheese (optional, for serving)

Instructions:

1. Cook the penne pasta according to the package instructions until al dente. Drain and set aside.
2. In a large skillet, heat the olive oil over medium heat. Add the minced garlic and chopped onion. Sauté for about 2-3 minutes until the onion becomes translucent.
3. Add the red pepper flakes to the skillet and sauté for an additional minute.
4. Pour in the crushed tomatoes, season with salt and pepper, and stir well. Simmer the sauce for about 10 minutes to allow the flavors to meld together.
5. Add the cooked penne pasta to the skillet, along with the chopped basil leaves. Toss everything together until the pasta is well coated with the sauce.
6. Cook for another 2-3 minutes, stirring occasionally, until the pasta is heated through.
7. Serve the spicy Arrabbiata pasta hot, garnished with grated Parmesan cheese if desired.

Notes:

• Feel free to adjust the amount of red pepper flakes according to your preferred level of spiciness.
• You can add other vegetables like sautéed bell peppers or zucchini to the sauce for added flavor and texture.

Recipe 2: Creamy Mushroom and Spinach Pasta

Introduction: This creamy mushroom and spinach pasta is a comforting and satisfying vegetarian dish that combines earthy mushrooms with the freshness of spinach. The creamy sauce adds richness and balances the flavors beautifully. With its simple ingredients and creamy texture, this pasta is sure to become a favorite among both veggie lovers and pasta enthusiasts.

Ingredients:

• 8 ounces (225g) fettuccine pasta
• 2 tablespoons butter
• 8 ounces (225g) mushrooms, sliced
• 2 garlic cloves, minced
• 2 cups fresh spinach leaves
• 1 cup heavy cream
• 1/4 cup grated Parmesan cheese
• Salt and pepper to taste
• Fresh parsley, chopped (for garnish)

Instructions:

1. Cook the fettuccine pasta according to the package instructions until al dente. Drain and set aside.
2. In a large skillet, melt the butter over medium heat. Add the sliced mushrooms and minced garlic. Sauté for about 5 minutes until the mushrooms are golden brown and tender.
3. Add the fresh spinach leaves to the skillet and sauté for an additional 2 minutes until wilted.
4. Pour in the heavy cream and stir well to combine. Let the cream simmer gently for about 3-4 minutes, allowing it to thicken slightly.
5. Add the cooked fettuccine pasta to the skillet, along with the grated Parmesan cheese. Toss everything together until the pasta is evenly coated with the creamy sauce.
6. Season with salt and pepper according to your taste and cook for another 2 minutes until the pasta is heated through.
7. Serve the creamy mushroom and spinach pasta hot, garnished with fresh chopped parsley.

Notes:

• You can add a squeeze of lemon juice to the sauce to brighten the flavors.
• Feel free to experiment with different types of mushrooms, such as cremini or shiitake, for a variety of flavors.
• For a healthier option, you can substitute half of the heavy cream with vegetable broth or milk.
• If desired, sprinkle some additional grated Parmesan cheese on top before serving.

_________________________________________________________

Prompts For an Operations Use Case

Here is one final example that is more relevant to how someone in Sales Operations would want output. Leveraging an API to an LLM will allow you to take the result set and take further programatic action on it.

___________________________________________________________

Role: You are a data analyst

Action: Research the top 10 towns in Morris County New Jersey in terms of population

Steps:

1. Provide the list in table format

2. The table should have the following columns: Town, County, State, Population

System Instructions: You should only respond with CSV Output. No other commentary or output should be provided.

__________________________________________________________

Town,County,State,Population

Parsippany-Troy Hills,Morris,NJ,53,238

Morristown,Morris,NJ,19,728

Randolph,Morris,NJ,25,234

Denville,Morris,NJ,16,635

Rockaway Township,Morris,NJ,24,376

Montville,Morris,NJ,22,635

Roxbury,Morris,NJ,23,699

Mount Olive,Morris,NJ,29,165

Hanover,Morris,NJ,16,694

Jefferson,Morris,NJ,20,949

__________________________________________________________

I'm still learning new and unique approaches everyday. Let me know in the comments if you have different approaches or ideas.