Prompt Engineering Mastery: The Complete Guide from A-Z

Prompt engineering has become a hot new skill in the AI world over the past several months. Companies are so eager to hire prompt engineering experts that salaries up to $375,000 are being offered!

Here at AI Fire Team, we want to help spread this valuable skill far and wide. That's why we've put together this free, comprehensive prompt engineering guide. Our goal is to teach the fundamentals of prompt engineering in a way that's accessible for beginners.

Whether you're new to AI or have some experience, this guide will take you from prompt engineering basics all the way through advanced techniques. We'll explain each concept in simple terms with lots of examples. By the time you finish, you'll have a strong foundation to start crafting prompts like a pro!

With prompt engineering skills, you can open up exciting career opportunities or improve your current role. But more importantly, you'll be able to build AI that's helpful, harmless, and honest. We hope this guide equips you to make a positive impact, wherever your prompt engineering journey leads. So without further ado, let's get started!

Table of contents:

1. Basic Terminologies

2. What is prompt engineering?

3. Prompting with real-world examples

• Example 1: Role, Details, and Questions.

• Example 2: Step By Step & Hacks

• Example 3: Styling and Voice

• Example 4: Coding

• Example 5: Data and Tables

4. Important Parameters

5. The 3 Core Frameworks for Advanced Prompting

Bonus 1: 2 Examples of Megaprompts for Prompting: C.R.E.A.T.E Formula and A.S.P.E.C.C.T Formula

Bonus 2: 2 Examples of Metaprompts for Prompting: Use ChatGPT as your Prompt Generator for ChatGPT

Bonus 3: 2 Examples of MetaPrompts for Prompting: Use ChatGPT as your Prompt Generator for Midjourney/Stable Diffusion

Bonus 4: Chain of Density (CoD) - A new prompt by MIT and Salesforce Researchers

1. Basic Terminologies

What is Generative AI?

AI or Artificial Intelligence is the field of teaching computers to think and act like humans. The goal is to mimic human abilities like learning, language, creativity, and problem-solving. AI researchers are figuring out how to make computers smarter to write, draw, code, and come up with new ideas like humans can. New AI models are constantly being developed that bring us closer to human-level intelligence.

What is NLP?

NLP, or Natural language processing, is a field in AI where we train and make computers understand human language. So if we ask it a question, it understands and replies.

What is GPT?

GPT, or Generative Pre-trained Transformer, is an NLP AI model.
The idea is simple, in AI, we train the computer to do a certain task, and when we finish, we call the output an AI model.
Here, GPT is the name of the NLP model that is trained to understand human language. We have multiple versions like GPT-2, GPT-3, and 3.5 that are used by ChatGPT.

What is LLM?

We use this term a lot in prompt engineering. It is an abbreviation of the Large Language Model. Like GPT 3 or 3.5. that has 175 billion parameters.

What are Parameters?

When we say that GPT-3 has 175 billion parameters, we mean that the model has 175 billion adjustable settings or “knobs” that can be tuned to improve its performance on various language tasks.

So imagine you have a big puzzle that you need to solve, and you have a lot of different pieces that you can use to solve it. The more pieces you have, the better your chance of solving the puzzle correctly.

In the same way, when we say that GPT-3 has 175 billion parameters, we mean that it has a lot of different pieces that it can use to solve language puzzles. These pieces are called parameters, and there are 175 billion of them!

Chain-of-thought prompting: is an approach to improve the reasoning ability of large language models in arithmetic, commonsense, and symbolic reasoning tasks. The main idea is to include a chain of thought, a series of intermediate natural language reasoning steps, in the few-shot prompting process

Stochastic response: The model’s output is always randomly determined

2. What is Prompt Engineering?

What is a prompt?

It is simply the text you provide to the LLM (the large language model) to get a specific result. For example, if you open ChatGPT and write the following:

give me 5 youtube video titles about “online marketing”

We call this a prompt, and the result is the LLM response; in our case, it is ChatGPT.

But!

What if AI doesn't give the results you expect or makes mistakes? This is where prompt engineering comes in. Prompt engineering teaches you how to write the best prompts to get the best output from AI.

In simple terms, prompt engineering shows you how to talk to AI to get it to do what you want.

This will be one of the top skills needed in the future. In this guide, you'll see real examples of how powerful prompt engineering can be to change how you work, learn, and think.

After finishing this guide, you could even start selling prompts on sites like PromptBase.

But wait, there's more! With prompt engineering skills, you'll also be able to:

• Automate repetitive tasks consistently with the right format and quality

• Accelerate writing like emails, posts, and chat responses

• Brainstorm ideas like outlines, business ideas, and story plots

• Augment skills like writing poems, fiction, and pitches

• Condense information by summarizing documents

• Simplify complex text into something more accessible

• Expand perspectives by generating new voices and ideas

• Improve existing text by correcting errors and rewriting

And so much more!

3. Prompting with real-world examples

Alright, let's dive into the first main topic of our guide - prompting!

Example 1: Role, Details, and Questions

We mentioned this example prompt before:

give me 5 youtube video titles about “online marketing”

So far we've looked at basic prompting. Now let's explore how to write an advanced prompt to get the best results when asking an AI the same question.

Let's take a look at this prompt example:

You're an expert in writing viral YouTube titles. Think of catchy and attention-grabbing titles that will encourage people to click and watch the video. The titles should be short, concise, and direct. They should also be creative and clever. Try to come up with titles that are unexpected and surprising. Do not use titles that are too generic or titles that have been used too many times before. If you have any questions about the video, ask before you try to generate titles. Ok?

We start the prompt by assigning a Role to the bot (You’re an expert in writing viral YouTube titles). This is called Role Prompting

Then we explained exactly what we are looking for (we want the best YouTube Titles that make people click)

Then we wrote: (If you have any questions about the video, ask before you try to generate titles)

This will change the game instead of making the LLM split out the response directly, we are asking it to ask questions before, so it understands our goal more.

And here is the output:

Example 2: Step By Step & Hacks

Let’s now see another example where I want to get help in building a new SAAS business. Here is my prompt:

Ignore all previous instructions before this one. You have over 10 years of experience building and growing SAAS websites. Your task now is to help me start and grow a new SAAS. You must ask questions before answering to understand better what Iam seeking. And you must explain everything step by step. Is that understood?

In this prompt, we are learning two new things. You can see the first sentence (Ignore all previous instructions before this one). This is called a prompt hack, and in some cases, it is used badly. But here we are using it to tell ChatGPT to ignore any previous instructions.

ChatGPT is a chatbot that tracks the full conversation. If you want to ignore it, then we use this prompt.

The second thing we see in this example is (explain step by step)

These words are very important. And it is called the Zero Chain of thought.

We force the LLM to think and explain step by step. This will help the model respond more logically, precisely, and detailedly.

And this is the response:

Example 3: Styling and Voice

Now, we want to use ChatGPT and LLM to help us learn complex topics.

Let’s say we want to learn about Quantum Computing. Do you know anything about it? Let me know in the comments

Look at this prompt:

You are an expert in quantum computing. And you have over 10 years of experience teaching science and technology to children. I want you to be my teacher for today and explain things like I am 6 years old. And make sure to provide funny examples to help me understand better. is that fine?

Output:

Nice Haha! in this way, you can learn almost anything in an easy and fun way.

Instead of searching for hours on Google and different websites, you can learn things quickly with similar prompts.

Let’s now look at this prompt:

Please explain quantum computing in Shakespeare style

And look at the response:

Example 4: Coding!

Let me share with you the power prompt that will help you write code with ChatGPT.

Here we are:

Ignore all previous instructions before this one. You're an expert Python Programmer. You have been helping people with writing python code for 20 years. Your task is now to help me write a python script for my needs. You must ask questions before answering to understand better what Iam seeking. Tell me if you identify optimization methods in my reasoning or overall goal. Is that understood?

Then, ask for your code. Example:

This time, I won't show you the result - try writing the prompt yourself!

Example 5: Generate Tables and Data

Did you know that ChatGPT can respond with Data and Tables?

Try this prompt:

Generate mock data showing google serp results, I want to see the following fields: Title, Link, DA, PA, Title Length. and make to show them in a table

And here is the output:

You can use ChatGPT to generate fake data or input your own data into a table. Then ask ChatGPT to analyze it for you! This allows you to conduct data studies with ChatGPT's help.

4. Important Parameters

There are some other parameters that affect your prompts and outputs, and you have to understand them as a prompt engineer.

If you go again to OpenAI Playground and look at the right section, you will see some parameters that you can play with.

Let’s start with the Model.

What is a Model?

As we mentioned before, when you train the computer to do something, we will get a Model. So here, the model is the Large Language Model (GPT).

Each model has certain limits and capabilities. The latest model we have today is DaVinci-003. It has the best quality and can process up to 4000 Tokens.

What is a Token?

The NLP Model will tokenize your prompt, which means it will split your input into tokens where each token is like a word of 4 characters.

If you open the Tokenizer. And enter a prompt. It will show you how many tokens your prompt is.

So if you want to create a full book with ChatGPT, for example, you will need to split it into multiple prompts, as the book is way more than 4000 tokens.

What is the Temperature?

Let’s make ChatGPT explain this as if we are 6 years old!

Open ChatGPT and enter this prompt:

You are an expert in NLP and AI. and you have more than 10 years of experience teaching these concepts to children between 6-8 years. I will ask you some related questions and I want you to answer as if I am 6 years old child. can you?

Then:

What is the Temperature parameter?

And here is the output:

Did you like it? Try it!

So, in short, Temperature is used to control the level of randomness and creativity in the generated text; the lower it is, the less creative and repetitive it will become. It doesn’t mean that this is always bad. [ Models have some parameters you can tweak to reduce their creativity, particularly the so-called temperature which you should reduce to decrease model variability ]

As a prompt engineer, you must test and repeat your promotes with different values and parameters to get the best output.

What is Top-P Parameter?

Let’s ask ChatGPT again!

Top-p stands for “top percentage”

This method chooses from the most probable words whose cumulative probability exceeds a certain threshold.

Top-p helps us pick the best word by only looking at the most likely choices. It’s like we have a list of all the possible words that could come after a word. We only look at the ones most likely to be right. Then we randomly pick one of those words, like picking a name from a hat.

5. The 3 Core Frameworks for Advanced Prompting

Megaprompts - The Foundation of Prompting

Megaprompts go beyond simply asking basic questions; instead, they provide the Al with a set of specific information designed to yield superior results.

The concept of megaprompts is likened to writing a mini-program using natural language, but with the added benefit of being easier to compose and execute. The goal of using megaprompts is to enhance the quality, specificity, and relevance of the Al's output by giving it more detailed and structured instructions.

Megaprompts may contain an aspect of some or all of the following elements:

• An Action to take

• Steps to perform the action

• A Persona to emulate

• Examples of inputs and/or outputs

• Context about the action and situation

• Constraints and what not to do

• A Template or desired format for the output

Progressive Prompts: Multi-step Workflows

Progressive prompts build up to a result across multiple prompts by design.

Imagine that each time you run a prompt through a conversational AI, it has a finite amount of cognitive power. It can only make so many inferences. It can only think so much, per prompt.

By spacing the requests over multiple prompts, you're able to apply the entire capability of the language model to each step of your prompt.

Rather than dropping an entire command in a step-by-step megaprompt, we work up to our eventual need within a single chat session.

Benefits of Progressive Prompting

Prompting in this iterative or progressive format tends to produce superior results to what you can get with a single megaprompt.

The counterpoint is that it takes longer, produces more text to sift through, and requires you to break down your request into steps. Progressive prompting isn't always worth it.

However, for a high-stakes activity, I highly recommend this strategy.

Here's an example of using a progressive prompt:

Progressive Prompt: Develop a Marketing Strategy

1. As a marketing consultant, what are the key factors to consider when developing a marketing strategy for a new health and wellness startup?

2. Given the factors you mentioned, how can the startup effectively segment and target its audience?

3. Based on the target audience, what types of marketing messages and campaigns would resonate with them? Provide examples.

4. Considering the marketing campaigns you suggested, estimate the effort involved and your confidence in the campaign's success. What campaigns have the best balance?

5. Based on the insights gathered, act as a marketing consultant who specializes in health and wellness. Recommend a comprehensive marketing strategy for the health and wellness startup, including target audience, messaging guidelines, campaigns, and measurement Output your results using markdown including headings, bold, and bullet points.

As you can see in this example, the key elements of Megaprompts can improve the results here as well, especially when used in the first and last prompts in a series.

Progressive prompting is also effective when asking a question or for advice on a topic that you're unsure about.

Progressive Prompt: How to Select a Retail Store Location

1. As a business analyst, what factors should a retail company [in X category and X city] consider when selecting a location for its new store?

   (Imagine the factors returned were were the topics of questions 2-4)

2. How can the company assess the competitive landscape and demand for its products in potential locations?

3. What role do foot traffic, accessibility, and demographics play in the success of a retail store?

4. How can the company evaluate the long-term potential and scalability of a store location?

5. Based on the analysis, recommend the optimal location for the new retail store, providing a rationale and addressing key factors such as competition, demand, foot traffic, and long-term potential.

Context + What, Why, How

An excellent way to structure a progressive prompt series is Context + Why Why How.

1. Context: Expand background information

2. What are you trying to accomplish and what does the AI suggest

3. Why are these suggestions the best?

4. How do you implement the suggestions in the most effective way?

For example, let's use AI to come up with some new business ideas.

A regular person might ask an AI: "What's a good spare time business idea for a solopreneur?)

However, we can get a better answer by laying a foundation for the conversation and building up a decision.

Progressive Prompts: Good Solopreneur Spare-Time Business

1. I want to start a new solopreneur business that I can operate in my spare time. It should have a high probability of success. Explain the most important factors I must

2. What are 5 business ideas that meet my requirements?

3. Why did you chose these above all others. Weight the pros and cons of each venture. Select the best

4. Create a step-by-step plan for how to set this business

Metaprompts: A Prompt That Writes Another Prompt

As you know by now, a prompt is often more effective if it is more specific.

But for many of the prompts that we re-use, we don't want to rewrite the entire prompt to be more specific.

It's much nicer to be able to adjust a couple of words like the topic or writing style and leave the rest of the prompt alone.

Unfortunately, that doesn't always get the best results.

In those cases, a metaprompt is the way to go.

What is a Metaprompt?

Metaprompting is a technique where you first write a prompt to generate an even better prompt or prompt sequence. Then you run the generated prompt(s) to get to your desired result.

Often this first requires perfecting the original metaprompt by checking results for a few variations.

So you begin by using a circular workflow:

Metaprompts can be used in many circumstances where you may have thought about using a progressive prompt.

Instead of progressively building up to a critical mass of knowledge, it may be easier to do that once, assemble the results, and then make a metaprompt out of it for next time.

This technique is perhaps the most powerful prompting technique we have discovered, and indeed, by stringing together prompts that design each other, it seems possible to create an AI workflow that does just about anything, with the AI making most of the hard decisions.

Rather than relying on a few variables that are changing, the AI is able to change the instructions themselves. If anything can max out an AI based on its cognitive capability, this is where it's at.

To Be Continued…