ChatGPT & Generative AI: Prompt Engineering for Business

1. [Chapter 1] Meet Davis, Your Prompt Engineering Instructor: Yeah, yeah, yeah. Hey, I'm Davis and I've taught almost 1 million people around the world technology and business skills so far. People use AI to create social media posts or proof read an essay. But that's like hiring a rocket scientist to screw in a light bulb with AI. As your partner, you're only limited by your prompt engineering skills. In this course, you're going to learn the ICO framework, how to architect a professional grade prompt. Four essential prompt, design skills, advanced prompt engineering techniques. Then we're going to build a professional grade prompt together step by step. And then you're going to get a hosted and verifiable prompt engineering certificate issued by easel. Now throughout this course, you're going to learn some key definitions associated with AI models and prompt engineering. You're going to take some simulations that give you real time feedback and build your understanding for the course outline and helpful definitions and citations and further reading links. You can refer to the study guide at easel link I guide. Now this course is designed especially for professionals. Using AI at work to make you more productive. You can use your prompts to do things like personalize anything, for anyone to write code, create something, strategize, teach, and so much more, so many things that before only humans could do. Now the good news for you is that it's been proven that when you have prompt engineering skills, like the ones you're developing here, you can get results from a general use AI model like those offered from all the large AI companies today that outperform even those results that came from custom trained models on a specific area. All right, so let's get started. 2. [Chapter 1] Prompt Engineering: The Power Behind the AI Business Revolution: Ask AI something simple and you'll get simple results. But when you can create detailed prompts that carve out a perfect little space among the trillions of possible outcomes. Your prompt engineering skills will start saving you time, lots and lots of time, and you'll be in a new enhanced professional reality Prompt. Engineering is the practice of designing questions or statements in a way that guides an AI model to provide the desired response. We use short prompts in everyday AI use, but you can build elaborate prompts involving lots of data and details that enable you to take advantage of the immense power of AI models and get more meaningful results. Prompt engineering skills can create a serious competitive advantage for you, enabling you to uncover the most accurate, relevant, and creative responses from AI models. So in summary, in the world of generative AI and large language models, prompts are what we use to get AI generated outputs. Think of a prompt as an elaborate digital key that based on its design, unlocks a response from an AI model. Prompt. Engineering is designing questions or statements that guide an AI model to provide the desired response. Prompt. Engineering skills can be used to dramatically increase the accuracy, relevancy, and or creativity of an AI models response. 3. [Chapter 2] Learn about Context Windows, AI Instructions, and the Most Powerful Statement: I might ask my friend Sean. Sean, what should I have for lunch? Maybe in reply he says, hmm, I don't know, Maybe chicken taco, right? But if I were to work with AI on this lunch decision, I might ask. I, I want you to help me figure out what I'm going to eat for lunch. Be my nutritionalist and help me find a lunch solution that's within walking distance. Now for breakfast, I had a scone and usually I try to eat one meal a day only that has gluten in it. Also, I'm trying to eat more vegetables, especially green vegetables like brussels sprouts. Also, I went on a really long run this morning, so I'm pretty hungry and you should know that I don't like things that are too spicy and also I'm allergic to avocado. Would you mind suggesting two lunch options that you think would work for me? Specifically, can you recommend two dishes, each from a different restaurant and tell me the name of the restaurant that makes it? Then the AI might respond. Okay. I'd suggest the Power Bowl from industry or the hippie bowl from Fresco. Neither have bread and both have brussel sprouts and lean proteins, plus they're both tasty. And both restaurants are within walking distance. Now one of the big benefits of working with AI is that it can synthesize enormous amounts of information, much more even than is in this example. When work with AI, your job really becomes asking the right question, Which is perhaps even the more important job than generating the response. Once a CEO of one of the most successful management consulting firms in the world told me that the right question is more powerful than any statement. Now here are some definitions you'll need in this section. A context window is the amount of information that an AI model is able to consider when generating a response. Instructions in prompt. Engineering is an optional set of directions used to calibrate how the AI model interprets your prompt. Now it's time for you to learn how to use the ICO, Prompt Design framework. 4. [Chapter 2] AI Prompt Architecture: Become a Prompt Engineering ICOn!: When you chat with AI behind the scenes, a record of your conversation is being used by the AI to determine what to say next. For example, if you say what's the smallest country by population, the AI is going to look at your question, your prompt, and do its best to deliver an appropriate response to you. It responds with the Vatican City with 510 inhabitants. If you then follow up with, okay, what's the second smallest and the AI? It's with a population of 10,876 How does this work? Well, when you chat with AI, you're building context that the AI can use to continue the conversation and deliver useful results to you. In this example, the context is your chat history. For example, you don't have to say what's the second smallest country by population because in your first message, you added that context to the discussion. When you asked country by population, context is the information that the AI uses to generate a more useful response for you in context doesn't have to be a chat history. You can add context through your prompts, and your prompts can include lots of information. The context window is the amount of information that an AI model is able to consider when generating a response. In most cases, it's extremely large in terms of text. Some models have context windows that can handle hundreds of pages worth of text. That's where advanced prompt engineering comes in. Your prompts can be elaborate and detailed. They can include information that enables an AI model to deliver completely personalized results. Introducing the ICO prompt design framework, use the ICO framework to organize your approach to AI. Prompts. Start off with instructions. Instructions are optional sets of directions used to calibrate how the AI model interprets your prompt. Instructions can be used to set the role of AI. You're a dietician set rules don't include more than three sentences in a paragraph condition. The style of the response provide analogies to simplified, complex topics, established boundaries, don't provide any medical advice. For example, it might make sense for you to use one set of instructions for prompts that you use to generate content and another for research work and another set for doing AI assisted analysis. After instructions add context to your prompt. This is all of the background information and data that you add to give the AI model what it needs to generate the result you're looking for. The two primary kinds of contextual information to use in your prompts are, first, background information, for example, information about you, an organization, a client, your industry, some policy or a situation. Second, examples. These are samples that will teach the model something like how you write what's good, what's bad, what's worked in the past, what hasn't now. Finally, after instructions and context, you can finish your prompt by telling the AI how you want the output to be formatted. This is optional, but if you're using AI for your work, a consistent output format can be meaningful. In the output part of your prompt specify exactly how you need the AI to respond. For example, tell AI to create a title, then a subsection with three bullet points, then another subsection with a four sentence recommendation, and then a final subsection with three sub subsections. Once you set this up, your output is dynamic. Each time the AI generates, it can fill out the form you've laid out using new information. In the case that you changed something in your prompt, maybe in your context. So for example, let's say that Rahul writes a list of instructions, then he adds a lot of context, then identifies how the output should look, he generates a result, then he changes a piece of data that's in the context part of his prompt and then he regenerates the output structure. Would stay the same way as in the first generation but would have different content because Rahul changed that piece of data in his context. So in summary, your AI prompts can contain huge amounts of information. Use the ICO framework to help you organize your prompts. Start with instructions, an optional set of directions used to calibrate how the AI model interprets your prompt. Then add lots of context to your prompt. Background information about the topic, and examples that will guide the AI. Finally, define an output format. This enables you to use your prompt with different data and keep a predictable structure. 5. [Chapter 2] Voices From the Field: What contextual information do you add to your prompts?: You know, we breed insects on food waste. But what we're doing is implementing AI with that development so that we can actually now be more productive form and the way that we breed the insects and how we breed them and temperature controls and the food waste that we use. And so it's just a lot of the little small processes because with just the insect from egg development to large larvae, there's already seven processes with each stage of development. I think that you start with very much like a biography. So you want to tell it if you're using it for business, you want to tell it about your business, what your business does, what kind of products you sell. So the AI has context. Same thing, if you're a student. You want to tell it that you're studying machine learning or computer science and that you write term papers or you do research. One of the best ways to JI is to give it examples. I've given an example of my writing right in the prompt. So if I wanted to do a copy, help me write my newsletter. I would tell it what I wanted to do and then I give it an example. If I give it one example, that's called one shot prompting. If I give it multiple examples, it's called shot prompting. And the limiting factor is the size of the context window. So, so based on the work that I do write, a lot of times I don't get to go out, so a lot of times I'm on the system sivingugh information, right. Information as regards how to sell a particular product to a target audience, basically. Right insight that I relevant to whatever it is I'm trying to build or trying to see or trying to speak to. Maybe bringing it back a bit, maybe a bit of market research. So like understanding the offerings that I have. Understanding, you know, something about a client or whatever I've collected about a clients. Just because I'm into like brainstorming a lot. So understanding, you know, a particular interest I have for me being like a sports professional with like all the athletes and stuff that I like coach at a very competitive level. But the more back background information that you give to the AI system, especially about the athletes, it's like what they trained, what they ate, a lot of the testing batteries and stuff that we do. Essentially what a testing battery is, you can just think up a testing battery as pretty much just a whole sequence of exercises that we straight coaches like give our athletes to see how they're performing. What you want to do is you want to take all of that information that you're putting into the spreadsheets about how your athletes and stuff are like performing. And then you want to plug it into the AI model. And plug it in as like background information, just context. Then based on the information that you pretty much collected and stuff from the tests and everything else about the athletes, then you can start asking more questions. 6. [Chapter 2] Interactive Learning Simulation: Let’s Build an ICO-Formatted AI Prompt: Okay, now you have the opportunity to participate in an interactive learning simulation that will enable you to practice building an ICO formatted prompt that's based on a real world scenario. Head to easel links and run through the learning simulation. You can even get an instant certificate of completion for the simulation after you finish. 7. [Chapter 2] Key Takeaways from the AI Prompt Design Section + What’s Coming Next: Congratulations on finishing this section. Let's do a quick review of what you've learned as an overview. I want you to think of a prompt as the most powerful comprehensive request for an answer that one could ever make. Now here are some key takeaways. A prompt can include instructions, context, and detailed output instructions for the AI model to follow. I, C, O instructions can be used to set the role of AI, set rules and boundaries and condition the general style of the AI's response. Your prompts context can include background information, info about you, your organization, your clients, your industry or a situation, and examples to teach the model about something. Think of the output section of your prompt as creating a form that after the model considers the earlier parts of your prompt, it fills out for you. Next, you're going to learn four prompt design skills, delimeters, definitions, markdown, and handlebars. Let's go. 8. [Chapter 3] An Introduction to Delimiters, Definitions, Markdown, and Handlebars: A act. So I like to cook. And there are some tools that I use in my kitchen that make my life a lot easier, like the stand mixer. It's a fantastic tool that makes my muffins, breads and cookies so much better and a lot easier to make. Now, when you're designing prompts, you have tools like this that will help you with your AI, like my stand mixer helps me with my cooking. Let me introduce you to a few concepts. A delimter is a sequence of characters used to specify the boundary around something in your prompt. A definition in prompt engineering is using a word or phrase to refer to something that you've added to your prompt. Markdown is a simple readable way of formatting plain text. Handlebars in AI prompting is the use of two curly braces with text on the inside that creates a space for the AI to fill in each time it generates. In this section, you're going to learn how to level up your prompt engineering skills using these tools. 9. [Chapter 3] How to Use Delimiters and Definitions to Build Magical Keys: Often you need to tell AI where to look. For example, this is where you'll find the policy and here's where you'll find relevant examples. Excellent prompts will often have many such points of reference in them. I'm going to teach you how to use delimeters and definitions in your prompts. Together there are prompt engineering superpower. A delimeter is a sequence of characters used to specify the boundary around something. In your prompt, you really only need to know how to use one demeter tags. You can use tags as a way of separating or chunking up information. For an AI like this information and you put your information here, slash information here. You just need to keep your tag structure consistent. Put the same word, any word in between the greater than and less than symbols. And the second tag with the same word as the opening tag. There, you need to put a backslash. You're closing the tag. You can use tags for a little bit of information like lyrics and then put some lyrics here and then close the tag or a lot of information like report. And then put a whole 30 page report here and then close the report tag. Now definition definitions are simple. In prompt engineering, a definition is using a word or phrase to refer to something you added to your prompt. Here's some example, simple definitions. Client background, Davis Jones is a teacher and software engineer. He likes cooking and music organization easel. After adding these definitions to your prompt, you can just say, consider the client background, capital C, capital B. Or in your output section, you could use handlebars, which you'll learn about elsewhere, like this organization, capital the AI will know what you're talking about. Now, for larger chunks of information, like a 15 page policy, for example, you'll want to add this to your prompt by telling the AI that you're going to add some information and you're going to refer to this information later by a given word or phrase. This is where we use delimeters and definitions together like this. In between the policy tags, you'll find a policy that governs my industry. Hereafter the policy with a capital P policy, and then you add the text from the policy here, Backslash policy. Now whenever we need to refer to that policy in our prompt, we can just use a capital P policy and the AI will know what we're talking about. Here are a few examples of how this might work. Please analyze the policy, then do something using what you find in the policy based on the examples in the policy. So in summary, a delimeter is a sequence of characters used to specify the boundary around something in your prompt. The most common delimeters are tags. Tags, or any word between two inequality symbols. The closing tag needs a backslash in front like this. A definition in the context of prompt design is using a word or phrase to refer to something you added to your prompt. Use delimters and definitions together, this enables you to refer to large chunks of information easily in your prompt. 10. [Chapter 3] Markdown: How You Add Information Hierarchy to Plain Text: Large language models are a form of AI that converts the text in your prompt, ultimately into ones and zeros. Then the AI model looks for patterns and its training data. This ultimately enables it to make a prediction. And that's your output When your prompt is pre processed and you've broken down into these little bits, it's all plain text, it's not formatted. So how do you indicate titles and things like that? Through markdown. Markdown is a simple readable way of formatting plain text. There are a number of ways to use markdown, including creating headers and subheaders, bold text, italic text, and many more. In prompt engineering, you'll most often need markdown for headers. Basically, to show AI hierarchy in your information, we use the hashtag symbol for headers. And the approach is simple. One hash is header one, the biggest header. Two, hashtags are the second biggest header, three hashtags are the third biggest header, and so on. When you're building a prompt and you want to add information about your organization, for example, you'll likely want to use markdown headers to give your information. Some structure like this. Hashtag, information about my company, Inc hashtag hashtag, our history, hashtag hashtag, our team. Because of your use of markdown headers, the AI understands that, for example, our team is a section that's a subsection of information about my company, Inc, which is the title. You'll also use markdown headers to tell the AI how to structure its output. Let's say that you want the AI to generate a weekly report for you using some data, and this data changes weekly. You've called this capital W weekly capital D data in an earlier part of your prompt. Now you're going to add a specific output structure to this prompt. Your prompt says draft a report for me using this format. Hashtag, weekly report here, generate three sentences summarizing the results you see in the weekly data hashtag. Hashtag progress on our goals Here using the weekly data generate one sentence that summarizes which goals we reached this week. Then in a second sentence, summarize which goals we did not reach because you've added markdown headers to the output part of your prompt. You'll guide the AI to generate a weekly report draft that's formatted exactly how you'd like it to be with a large title that says Weekly Report with the appropriate text underneath then a subsection title that says Progress on our goals with two sentences underneath it. In summary, most AI models we interact with are large language models, or LLMs, designed to work with plain unformatted text. Markdown is a simple readable way of formatting plain text. We often use its hashtag based approach to formatting headers. In markdown one hashtag indicates the biggest header, two, the second biggest header, and so on. You can use markdown headers to structure your prompts, contextual information, and to guide output formatting. 11. [Chapter 3] Let’s Ready a Document for AI work Using Delimiters, Definitions, and Markdown: Okay, let's take a document that we want to use with AI and format it so that it's clean and AI is ready to work with it. You're just looking at an empty note here in this prompt management app that I built. But what we're going to do here could be done on any document management tool like Google Docs or Microsoft Word. If you keep your prompts in there or you want to design your prompts in there or just straight into like a AI interface like Chat, GPT or anything like that. This is what we're going to be doing here. I'm going to open this expository scoring guide. This is just from a prompt that I built earlier. What this is, this is the State of Texas Assessments of Academic Readiness, which is this educational test that public school students have to take. This is this scoring guide from a couple of years ago. What we want to do here is take the scoring guide and make it ready for use with AI so that we can integrate it into a prompt. Now what we're going to do here, skills wise, can be applied to many different types of documents and information. It could be applied to information about your company, or you, or a client, or a situation, or an article, or pretty much anything that's text based. All right, so what we're going to do here, let's just do this section right here. I'll show you how we're going to use markdown, delimeters and definitions here in structuring this content. All right, what I'm going to do here to start with is just paste in the raw text. Right now we have this text without any formatting, which is how AI models need text. Ai models don't support formatted text when you're using large language models. By default, basically we've just got this raw text. How can we make this text formatted in a way that AI is ready to use it? Well, I find often that what you want to do here is tell the AI that you're going to give it some information. A good way to think about this is you're creating a boundary around some information. And we're going to do that with D limiters, something like this. Let me just go through what I'm doing here a little bit. I'm saying in between the guide tags, you're going to find information about a scoring guide. And hereafter, this is the guide with a capital G. By saying here, after the guide, the things that we put inside the guide tags, we'll be able to be referenced throughout our prompt by just saying the guide with a capital G. Then what we're going to do here is open our delimitor like this. Then I'm just going to go ahead to the bottom of this and I'm going to just close the tag by doing backslash and then guide. All right, basically anything that is in between these tags is going to be the guide with a capital G that we can use throughout our prompt. All right, this is actually the title of the guide. Now we're going to start using some markdown to give our information some hierarchy. With markdown, we do one hash tag to represent the biggest title or title one. Then we would do two hashtags to indicate title two or the second biggest title, something like that, like score 0.1 What I want to do here is just go back and make sure that I'm mirroring what is going on here with the actual scoring guide. Yes, here's the title Then Basically what we do here is we have this subtitle and then we have these sections which are like sub subtitles. These are like the third section down of this guide. Let's go and create that with markdown, this part here where it says the essay represents a very limited writing performance. You see that this is basically describing what score 0.1 means. This is basically like body text that's associated with score 0.1 Then this organization progression thing, this is a new section right here. Let's go ahead and do this. This is a subsection of score 0.1 Then each of these are different bullet points basically. We'll just go ahead and make those all bullet points like that. All right. And I'm just going to double check that we have three bullet points. 123, make sure that that mirrors what we have here. Yes, 123. All right, great. What we've done by adding three hash tags to this organization progression. Sub subtitle is, we are in between our scoring guide tags saying that here's the title and here's one subsection and then here's a sub subsection. Basically what we're saying is that this is how you know how to score the organization and progression of something that is score 0.1 I'll show you a full version of this so that you can see how this looks. I'll just navigate over here to the full prompt. Great, here's how this is done in practice. I did the exact same thing when I built this out. For some teachers, I call this scoring guide instead of guide. But I took this entire expository scoring guide, which I found is just public information. Then I just structured it here, going down, and just basically converted it from this PDF into something that's usable for a prompt. I'll show you through a real life example how I use this scoring guide, capital S, capital G. You'll notice down here in the output section of the prompt, where we get to the bottom and I basically say, hey, okay, so the students submitted an essay, et cetera. You're going to see here I say, I want you to help me grade the essay. This is referring to the essay that's above in the prompt according to the scoring guide. Now what I'm telling the AI model is that you're going to use the scoring guide again, which is inside our tags. It knows where to look to grade the essay. Here is where the teachers put the essay. That's an example of using delimitors definitions and markdown to convert something like this. That is not a very useful piece of context because it's a PDF. It's not organized in a way that AI can work with it, and converting it in to something that is usable by an AI model using the skills that you now know about, which are these limiters, the definitions, and the markdown. 12. [Chapter 3] Handlebars: Reliable Output Formatting with Dynamic Information: In AI prompting, you can think of handlebars as a stage. The stage itself stays in the same place, but there's a different band, a different dance every night. Handlebars in AI prompting is the use of two curly braces with text on the inside that creates a space for the AI to fill in each time it generates. Handlebars are a tool that we use almost exclusively in the output section of our prompts. The simplest application is to have the AI replace the handlebar space with a word or phrase. Maybe from some information you added in an earlier part of your prompt each time it generates. For example, maybe your prompt has contextual information that includes a client's name, Like this client name, Davis Jones. Now in the output section of your prompt, you might have a little line like this proposal for client name. In handlebars, the AI would then generate proposal for Davis Jones as its output. You can also use handlebars to give AI specific instructions about what to generate in that space. Here are two examples, first a simpler one, then a more complex one. Let's start with the simple one. Generate a funny title for a business proposal that includes client name here, make it less than 60 characters and here's an actual result. Davis Jones pie charts and puns a slice of success. Now here's a more complicated example. Client Name is struggling to lower their stress levels before bed. Use your training data to generate two exercises that client name can do each night before bed to build stress reduction skills. Structure these two exercises like this hashtag exercise name details about the exercise in under 100 characters. Encouraging words for client name. In this part of the prompt, we're using Definitions, Markdown and Handlebars within Handlebars to customize the format and style of our output. Here are two actual results. Mindful breathing, inhale deeply for 4 seconds, hold for seven, exhale for eight, repeat. Calmness awaits Davis. Here you can see that Markdown created this title, and also that the nested handlebars created calmness awaits Davis. Those are the encouraging words for client name. Here's the second output, Gratitude. Reflection, List three things from your day. Positivity breeds peace. Davis, keep it up. In summary, use two curly braces, handlebars with text on the inside to create a space for AI to fill in something for you each time it generates. We typically use handlebars in the output portion of our prompts as a way of guiding AI as it generates dynamic output. The simplest use of handlebars is to have AI dynamically add a word, phrase, or number to its output, like name or score. Handlebars can give AI detailed instructions that include markdown definitions and even nested handlebars. 13. [Chapter 3] The Four Core Prompt Engineering Skills (Recap) + What’s Next: With these four skills, delimeters, definitions, markdown and handle bars, you can build prompts that do almost anything. Let's summarize what you learned in this section. Put information between tags and define it with a word or phrase. This enables you to refer to that information easily throughout your prompt. Use markdowns, hashtag based headers to give structure to your prompts, contextual information, and to style your output. Handlebars can be used to create dynamic output based on instructions you add between curly braces. These four tools complement each other, Use them to construct high quality context for the AI and highly customized outputs. All right, in this next section, we're going to learn some advanced prompt engineering techniques. 14. [Chapter 4] Peru, Mongolia, and a Diplomatic Dish Designed with Generative AI: Imagine that you're a government official and you're go at a dinner that's celebrating a historic agreement between Peru and Mongolia. The first dish comes out and it's a perfect example of a traditional Mongolian stew, but it's made with Peruvian ingredients that few people know about. It's amazing, unique, and creative. And you ask the chef, how did you come up with this dish? She answers, well, I had an AI lead a simulated collaboration session between an expert in traditional Mongolian cooking, an expert in rare Peruvian vegetables, and an experienced Peruvian fisherman. Crazy, right? Well, in this section you're going to learn some advanced AI prompting techniques like this. First you're going to learn two easy techniques that you can use to dramatically improve your results. Then you're going to learn what having AI step back is all about. And finally, you're going to learn about multi agent or SPP prompting, and the Tessa technique. Let's get started. 15. [Chapter 4] Two Easy Prompting Techniques to Improve Your Results: Think of a time that something appealed to your emotions. A person, a movie, a song. Something that inspired you to do something or think differently. Did you know that AI models respond to emotional appeals? Let's learn a couple of simple and effective techniques that will improve your AI results. The first technique is to add emotional appeals to your prompts. Researchers at Microsoft have confirmed that modern AIs are capable of understanding emotional appeals. And that adding them to prompts improves results by up to 8% based on a variety of metrics. For example, Sanjay is using AI to help him prepare for a job interview as a propulsion engineer. So in the output part of his prompt, he adds the following emotional appeal. This job opportunity isn't just a step forward in my career. It's the fulfillment of a dream I've been working towards for a long time. Then he continues. Now I want you to help me prepare for this interview by filling out the following form that will help me learn about trends in the industry. Ai models understand these emotional appeals. Second, I'm going to teach you the according to technique. Here you'll ask the AI to use specific parts of its training data when generating results. It's easy. Basically, you add a phrase like, according to data from Wikipedia to your prompt. Or similarly respond by using information from official government sources. You'll often want to add these according to phrases to your prompt in the instructions or output portions of your prompt. Here's an example. Sanjay may upgrade the prompt he was working on with. Now I want you to help me prepare for this interview by using information from peer reviewed research in your training data to fill out the following form that will help me learn about trends in the industry, according to basically tells the AI where to look for answers. So in summary, AI models can understand emotional appeals. Use them in your prompts to enhance AI's reasoning abilities and your results. Ask AI to use specific parts of its training data when generating results. For example, specific sources or types of information to view the original research on emotional stimuli and prompts visit easel dot links emotion to view the original research on according to prompting. Visit easel, link slash according to. 16. [Chapter 4] Stepping Back: From Answering Questions to Questioning Answers: Have you ever been working on a problem? And maybe you got a little frustrated, then you decided to calm down. Take a step back, and give yourself some space to think. Maybe you slept on the problem, and then you worked on it the next day, and then you had a breakthrough. Well, it turns out that AI models exhibit similar behaviors. Researchers at Google found that when AI is asked to step back, think about a topic at a high level and then move forward into some more detailed analysis. Ai models perform up to 27% better. For example, Kendra is going through a professional transition. She wants to become a nurse. And she's looking to work with AI on her career transition plan. So early in her prompt, she might ask the AI to step back, consider what it knows about trends and medicine. And then she proceeds to ask the model to help generate a career transition plan specific to her use case like this. Now take a step back and consider what you know about trends and medicine and nursing after you've done that. And then she continues, Okay, here's another example. Jenny is looking for a text strategy that will help her sort through lots of information more quickly. She might ask the AI to step back and consider how people have successfully handled information overload. Then proceed to ask the AI to make a specific technological solution recommendation like this. Now take a step back and consider what you know about information overload and turning lots of information into valuable insights now. And then she continues. Finally, remember that AI models are trained on vast datasets. Often you just need to use your prompt to have the AI recall data already in its dataset. Add phrases like using your training data or using what you know about physical therapy to your prompt to explicitly tell an AI model to bring its immense training data to bear when it generates a response for you. So in summary, when AI is asked to think about a topic at a high level, then move into detailed analysis, results can be greatly improved. This technique is called stepping back. It's especially useful when you're using AI to generate results that involve specificity. Ai models are trained on vast datasets. By telling an AI what training data to use, you're prompt can access that knowledge to read the original research. On the stepping back technique, visit easel, link slash stepping back. 17. [Chapter 4] Simulating Multiple Perspectives with the TESSA Technique: Have you ever worked with someone who looked at things a little differently than you did? You found that their different view was interesting and helpful? Well, you can basically simulate this with AI. Researchers at Microsoft have developed a prompting approach called solo performance prompting or SPP prompting. It has an AI assume multiple personas, each with a different kind of expertise or point of view, and then engage in a simulated collaboration with each of these assumed personas and then deliver its result to you. This prompting method is great for solving complex problems or generating really creative results in experiments. Ai models prompting with this approach deliver results that are quantifiably up to 20% better. In this module, I'm going to teach you how to use this technique with the Tessa framework. Let's set up an example. Let's say that Sahid is working on a marketing strategy and he needs to know how to position a brand. He needs to understand consumer attitudes across three different demographics that the brand targets. With the SPP approach, he can prompt the AI to take on personas. For example, a teenager, a working parent, and a retiree. And the AI can then simulate a discussion among these personas before generating a recommendation for Sahid. To build this prompt, shied uses the Tessa framework. Tessa is a step by step process for building this prompt like this. First you name the task, then the experts, then you start the discussion, then synthesize, then find agreement, and then get your results. So let's go through this task. First, we introduce the task to the AI through our prompt like this, I need help with the branding strategy now, experts. Here you're going to name all of the hypothetical people. Subject matter experts, audience representatives, whatever. Into the discussion like this, let's bring together people who represent different audiences. One, a teenager interested in computer games to a working parent who plays games 4 hours a week. Three, a retiree who likes technology and plays games for 10 hours a week. Then you tell the AI that it's going to be in the discussion and it's going to lead it. Now we start the discussion. You'll do this, for example, by telling the working parent to share what they're looking for from a brand like this, the teenager. To share how it would like to interact with this brand on social media. And then the retiree to share what they like about the games that they play, synthesize. Now what you'll do is tell the AI again, through your prompt to synthesize the ideas of the personas, and then generate, in this case, a branding strategy. Now, in other prompts, it will be whatever the task is and finally, agreement. You'll tell the AI to have the personas work together until they agree on, again, in this case, an amazing branding strategy for Sahid, and then deliver those results to you. And it's important to note that this prompting technique works perfectly with the ICO framework. You simply add any instructions that you have at the top of your prompt, add any context that you need to add, then add the Tessa approach to your prompt, and then have the AI deliver any output that you'd like it to deliver to you. In summary, the SPP approach enables AI to simulate multiple personas, improving its ability to solve complex problems and be creative. Use the Tessa framework to use this approach in your prompts. Task experts start the discussion, synthesize agreement. This approach works nicely with the ICO prompting structure. You can add instructions, context, then Tessa. Then an output structure to read the paper on solo performance prompting. Go to Sellin Multi agent. For an example, see easel DolinksPP example. 18. [Chapter 4] Video Quiz: Improve Your Recall of these Advanced Prompt Engineering Techniques: Imagine that you're reviewing investment opportunities and you see an opportunity that comes across your desk to invest in a Polish coffee shop chain that's raising money so they can expand to parts of Germany. You'd like to work with a general USAI model as you evaluate this business plan and you're deciding which contextual information to add to your prompt. Which of the following assumptions can you make about the AI model? The AI knows about the current coffee consumption trends in Germany. The AI has information about how Germans tend to consume coffee. The AI has been trained with multiple coffee chain business plans and related documents. The AI is capable of using source documents written in Polish, German, and English in a single prompt. The answer is that all assumptions are valid except for assumption one. It's not safe to assume the AI model has been trained with current coffee consumption data from Germany. Now you'd like to do everything you can to ensure the model returns accurate information to you. What's the best way to do this? Add an instruction telling the AI not to speculate. Use the according to method. To have the AI model leverage training data only from sources you trust. Integrate an emotional appeal like this is really important to me into your prompt or all of the above. That's right. All of these strategies are valid and they can be used in combination with one another, All right? You'd like to leverage the solo performance prompting or SPP approach when you build this prompt, you'll do this with the Tessa framework, setting the task as generating an investment evaluation. Now, which group of experts will you introduced into the prompt? Which expert will lead the synthesizing part of this prompt? An owner of a chain of European coffee shops, a European private equity fund manager, a German coffee enthusiast, and a Polish CEO leading the synthesis. An expert on German coffee culture, a coffee supply chain expert, and an investment banker with the AI leading the synthesis. An expert in German commercial real estate, an expert in marketing to German consumers, a Polish restaurant supply chain expert, and a German investment banker leading the synthesis. While all of the experts presented are valid personas to include in the exercise, only choice two has the AI leading the synthesis, making this the correct answer. 19. [Chapter 4] Recapping Your Advanced Prompt Engineering Techniques Learnings + What’s Next: All right, in this section, you learned some advanced prompt engineering techniques. Let's review them. When you include emotional appeals in your generative, AI prompts like this matters a lot to me. Ai will generate statistically better results. Use the according to technique to have AI models use specific parts of its training dataset when generating results. If you're doing AI work that involves specificity, tell the AI to step back and recall what it knows about a concept that you're using in your prompt for complex problems or synthesizing viewpoints. Use the SPP method. With Tessa task experts start the discussion, synthesize and agreement. Now, in the final section of the course, we're going to build a professional grade prompt together. At the end of the section, you can find out how to get your certificate. 20. [Chapter 5] Don’t Hire the World’s Best Chef to Come Chop Onions!: The way that many people use AI, it's like hiring a chef to come to your house to chop your onions. Maybe that's why one of the partners at Y Combinator, the most important start up accelerator in the world, based in San Francisco, had this to say. Thing I'd love to see more start ups working on is the use of LLMs to automate complex back office processes in large enterprises. So for example, in a bank, you might have a customer service team answering loads and loads of queries from customers. And people are already working on automating that. But what lots of people don't realize is that there's then a compliance team that's spot checking one on 100 of these conversations to make sure that things like complaints are handled appropriately or that financial advice isn't given if the agent isn't qualified. And that's done by a massive team of people who are going through reams and reams and reams of text. That's a really good task for an LLM. Okay, so now we're going to build a prompt designed to make a big impact in a space that you may not have thought of industrial mining equipment. Let's stretch our minds a little bit and imagine new areas where prompt engineering can make a big business impact. Here's the case. Manufacturers of industrial equipment receive many warranty claims. In each of these claims can take a human lots of time to review. Let's use AI to make this system exponentially more efficient. By designing a prompt that qualifies warranty coverage submissions for gearboxes in mining conveyance systems. Now you don't need any prior experience in this area to understand what we're going to do with this prompt. Now before we go on, I'll be using the easel prompt management app that I built. As we build out this prompt, you can access the prompt that we're going to build together at easel link gearbox prompt. What we're doing can be done in any word processing system you like and with any AI system that you like. If you want to copy the text of the prompt, just click here and copy it to your clipboard. And at the end of this section, I'll tell you how to request your prompt engineering certificate from easel. 21. [Chapter 5] Let’s Lay Out this AI Prompt Visually Using the ICO Framework: I often find when I'm designing prompts that it's helpful to start designing with the goal in mind and then work backwards here. The goal is to help the manufacturer of these gearboxes pre qualify these warranty requests. I'm going to have the AI put the warranty request into one of four categories and then I'm going to have it give a one paragraph summary justifying why the request was put into that category. So this means that the core function of our prompt is going to be to put a warranty request into one of these categories. Which means I need to define the categories. So that's going to be part of my context. Now, in order to teach the AI how to put the warranty request into one of these categories, I'm going to need some examples of previous requests that have actually been received and categorized. This will dramatically improve the accuracy of this prompt. I'll also need to teach the AI what should be in a warranty request submission. So I'm going to get that information from the company and put that in the prompt to, now that I've got the categorization parts, I need to enable this prompt to justify what it's doing and support its ability to categorize warranty requests that aren't exactly found. In the examples that I've taught the AI, what I'm going to do is get the product catalog and put the relevant parts of that product catalog into our prompt. This way the AI will know exactly what these products are, how they're rated, and how they're meant to be used. Now I'm going to design the output so that it's consistent each time it generates. Finally, I'm going to do the instructions. I'm going to do this last because I'll then know how the prompt works and what it is supposed to do. This will enable me to set appropriate roles and boundaries for the AI for this prompt. Now before we finish, we'll need to create a space where we add our actual warranty request submissions to our prompt. So to design this prompt, I started by looking at what I want the prompt to do. Then I identified the information the AI will need to have in order to do that. That information will be the prompts context. Finally, I added the output and instruction sections to the prompt and where we're going to put the warranty submissions. Now as we build the prompt, I'm going to add some natural language to connect these elements. For example, you'll see me tell the AI at some point to step back before it continues into something. Okay, let's build the prompt. 22. [Chapter 5] Side by Side: Let’s Add Context to Our Prompt: Okay, context. I've added information that the AI isn't likely to have and it needs in order to do its job of assessing these warranty requests. This approach is technically called in context learning because we're not changing the model itself, we're just teaching it as we prompt it. All right, to start with, let's look at these parameters. These parameters are what the company requires, the person or business making the claim, What information they have to submit in order to make the claim. You're going to see a pattern here that you're going to see a bunch in this prompt and you've learned in this class. We're going to set up the delimitter tags here, then we're going to just describe what those delimitter tags are encapsulating. And then we're going to give it a name, the parameters. If you want to see the source document where this came from, the actual business document, you can go to Easel link Dodge Warranty. You'll see that here on the screen. I just straight up took this information from the company's warranty claim requirements document. All right, let's continue down in the category sections. This is where we actually set up the categories that the prompt is going to put one of the claims into. This uses most of the techniques that you've seen in this course. Here are our categories tags. We're going to open it right here and then you're going to see, we close it right there. We're going to give it a definition here. After the categories with a capital C and then within the categories we're using markdown, we're creating headers here like category one, very likely covered, then this is text that's underneath that header. The AI can understand that we are establishing 1234 categories. Then we're going to create category four or we're going to make category four, this catch all category. Because if the submission doesn't contain enough information to properly assess it, I'll go ahead and fix this. Here doesn't contain enough information to properly assess it, then place it in this category. All right, great. Now going on, we're going to establish this catalog information, the same pattern. We're going to create a tag, a dilimeter. We're going to add a definition that relates to that particular dilimeter. Then we're going to use markdown. This comes directly from the company's product catalog. And you can actually look at it if you like, at Easel Gearbox catalog. You're seeing that link here on the screen. I have honestly found that this is one of the actual hardest parts, at least in terms of time consuming work, is when you are basically translating business documents like PDFs and things like that into plain text for prompts. One thing that I did to actually do this is I used AI to help me do it. You can do that too. A lot of AI systems you might have access to can accept documents as uploads. And you can say, hey, I want you to summarize this. You can use prompts to actually accelerate the development of the business. Prompts that you're building by having it do stuff like this. Sometimes you'll have to clean it up a little bit. And what we've done here is basically use mark down again to just name the features, the nomenclature, the special options and things like this. The selection process if you're building out true business grade prompts. A lot of times this is just going to be where you're going to spend a lot of your actual labor and time is to be in putting in the information that the prompt needs to have in order for it to know about a business or an organization's particular processes so that the prompt can do its job. All right, now let's go down to the examples. If you remember what you learned earlier in the course, context is basically about background information and examples. These first three things that you learned, the parameters, the categories in the catalog info, that's basically background information that is not probably in the AI's training dataset. We are teaching the AI about that in the prompt. Now what we're going to move on to is examples. Examples are very important to enabling an AI to do a great job. What we're going to do is provide an example from, from the businesses data about what falls into each category. And we're going to do that with a common pattern that you're seeing again and again, which is in between the example tags. So there's our Dlimitter, we're going to name it examples, capital E. And then we're going to add our example here, that's category 123.4 We're teaching the AI. This is an example of a warranty request that would fall into a given category. And this is going to make it again much more accurate. All right, now we're getting into our output area. We'll do that in the next module. 23. [Chapter 5] Let’s Create a Reliable Output Structure for the Prompt: Easel Prompt output directions can be really simple, like generate an E mail. We're not going to do that. We're going to control the output, specifically using markdown and handlebars so that each time this prompt is used, it generates output in a defined format. This lends our prompt and AI work to integration in advanced business systems. Like other applications, well defined workflows where you would expect the documentation or the output to come in a consistent format. Let's go through this. This is the output area right here. You're going to see here that I'm using some natural language to tell the AI model to step back and consider what it knows about quarries, mining processes, and industrial technical sales management. Now, these are not things that we taught it in the prompt. We didn't talk about quarries or mining processes or industrial technical sales management. However, AI models are definitely trained on those topics. By having the AI model step back and recall its training data, that is going to increase the accuracy and effectiveness of this prompt. We're also telling it to consider what it finds in the capital C catalog capital I inform the catalog information and the parameters. These are our definitions that are referring back to the information that is encapsulated in our delimitters. You can see that I'm referring back to the information that we taught the AI earlier here in our output. Then using the examples, capital E to guide you, assess the submission, which is going to go right here, placing it into one of the categories, and fill out the following form for me. All right, this part is basically at the beginning of the output. What we want to do is essentially bring it all together and tell the AI this is the data I want you to consider as you start to execute what I want you to do. Then I like to have the AI fill out this form for me. I find that by telling it to fill out a form, the AI is able to understand that I want it to keep a consistent formatting here. By doing a single hashtag, we're going to create a title basically. And the way we're going to create that title is through handlebars here and here. And we're going to tell the AI to generate a clear title related to the capital S submission and your assessment that's less than 40 characters. It's important to be very specific. Less than 40 characters. Ai is probably not going to go bananas and make a title that's like the length of a book. But in business prompt engineering, the more specific you can be, the better. Because for example, let's say that you want to flow this output into a system and that system has some length limits on the titles. This is how you would get into specifying what those length limits should be. All right, and then what we're going to do is set up a subtitle. And by putting this category outside of the handlebars, we're going to ensure that the AI says category semicolon. Then inside the handlebars, we're going to say identify which category you selected. Here we're giving it instructions inside the handlebars that are related to the submission. Now we're going to do another subsection which is rationale, and we're telling it to generate up to five sentences that provide insights into why you selected the category you did. Now in this case, we're going to give the AI a little bit of leeway up to five sentences. It's important to note that if you say three sentences, the AI will generate three sentences, 455 in some cases. In this particular prompt, we might not need all five sentences. I've left it at up to five sentences to give its insights into why it selected the category it did. That is our output. When we demo this, you'll see that every time we use this prompt, even as we change the submissions, our output structure is going to stay consistent. Okay, before we move on, here is where we're going to put the submission, and you've seen this pattern before. In between the submission tags, you'll find a warranty claim that we recently received hereafter, the capital S submission. And you'll see that that is referenced here in the output as well. When you try this prompt out for yourself, you'll just take out this here. So you can either use the easel lap or you can copy to clipboard. And you'll take out this part right here. And then you'll put in one of the sample test submissions that you can find in the study guide. 24. [Chapter 5] AI Prompt Instructions: Let’s Set the Role and Rules: Okay, so you're actually looking at a different prompt here. Why is that? Well, as I worked on this prompts instructions, I actually used AI to help me. If you go to easel links, meta instructions, you'll find this prompt which I used to teach the AI the goal of the prompt and also the contextual elements of the prompt that we're building here. And then I shared some general custom instructions that I use and you can find these in the study guide to I had the AI help me create these instructions that are appropriate for the gearbox prompt. You're going to see here that I actually encapsulated a lot of the prompt that I had already built out. The contextual parts of the prompt in between prompt tags. This is just to show that you can have delimiters that are inside delimiters. And I named the prompt here. And then you're going to see after adding these contextual elements, I added these default instructions, which I use in lots of my prompts. Then here at the bottom, I said, consider my prompt and default instructions and then step back and consider what you know about prompt engineering and create my sets of instructions. I just want to illustrate here that I tried this with a couple of different AI models. I tried this prompt that generates instructions with a couple of different AI models that I built into the easel app before finding results that I liked. I liked these Gemini Pro results better. Which is just to say that sometimes it's worth testing your prompts on different AI systems because they're all a bit unique and the models do sometimes evolve over time. And also, you should know that different models have different pricing structures that can make a big difference when you're using your prompts for business. For example, when you look here, you'll see that Gemini Pro currently is priced at this much. To generate this prompt, I mean, it's way less than a penny, where GPT four is about $0.10 That's just to say that these models are priced differently and they're better at different things. You might have to try your prompts on different AI systems to get the results you're looking for. 25. [Chapter 5] Putting the Prompt to the Test: Let’s Demo Our New Prompt with Test Submissions: Es. Okay, let's demo our prompt. I'm first going to copy the prompt to my clipboard. Then I'm going to go to an AI model and paste in the prompt and add one of the test submissions from the study guide. And then I'll generate, okay, here's the result. Looks good. Now I'll try this with another test. Submission looks good. Okay, so at scale, you might set up a system that would integrate this prompt into a business system through computer code. Or use some application that enables you to process lots of warranty claims at once. Okay, now let's get you your prompt engineering certificate. 26. [Chapter 5] That’s a Wrap! Congratulations Prompt Engineer!: Great job. Thank you so much for learning with me. To request your prompt engineering certificate head to easel link certificates and just follow the instructions there in the certificate request form. Now finally, remember that prompt engineering in some way is an art. There are so many ways to design prompts. I encourage you to be creative and try to retrain your brain so that you remember to use AI to help you with your work. This is a new thing. Ai can do almost anything. You've just got to ask it for help with the right prompt.