ChatGPT Prompt Engineering: Learn to Communicate Generative AI in 2 Hours

1. Welcome to Essential Prompt Engineering Class: Keeping up with AI advancements is a daunting task. One day you'd find out that being polite with AI gives you better answers and you go spend time testing it. The next day you'd see some sort of a magical symbol that supposedly makes Chajipeti open a Pandora box of intelligence. In reality, that's a huge waste of time. Instead of focusing on random posts, tips, tricks, and hacks, dedicate your attention to mastering the long term principles of prompt engineering. In the following lessons, we'll dynamically go through the principles that won't change regardless of the latest half an hour ago release. This will especially be useful to someone working on marketing tasks. But all of these approaches can be applied to other professional fields. Many of these approaches have a much more complex nature and come from machine learning. But what we casual business workers and marketers need, we just need an approach that works with AI on a daily basis without writing any code. Let's think of this as learning to communicate with generated AI tools like Ajipet. By the way, one exciting side effect that I noticed on myself and many other people is that prompt engineering is a transferable skill. If you create or communicate tasks to other people, you find yourself being more structured and effective in your communication. But pay attention, this is not a technical course for developers, it's therefore business related professionals like marketers, managers, content creators, and others you need to learn. Writing prompts to solve daily tasks effectively. If that's you, you're in the right place. The generative AI field is often called a once in a lifetime breakthrough. There's really no reason not to learn to use it. This field is so dynamically evolving that I have to update this course regularly with fresh content to make it relevant and interesting for I address every single review that you guys leave. Trying to double down on the strength and fix any issues that you bring to my attention. Thank you for enrolling in my course and let's get started. 2. 1. Understanding Strengths and Limitations : In this lecture, we are going to discuss Chagipt, what are its strong and weak sites? Chagpt is a natural language processing chatbot driven by generative AI technology. It allows you to have human like conversations and actually much more Gp can answer questions and assist you with tasks such as composing e mails, essays, and even writing code. If you dig really deep, it can be integrated with a lot of other tools to assist you in more tasks. Here are some of the downsides. The first limitation is that Chagipeti is not a reliable source of information. As of now, Chagpti doesn't provide accurate sources of information that it uses. It was trained on billions of information entries, but it doesn't know what these entries really mean, Which of these are more accurate and which are less accurate. Moreover, you don't have much control over the generated data. Each regeneration would result in a slightly or even significantly different answer, which is actually a blessing and a curse. That also means that the response that you'll get might be different from the one that you see in the demos of this course. Even if you enter exactly the same prompt, you might think, Leo, they already have this browsing feature. You can browse with Agape well in 2024, this feature is still unstable. Although Cha GPT now has this browsing functionality, it's not currently very good and accurate at searching and processing information. It often produces mistakes. Blends them with hallucinations very confidently that it's hard to notice it. I expect some improvements in the future, but that's what we have right now. For the time being, I would recommend using research specific tools like being Perplexity or U.com for any tasks that involve browsing. The second downside is a continuation of the first one. The confident mistakes that ChagiPT makes are called hallucinaus. You see, Chagpt is trained to write using a language, many languages actually. It can produce faulty information very confidently without you even noticing it. It can generate non existing sources of information that look very, very believable. It can even pretend it knows the content of a link, but in fact, Ch GPT is just guessing the topics within the URL slug. Here's a high level, but very practical tip, your good use case, no hallucinations, bad use case, lots of hallucinations. An offensive prompt wouldn't fix a bad use case. The next downside is its lack of emotional intelligence. Gbt can simulate natural conversation, but it lacks the emotional and real world experience. It's the intelligence of a human conversation partner. Gbd can have the difficulty understanding and responding appropriately to subtle nuances in communications. But hey, we humans, we are sometimes even worse than this. Let's move on. The next downside is that Jip itself is not very good at maths. At first, this was a huge problem, but now GPT users can access some of the improvements, for example, by triggering Chagpt's advanced data analysis functionality. Or by using, well, from a custom GPT. The next drawback is about privacy and security concerns. The use of GPT requires the exchange of data and information with the system. This poses potential risks in terms of data protection and security. It's important to take appropriate security measures to ensure that the sensitive data is protected and doesn't fall into the wrong hands, which actually already happened. Number six is the computational cost. Gp is a highly complex and sophisticated AI model. It requires substantial computational resources to run. Organizations should carefully consider their computational resources and capabilities before using PT. Let's say on premise, I believe this problem is temporary and with time we might be even able to run large language models offline on our smartphones. It's actually something that Apple is working on right now. But at the moment of recording this lecture, it's far from being in production. By the way, Sam Altman, the founder and CEO of Open AI is actually looking for energy resources to scale the product to new levels. So like most other AI technologies, TGP is great at finding patterns and analyzing data that you provide. Ch GPT does way more than just writing social media and blog posts. It can provide almost an infinite amount of ideas and points of view in a matter of seconds. You just need to prompt Ch GPT write and be aware of its limitations. The advantages actually outweigh the disadvantages by far far. One other advantages that Chachipt users don't utilize enough is it knows professional frameworks, whatever copyrighting formula you have, whatever format of business or marketing analysis you want, it can handle it. The primary advantage of ChagPt over other large language models is its simplicity in customization and personalization, and it's very easy to integrate in your workload. You can set custom instructions, custom puts, and incorporate Chgpti with other tools, for example, Harpo U.com Descript. All of these tools utilize Chagipt to enhance their multimodality. By the way, multimodality is an important term in Eni. It essentially refers to processing various types of media such as text, audio, video two D images, three D images, and so on. The list can go on, but it's important to remember that Gp is just one tool in your huge professional toolkit. The most important tool is your critical thinking and your experience. Yeah, let's move on to the next lesson where we will look at this tool in more detail, see in a few seconds. 3. 2. Data and Privacy: Hi everyone. In this lecture, I want to draw your attention to an aspect that I'm sure many often overlook. Something that even a few employees at Samsung have done. By the way, check out this story on tech crunch. Just Google Tech Crunch, Samsung GPT band, something like that. And you'll see the first link. All right, so if you're like me working in an enterprise setting, handling corporate information, there is a crucial setting that we should remember to utilize. Particularly when you work with data that is confidential or at least close to confidential. Here's how to ensure at least partial confidentiality within Chap. First, you'll need a GPT account while working with confidential data. Then within chat GPT window, open settings. Then in data control step, disable chat history and training. The reason I say partial confidentiality is that the data is stored on open AI servers for 30 days. Anyway, I'd suggest removing any brand mentions, sensitive data, or any data that can identify your business and you'll probably get away with it If you're using a free version of Cha GPT, I would recommend to not enter any confidential data at all. It's also related to Bin Chat and Google, Gemini, and Claude. As these large language models use your input to improve their models. I know you came here for other things, but safety comes first. So please be safe and never underestimate the privacy as it starts with you. All right, cheers. And seeing the next lecture for stuff that's a bit more fun than data and privacy. 4. Prompt Engineering Approach and Components: What makes a good prompt? What are the components of one good prompt and how to build yours? Prompt is the main instrument that we use in large language model like Cage PT when we interact as users. And it's what JAI needs to, well, generate something. That could be our question, a step by step instruction, or a task, depending on how much information and instruction you give to Cage PT? ChagpT is going to either come up with. This could be our question, a step by step instruction, or a task, depending on how you give this instruction to Cage P? GPD is going to either come up with a high level generic answer or a more profound more detailed response. So what are the components of a good prompt? We already discussed this on a high level, but let's go into a bit more details. First, it's an exact and detailed task. Think of the large language model, a bit like an SEO. You need to put the right keywords to activate the right parts of the model straining data. If you struggle with it for your prompt and you don't have the inspiration or time to optimize your prompt, try the tool called prompt Perfect. The link is over here. And also, it's in the resource section. It automatically optimizes your prompt by either removing excessive words or adding more details depending on what you put into the system. Okay, let's go back to talking about the prompt. So it's actually the reason why assigning a role in the beginning of a prompt is a good idea. It's the reason why asking to follow specific frameworks related to your tasks is also an amazing approach. Second, a good prompt should have a context and a task. A good prompt needs to provide information or context needed to complete the task. Think of this as if you're explaining your task to a new employee or a freelancer. Third, your prompt needs to balance clarity and precision with details. Avoid contradictions and unnecessary words. Try to structure and format your prompts. The fourth component of a great prompt is singularity. It's one prompt, one goal, one task. Bona would definitely like that one. All right. Try to use simple language and avoid confusing the large language model. The next step is to include negative instructions where necessary. Meaning setting limitations on what not to do or what to avoid. Just as we would add negative keywords on Google search ads, this also helps us avoid any irrelevant answers. Tip number six, if you see yourself repeating the same use cases, and the same type of prompt, invest in testing and iterating on developing your prompt. Refine it. Add what you think needs to be added. Once you get impressive results multiple times in raw, save that prompt in your notes or a template library. Now, let's put it all together into one prompt. We'll start with assigning a role, mentioning the industry, the company, what it's doing. Then we'll continue saying what we will provide, what kind of output we expect from Chachi Pt. Then we set rules on what it can do and what it cannot do. Then finally, we provide any relevant input or like market data, for example, that universal all round prompting technique will save you hours of experimentation and will get you thinking in the right categories while prompting. All of the other prompt engineering techniques that we will discuss here will relate to these components and principles in one way or another. So see in a few seconds for the next lecture where we go and discuss those techniques. 5. 0-Shot, One-Shot and N-shot Prompt Engineering in Text, Audio and AI Image Generation: In prompt engineering, the term shot basically means an example. Shot prompting is a technique for guiding gipty and generating specific responses by providing specific examples or also known shots. This approach is particularly useful for marketing professionals and business professionals who need to create structured content, such as blog posts, social media post, product descriptions that align with their brand tone and style. So Why use shot fronting and gupty. Reason number one is improved content relevancy. So shot fronting ensures that the generated content meets specific requirements and is tailored to the brands voice and style. Reason number two is increased efficiency. By providing clear examples in context, shot prompting reduces the amount of editing and rewrites, allowing us to save time and focus on other tasks. Reason number three is enhanced creativity. Well, Chad GPT's ability to generate content based on a few examples, encourages creative thinking and experience. And you can lead to innovate ideas that might not have been considered otherwise. For example, why not search for inspiration from more creative industries than the one of your brand? If you work in a tech company, why not use inspiration from a fashion brand? To use shot fronting and Chagpt, identify the specific goal or task that you want to accomplish, such as creating a social media post or writing a product description or series of CO meta descriptions. Then find relevant examples of content that align with your goal or your inspiration. That could be a successful social media post of your brand or an example of the best selling product description. In other words, use successful examples to guide ChagpT in creating similar content. Number three, construct a prompt that includes context and examples, and then use Chad GPT to generate the content. Let's suppose you want to create a social media post for a new product launch. You find a successful post from a similar brand or competitor. Use it as an example to guide T GPT. Here's an example. You're a world class marketer for a tech company. Your task is to write social media post. Using specific context that I give you. Then you enter your context. This can be as detailed as you want, and it could actually have multiple sections or even file tat. And then provide a post example. So if you don't provide an example, that would be a zero shot prompt. If you provide one example, that would be one shot prompt, and if you provide more examples, that would be N shot prompting. By the way, five shot prompting is one of the standards for benchmarking large language models. So, yeah, let's add a post example. By providing this context and example, GPD can generate a social media post that is similar in style and tone, ensuring that your brand's message is effectively and consistently communicated to your target audience. By the way, not many people are talking about it, but shot prompting is actually a technique that is also used in other modalities. Image generation, you can provide a reference and generate a similar image. This way, you get so much more control over the colors, the composition, and the overall style, and you actually have to prompt less because the model would take a lot of parameters from the image itself. While it won't keep the reference face the same. The images are still going to be quite similar. Let's have a look at an example in Adobe Firefly. Let me open my Adobe Firefly and go to text to image functionality. Then I'll applaud this image of myself as a reference. And enter a basic prompt. In this case, the reference image is a shot. Therefore, we get one shot prompt. In consumer image generation tools, one shot is typically all that you get, but it's already helping a lot. Now, let's compare an image generated without an example, which would be a zero shot. With an image with example, one shot, not to say that one is better than the other. But to me, it is so obvious how much more control you get and how you increase your chance of getting closer to your reference. Now, is it it? Well, almost, let's look at how it works in audio. In audio, your shot is an audio sample. Let's call to Stable audio and follow the same logic we did with the images. By the way, stable audio is a tool that allows you to use NAI to create music without lyrics. It's actually ideal for background audio. And by the way, it's a licensed stock music, so you can use the audio commercially if you have a subscription. So here we are in St audio. Let's upload this audio sample. And now let's give a prompt. Let's listen. It's a new track, but you can already hear the impact of the example that you provided. And you can actually adjust it as well. Now, let's discuss a couple of more practices for shot prompting and Cha GPD. Number one is use clear and concise language. Ensure that your prompt is easy to understand and includes all necessary context and examples. But try to make the prompt itself relatively simple. Different tools and modalities may have their prompt guides, so please check out as a rule of for text models, start with a verb like Search, do perform for images. Don't start with the verb. Start with describing the subject. We'll discuss this in more details. For audio, don't use verbs as well. Instead, describe the style, temple, instruments, et cetera. Then next, provide relevant examples. Use examples that are relevant to your goal and align with your Brand stone. It's a great way to get closer to the intended brand tone of voice. You'll just have to make less edits. Isn't that beautiful? Well, depending on what you're trying to achieve, you can either provide reference examples of great campaigns as an inspiration or use your past best performing work. Example. You could use posts with the highest engagement right on reach. Or if you're doing Instagram, you can choose the posts that were shared or sent to someone else the most, as Adam Moser recently said that they actually consider share on reach metric. By the way, one more great source for examples is the website that I follow, and I love. So it's here marketing examples. Really great resource, highly recommend you. So once you get the output, don't forget to actually read it and refine the generated content. Just to make sure that it meets your objectives and quality standards. Unlike many instructors, online, I highly encourage not to let the AI do an unsupervised work for you, especially if it goes public. But with the help of these methods, you'll have significantly less work at this stage. So remember this idea, create content with AI. Not by AI. So by incorporating shot fronting into your ChagpT workflow, everyone can create a more consistent content, ultimately driving better results for their campaigns, presentations, e mails, communications, and more. All right. Hope that now at the end of this lecture, you're ready to go and use this technique. We'll discuss more ways to communicate with JAI and I'm excited to see you there. 6. 5 Chain of Thought Prompt Engineering Technique: Welcome back everybody. Most of our tasks in life are multi step, meaning we need to take a few logical steps to solve a problem. Like all of the other techniques we discuss here, this one works equally great, whether you're using a ChagPT, Google bard, or tropics clod, doesn't really matter What is chain of thought, Prompt engineering technique. Chain of thought is a prompting technique that helps us guide Gibt or any other large language model by breaking down the reasoning process into clear digestible steps. This way we can ensure that the answers are transparent, easy to understand, and highly valuable, especially in complex problem solving or even analytical tasks. Why is this technique so important in generating I? First, it's clarity. Dividing reasoning into simple steps clarifies complex processes. Transparency helps us see how solutions arise, building trust and identifying potential errors and analysis. Analysis takes multiple steps of processing data and information and that's where chain of thought shines. How do we use this prompt engineering technique in real life? It's very simple. Start your question as usual. For example, what are the potential competitive advantages of an AI driven video editing app? And then simply add, let's think step by step, but what if you want more flexibility as to how Cha Jet thinks? In this case you might want to provide your sequence of actions and logic. This works particularly great when you use advanced data analysis, also known as code interpreter, which we'll discuss in a dedicated lecture a bit later in the course. The more flexible chain of thought formula is as follows, your context plus goal, then x step by step, and then provide your logical sequence of actions. This type of fronting guides the AI to provide a response that includes each step in its thought process. This way you can easily follow along and understand the motives, reasoning and actually have more control over the logic. There you have it. A simple, effective way to guide your AI in conversations, particularly when it comes to marketing strategy and working with market data. Now here's a little practice for experiment and combine techniques to achieve better results. For example, try to combine chain of thought with few shot prompting. This is a great way of thinking about how to solve your task with the help of Ch GPT or any other large language model. All right, see in a few seconds with the next technique. 7. Feedback based iterative prompting: Imagine you're working on a complex problem and GPT isn't quite getting it right. Just something seems off. Well, this is where feedback based prompting comes in handy. Without further ado, let's go for a step by step guide. Step one is initial prompting. Start by giving ChagpT your usual prompt and receive its first response. As a step number two, ask CGPTi, to criti its own response and suggest improvements. This step can significantly enhance the quality of the output. Step three, use perspectives. Think what your stakeholders might say or ask and who they are. When prompting GPT specify the perspective, such as, let's say, provide feedback from the copywriters point of view or from CMOs, like Chef Chief Marketing Officer. That could be asking to provide feedback from the copywriters point of view or CMOs point of view or CEOs point of view, right? And this would actually add a valuable feedback layer to your process. Once done, identify the suggestions you find most valuable and ask PT to elaborate on those points just as you normally would. To make it a bit more practical, use the prompt templates from the Prompt library to get started or to revise the materials from this course. As I'm signing off, I'm getting ready to record the next lecture. I wanted to check out this conversation with CGPT as a practical example, the real world example. If you're ready, try this technique on your own and see how it can improve your interaction with ChagpT. Any way? See in the next lecture. Cheers. 8. 6 Self Ask Prompt Engineering Approach: Self asking prompting technique is the technique that we use with ChagPT to ensure that the generated responses are accurate, precise, and tailored to our needs. In this lecture, I'll show you how to use this prompting technique when you can use it. And I'll also add one additional twist and a productivity tip for this approach as well. At the very end of this lecture, we'll go through a ChagPT conversation example that uses this technique. South asking technique involves guiding Chagipeti to seek clarification and ask for additional information before delivering the final answer. Let me draw you a little situation when you might want to use this one. So imagine you're a PPC, or like paid advertising specialist, who suddenly is asked to write a website article or create a social media content plan. Which actually happens a lot at start ups for example. So you have a general understanding of marketing, but you don't know all the process and all the information that you might need to do it. In this case, asking for clarification, using the self asking prompting technique will save you. In other words, this approach is valuable because it enhances the quality and relevance of your responses. But it also helps when you don't really know how to accomplish something or you don't have enough context. By encouraging Agupeti to ask questions first, we ensure that it gets a clear understanding of the task and context before generating its answer. And it actually reduces the chances of misinterpretation, hallucination, and improves the overall quality of the responses. Moreover, you can figure out something along the way yourself. Implementing the self asking prompting technique involves two quite simple steps. First, just write the prompt as you would normally do using a short prompt or whatever it is that you're trying to do. Then you might want to mention what kind of questions and how many of them you might want to ask. If you simply say ask me questions in order to produce better results, Chagipeti can ask you too many questions or ask you questions that are too vague and it might be hard for you to respond to all of the questions. It would be hard for Agipeti to understand everything that you've given because that could just be too much of a prompt. So I find three to five questions are around the golden middle and it keeps the response and information short and saves a bit of your time. While Chajipeti is trying to ask you some of the most important information, let me share a quick example with you. Here is the prompt. You are an expert in digital marketing. I'll be asking you to draft out a campaign plan, but if there's any ambiguity, feel free to ask me three to five questions before delivering the plan. Here's cool productivity tip for this prompt engineering technique. The problem with it is that it takes a lot of time to type in all of the response to the questions asked. If you're using a Mac, then you can simply press a five and start speaking. For any other systems, you might use an extension called GPT Mic or any other built in voice input. Don't worry if punctuation or a few words are incorrect. Gpt is trained to figure these things out, as with many lectures in this course. Here's a cool twist. Again, if you work with many stakeholders, customers or requesters, you most likely use a brief of some kind or a form that your customers or stakeholders fill out. You can dramatically increase your productivity by aligning your brief with a GPT prompt. Using this technique and the chain of thought prompt engineering. For that, you'll need to have a base prompt and a form that will collect the information in the exact same order. Here's an example of a prompt that you will use as a base. Guide me on creating a copy step by step. First, study the context, then ask me five questions about the audience goal and information needed for social media post creation. Describe the audience, describe the goal of the post, and provide the information needed for the post. By the way, keep in mind that this isn't just an example and you might want to include different fields for your prompt and your brief based on the needs and traditions of your company. Just make sure these are aligned right. Let's quickly summarize this one. The self asking prompting technique or ask before answer. This technique ensures the accurate and tailored responses from Apt. It involves guiding Chagpet to seek clarification and ask additional questions before delivering the final answer. This technique is useful when you lack knowledge or context in a particular area. Implementing self asking prompting technique involves writing a prompt and specifying the number of questions and three to five questions strike the balance between the response length and gathering important information. Voice input can be used to save time when providing responses to Gps questions. Aligning a brief or form with ChgPT prompt can increase productivity and ensure you collect the relevant information. The prompt and brief should include fields that match each other. If you feel like you need more flexibility, you can simply ask Gupte to specify what kind of questions you wanted to ask you. I don't use it that often myself, but there are cases when this really helps me out. Anyway, I hope that you had fun experimenting with this one and take care, see in the next lecture. And meanwhile, stay here in this lecture to check out how these conversations go. I'll put some music here so that it's not that boring, right? See in a few seconds. 9. 7 Self Ask Prompt Engineering 5 Why Approach: Hi again everyone. In this lecture, I wanted to show you a very cool twist on the ask before answer Prompt Engineering Technique. And this one is going to be a bit different in its logic. We are going to use the five Y technique in order to get to more in depth and more relevant solutions to our problems. This is a great way to go towards your desired output through a little bit of a problem solving approach. Let's go to my screen and I'll show you how it's done. Typically, when we ask GPT to ask us questions, we ask to do it right away, Then we provide the response, and then GPT comes up with the desired output. In this case, we go a bit more sequential. The five wise technique is a problem solving framework that allows to get into the matter of the problem by asking five questions. Let's start with why and do it step by step. Here we are in a pet, and let's start with the first example. The context is, I'm preparing a local media press release for a fitness studio. The prompt would be, ask me questions using five Y technique, one question at a time. Once done, suggest a list of five topics and headlines. Here's the first question, why is the fitness studio issuing a press release at this time? I would say to improve website authority and awareness. Then why does the fitness studio believe a press release will effectively improve website authority and awareness? I would say they have a good domain authority and website traffic was Y number two. The Y number three directly responds to my previous answer saying why does having a good domain authority and website traffic make a press release suitable tool for further improving website authority and awareness. I would say that it helps me rank higher on my target keywords and get more website visitors and leads. Then I have the same question, but to my last answer, why is ranking higher on target keywords crucial for your fitness studios, online success? I would say leads and customers with high LTV. And then the fifth y actually strikes the real reason. And it really happens like this in real life. When you interview customers for your products and product interviews, then you really get to the truth in the fifth, why. The response is that competitors in my city don't invest enough in SEO and getting good search traffic. And this approach gives me a marketing advantage. Now finally, based on my input, here are five potential topics and headlines for your press release that could leverage fitness studios, SEO, and competitive advantages. Of course, I did not describe the advantages of the studio itself, but let's have a look at the examples. The reason why we have a list of five topics and headlines is that some of those are just not going to be good. And let's accept this fact. Let's have a look at these five options. Topic number one says unique fitness approaches. The headline is innovative training techniques. How studio name stands out in the city. Out say it's too blunt. Let's move on. The next one, transforming lives. Real success stories from studio name, Kind of generic but better. And the third one is the one that I actually like, building a healthier community, fitness studios impact in the city. And that is actually one that I would probably choose. Let's have a look at the two other. Number four, meet the experts behind Premier Fitness Studio. I would play around with it, but I like the experts behind, or I would even change it to people. So you see you don't rely on Chagipeti in the final, final, final, everything you are submitting this, so it's our responsibility to come up with a final result. And don't be afraid to tweak a few words here and there. And the last one is actually very GPT style with revolutionizing fitness in the city. Not realistic at all. Sounds a lot like Chagapeti. I would skip this one, but at least I have number 3.4 here, which are actually workable results and something that I would need to think of on my own maybe for a bit longer. All right, so let's summarize using five is with self asking or ask before answer. Prompt Engineering technique is a great way to arrive at new ideas through problem solving. I hope you have a chance to give it a try and you like what you get out of it and see in the next lecture. 10. 9 Universal Prompt Engineering Technique for Chat GPT Marketing: Hi and welcome back. As we discussed previously, the true power of AI is finding patterns, analyzing data, and learning from feedback. We are going to use this power to engineer sequences of prompts that can solve your marketing problems tasks. In this lecture, we are going to discuss five main building blocks of prompt Engineering with Chip. The first building block of prompt engineering with ChagPT is to assign a role and give clear context. Imagine giving a task to a freelancer that doesn't know anything about what you're doing. You need to provide the context. Otherwise, it's hard for technology that doesn't know to be aligned with your objectives. The second building block is market data. You need to provide Chagapet with market data that you can rely on that you know and that you can validate. The third building block is your audience. You need to provide this information to Cha GPT so it can adapt its answers to your particular case, your particular audience. Otherwise, it's going to give you more generic high level answers which are not going to help make any conclusions. Obviously, sometimes these answers are going to disappoint us. We want to provide this information upfront. The building block number four, iterate using marketing frameworks. As we discussed in the previous lectures, Chagpetin knows most of the world's frameworks in most of the world's fields in marketing. There are so many frameworks that can help us elaborate on the information that we already have. For example, if we talk about our target audience, there are different ways of segmentation. There are different ways of conducting competitor analysis. This is where Chagpt shines and where we can benefit a lot. Building block number five. Last but not least, ask for creative ideas. Once you have the context, the audience data, the market data, you've elaborated using frameworks, you can now generate some awesome creative ideas. It actually meets the process. It reflects the process that marketers usually have when working on their projects. We start as strategists and finish as poets, as artists. Well, that's exactly what we are going to do with Chad Pet in the following lectures. I know it sounds a lot like philosophy, but it's really a mindset that can help us make most of GPT. The next lecture, we'll have a look at a practical implementation of such approach. 11. Prompt Engineering RAG-Inspired Approach: So what is retrieval augmented generation? Introduced by Meta, Retrieval Augmented generation, also known as Rag, is a technique that combines information retrieval with language models to generate more accurate and contextual responses. Reg models pick relevant information from a knowledge base and use it to guide the generation process. This results in a more relevant output that is grounded in facts and better aligned with a given context. However, building Reg is a technical process, quite advanced one. And it requires a specific set of programming skills and resources to build a reg application or API integration. With the techniques in this lesson and some limitations, of course, we can take advantage of g inspired approach in GPT. That's why it's more correct to call it inspired prompting rather than a pure g application. Over the next couple of minutes, we'll discuss why use Rg, the best rag use cases, how to build a no called g inspired custom GPT, in char PT, how to minimize hallucination in the custom GPT Rg. So without further ado, let's go. First of all, why use g in prompt engineering. Reg offers several benefits for prompt engineering. Number one is improved accuracy. By retrieving relevant information, Rec can generate responses that are factually correct and less prone to hallucinations. Though, speaking of hallucinations, there is an nuance that we'll discuss a bit later in this video. Number two is enhanced context. Reg uses contextual information from the provided documents, enabling them to generate responses that are more relevant and reliable. One of the most important benefits is the expanded knowledge. Reg allows language models like GPT access information beyond their training data. It enables to handle prompts about current events, user specific data, or company documents, or let's say product information, pricing, et cetera. Now, let's talk a bit about the use cases in business. How can you use Reg? Well, some of the most promising reg use cases for business environment include customer support chat bots. Of course, by accessing up to date product information, Rec empowers chat bots to provide more accurate and contextually appropriate responses. Number two is business intelligence and analysis. Well, you can generate market analysis reports or insights by retrieving and incorporating the latest market data and trends. You can also use your brand guidelines, strategies, tactics, and let's say product or service information to for example, consult generate ideas for content, create content and maybe some presentation decks that can really speed up the process and improve the relevancy of the content that you create. Speaking of content, content creation is a very important use case. Rec can improve the quality and relevance of the content by pulling in accurate current information from various sources. And it would make your content more informationally saturated, so to speak. Reg approach can be used to maintain your tone of voice as well. Let's talk about how to build a rag with no code in GPT. Well, as I said before, Rg is a technical term. Normally, you'd need a technical person or even a team to build a g app, but non techs can also take advantage of this approach. For example, by creating custom PTs, which are basically rags in their nature, or simply prompting an LLM that supports text file attachments. Let's go through step by step instructions on creating your rag with no code, by creating a custom GPT. Step number one is always is to define the goal. This is a crucial step that will impact how you prepare your knowledge base and prompts. And the best way to start, if you choose one primary use case and work your way from there. Then prepare your knowledge base. Gather the relevant documents, articles. Data, you want your C system to have access to and refer to. Just copying and pasting and throwing tons of information into one document wouldn't really work to its best. Only to some extent. With your intended use case in mind, make sure that all titles and final names are named so that it's easy for the LLM to scan. Make sure you use the same keywords in the knowledge base and your prompts to maximize the scanning accuracy. This way, the system would trigger the right parts of the document. The best formats for GPT are Doc X and CSV, and of course, DST. Images are ignored. It's critical that you know your knowledge base quite well and can access or updated when you need it. From my experience, I'd recommend to stick to one document as it works better than multiple ones. Let's go and create Custom GPT. So go to GPT, click Explore PTs, and create a new GPT, and click Configure to manually enter the instructions and upload the knowledge base. Use a combination of prompting engineering techniques, give a role, context, task, and step by step instructions. Now, this prompt part will be important because otherwise, GPT will blend the training data and data from your knowledge base and produce a hallucination. For example, it could pick a product description from the knowledge base and hallucinate its price and availability by triggering the general training data. To avoid this in your custom GPT configuration, use the phrase, search, the knowledge base two. But This will trigger the data analysis function in CI GPT. Now, step number four is to test and refine your prompts. Save Custom GPT and test how it retrieves information from your knowledge base. Challenge it to provide responses with factual information from your knowledge base. Review your instructions, prompts, conversation structures, and knowledge base. Make sure to optimize the vocabulary and keywords to ensure the perfect match between Rag and prompt. Now let's go over some examples of Rag prompts for different use cases. Number one is customer support. So you are a customer support agent for let's say your company name. Use the provided product information to answer the following question, and then customer query. Market analysis. You are a market analyst for industry, provide a summary of the current market trends based on the latest industry reports and data. Then you would attach your industry reports and data, of course. Now let's have a look at the content creation prot. You are a content writer for brand or type of brand, if it's not very popular. Use the provided brand guidelines and target audience information to write type of content about topic. And then you can even provide a couple of notes. Here's an extra layer of protection from hallucinations. Once you have the answer that you like, prompt this. Double check the details in this response for alignment with my document, or knowledge base if that's your case. Find out if there are any discrepancies between this text and the document provided. Let's summarize. Using rag techniques and prompt engineering, marketing and business professionals can generate more accurate contextual and effective content to support their business objectives. And while some of the examples that I provided are from marketing, most of them are from marketing, because that's my experience. This can be applied to anything. For example, if you're learning, you can take notes from a course that you are taking, summarize them neatly, add your particular information of your project, and then create a Rg just like this, and you will communicate with the knowledge that you've gained. And in my experience, that creates a great way to implement the knowledge that you gained and to practice it, because we often take a course and forget most of the stuff in the span of a couple of months. So this helps to actually apply the knowledge that we gained, not forget it. All right, that's it for this g inspired prompt generation technique, see in the next lecture. 12. 11 How to Refine Your ChatGPT Responses: Let's talk about how we can significantly improve the result using the prompt refinements. Tip number one is to simply regenerate response. You'll be surprised at how different the responses may get from attempt to attempt. Tip number two is to rate the response. When you rate a response in charge pet, it asks you exactly what you didn't like. After your feedback, it suggests an alternative response. And for me, work quite well. Way number three is just ask what you need. Refinements can be like beacon size or right in simple English, or this by alphabet, or by value, or whatever criteria you may find effective. You can also ask to summarize this into a particular amount of text. I'll provide more of these templates in the resources of this lecture, so be sure to check it out. And by the way, it's also in the prompt library, right? Let's move on. Number four is my favorite and it's about the output format. You can either put it in your prompt right away or ask to rewrite in the refinement I'd like to ask to write in a table format because for me, it's a comfortable way to quickly digest information. Also, I notice that it improves precision in the output. By the way, if you need to paste your table into GPT, just type put this data into a table and then paste your table. It will look messy in your prompt, but most of the time it figures it out correctly. Alternatively, you might want to ask to output the format in markdown or HTML. All right, let's summarize. Use refinements to significantly improve the results that you get from GPT. Regenerate the responses. Ask for refinements that you need and customize the formats. I hope that these four simple tips will help you improve the results that you're getting with Cha GPT. And I hope that it's going to be fast and effortless. See in a few seconds in the next lecture. 13. AI Image Generation Prompting (Works with Midjourney, Adobe Firefly, Stable Diffusion and Others): Like it or not? AI image generation has taken its niche. Interestingly, it doesn't really substitute a real photography, but has an interesting function, if used correctly. You have ever tried to generate an AI image, you probably noticed that writing a prompt for image generation can be significantly different from writing a prompt for text generation. That's why this video we'll cover the specific techniques for writing AA image generation prompts. With slight adjustments, these techniques can be used practically in any AI image generator, such as mid journey, adobe Firefly, staple diffusion, gt images, or whatever other mage generation comes out. We won't go into the specifics of each particular tool in this video. Instead, we'll focus on writing those prompts and getting more creative ideas flowing. Okay, let's start. So the main difference between writing prompts for text models and writing prompts for image generation models is the language and the goal. Prompting a text model, you would often guide the model using action verbs like search, describe, analyze, write, rewrite, prompting an image generation model requires a slightly more descriptive approach. That's why you would want to stay away from guiding the model this way. Actually take a look at components of the prompt for image generation. These components are subject action, environment, composition, style, and visual effects. The main component, of course, is a subject. The rest can be optional and added for more control over the generation. So if you put one word, let's say a cat, you'll already get some result. Let's go through each component one by one. So the subject, name and describe your subject with attributes. For example, if your subject is a cat, you can describe it as a fluffy orange cat with green eyes. The more specific you are, the more control you'll have over the generated image. Next component is action. What is your subject doing? What's going on in the frame? What's happening around the subject? For example, a fluffy orange cat with green eye, lazily stretching in warm sunlight. The next component is environment. Here, we specify where all of this is happening. So that would be a fluffy orange cat with green ice, lazily stretching on a cozy plush, pillow in front of a roaring fireplace. Well, these are more words, right? It's building up. The next component is composition and angle. This one requires understanding some basics of photography. The cat in the center? Is it a close up shot or a wide angle shot? Is it shot from low angle from below or from above or from the top? Let's add this part. A fluffy orange cat with green eye, lazily stretching in the warm sunlight on a cozy plush pillow in front of a roaring fireplace shot from low angle, with fireplace, softly blurred in the background? Next comes style and visual effects? Scribe the visual style or specific effects? Is the image black and white or color? Is there a specific color palette that you're looking for? Is the image supposed to look like painting, art or photograph? Is there a specific lighting effect that you want? For example, harsh daylight, warm sunlight, golden hour, blue hour, counter light? Are the shadows soft or hard? Do shadows create leading lines? Or maybe you want to use rembrant lighting on a portrait? Well, of course, if you're generating a portrait? You can see, even if you have experience in visual arts or photography or videography, there's still plenty of room to expand your vision and creativity and just go and learn. So let's expand our example even further. A fluffy orange cat with green eyes lazily stretching in the warm sunlight on a cozy plush pillow in front of a rolling fireplace, shut from a low angle. Hard shadows create leading lines. Okay, last but not least is gear. This one is quite optional. But advanced users who know the specifics of gear can gain even more creative control. By naming photo parameters and gear, you can control the focal length, the blur, the bok. However, some tools like Adobe Firefly don't allow brand naming, so you need to be a bit careful with this. Sometimes it works, sometimes not really. But let me give you a couple of ideas that work for me. So, I've never actually shot on film. I like the looks of Portra 400, Kodak Gold, 200 and Fuji Ostia. But yeah, if you remove the brand name, it still can pick the relevant training data. So let's expand our example even further. A fluffy orange cat with green eyes lazily stretching in the warm sunlight on a cozy flush billow in front of a roaring fireplace shot from low angle. Hard shadows create leading lines, shot on 85 millimeters F 1.4 L port 400, and let's have a look. Beautiful. Last but not least, don't forget that you can use images as a reference. In this case, the model takes care of a big part of what we've just discussed. So your prompt should be much shorter. So here are a few more practical ways to improve your image generation. Learn the limitations and current biases. You'll be surprised at how biased image generation models can be, especially if you start talking about races, genders, and nations. So if you're trying to generate something like this, be really cautious and pay special attention to understanding whether you're actually hitting a stereotype. Make sure that you have this human quality control and diversity, equity, inclusion vision built in in your mind, so you are the best biased filter. Then next step is, sometimes, less is more. The very simple prompt can really work well. From my experience, sometimes long, detailed prompts like the one we created can create unique and controllable results, but on the other hand, these may result in more unexpected artifacts and glitches on the image. Learn what's adjustable and fixable. Through community images and pay attention to prompts. Get inspired, learn photography terms, styles, and gear. Avoid negative prompting in the actual prompt. But if the tool provides those ugly artifacts like fingers or hands or something ugly that you don't like, and the tool that you're using allows you to add separate negative prompt, then use that negative prompt and input everything that you don't want, for example, weird fingers, crossed fingers, and so on. So now in the end of this lecture, I want to just show the image results from different types of complexity of prompt, and the examples of the techniques that we applied. So yeah, let's turn on some music and watch. O 14. 12 Practice Activity: Now we've learned how to use the most important prompt, engineering approaches for marketers using Agipet. Really, passive learning without practice and feedback isn't as effective. I want you to take the most of the time you spent with the course. Here's an activity for us. Define the goal of your prompt, then use the prompt engineering techniques that we've discussed in this section. Create a prompt for your typical task. Try to combine different prompt engineering techniques if needed, then submit it in the Q and A section or just send me a message if you want to be more private, I will provide you with my detailed feedback on your prompt. Sure, this won't take as much of your time, but it will definitely improve the way you use these prompt engineering techniques and how flexibly you can think about them. That's it. See you there. 15. 13 What Are Custom Instructions for ChatGPT and How to Prompt Them: Hi everyone, I'm excited to finally talk with you about apt custom instructions. Let's talk about their benefits, the limitation. We'll have a few presets for custom instructions. The general Use 1.1 that you can customize for yourself, particularly to enhance and automate your Chap functionality. To make your marketing efforts even more efficient, let's start. What are Chagpti custom instructions? Customer instructions are a new feature still in beta added to Chagpti by open AI. And it makes ChagpTi consider your personal instructions for every prompt, both information about yourself and the way you want Chagpt to respond. The great thing about customer instructions for Chagpti is that they are available for all Chagpti plans. Both free one and the plus one can be accessed through settings and beta window on Chachi PT website. The moment it's actually the competitive advantage of GPT as Google, Bart and Cloud don't have this functionality yet. What are the benefits of PT? Custom instructions. You can tailor ChagPT responses to your recurring needs. You can save time by typing less things all over again, which is already a bit annoying. Sometimes it can improve your communication, make it more efficient. If you test really well, it can even make ChagPT sound a bit more like. What are the limitations of Chap custom instructions? The main limitation that I can't handle with some kind of a workaround is that GBT custom instructions don't follow your instructions partially or completely. Well, the future is still in beta and we can expect it to work better in the future, but it's still be beta can work better anyway. Let's move on. There is no preset switch yet. If I have a couple of sets of custom instructions, I can switch back and forth between them quickly and I can change them within a conversation. Sometimes that's a bit uncomfortable, but maybe in the future iterations that would be fixed open. Ye okay, let's move on. Your context may change more frequently than custom instructions can handle. For example, if you change the context within a conversation or you need your conversation for a few different purposes and a few different contexts. Then I would turn the customer instructions off and then set the dynamics myself. We can define two different types of custom instructions and marketing that could be either very general marketing use. You can consolidate your personal level information, organization information, the industry you're working on, and can help you maintain that consistency across all of the tasks that you complete. We'll have a look at how it works in a few seconds. Specific use, you can customize for your very specific job function that's repeating. For example, for your PPC workflow or your copyrighting or management or data analysis. Here are a few tips for creating effective custom instructions. Maintain a singular focus when you take a photograph. In most cases, you want one object to be in the center of attention. It's here, avoid merging multiple specific uses in a general purpose custom instruction. Avoid any contradictions and oximerance. Don't ask VT to be both analytical and super duper creative at the same time because the responses will be very mixed and confused. I would also recommend to outsource your repetitive tasks. Think of things that you repetitively type into ChargPT or things that you currently do yourself. How can you allocate them into the customer instructions that you don't have to explain things all over again. The shorter customer instructions, the better. I suggest that you think. As an SEO specialist, think in terms of keywords. Optimize the keywords for better word to meaning ratio. You need to have less copy but more actionable words. More things that Gibt's language model can stick to when choosing how to answer. That way, you'll have the most stable results. Actually, it's something that I noticed from many thought leaders in this AI area. When they share their custom instructions, they give long sentences, very emotional. But in fact, you could cut that sentence into three to four words that chip would understand. For example, asking it to give no disclaimers. Yeah, don't forget the limitations and the amount of symbols at the moment. It's 1,500 symbols in each of the boxes are given here for a reason. One more important tip, and I think that's the most important one. Test and calibrate every change that you make to your favorite custom instruction. Don't get tempted to change multiple things at the same time because that way you won't know what worked for you better as the feature still doesn't work 100% stably. Test your changes with ten to 15 prompts that you regularly do and refine them accordingly if you see that something doesn't work. Here's the structure of general purpose marketing, custom instructions that we'll work on in this course. We actually have an example of such instructions that you can simply copy and paste into your Chagupet. You can access it in the resources of this course, I will show you exactly where. Here's the structure of a custom instruction that works for me. I tested a lot of them. In the first section, you tell about your experience, your background, your industry, and typical tasks that you will give. You can also say what kind of answers you prefer all of the time. But that would be something that I would ask you to customize for yourself. But later on in the second section, when you explain how you want Gibt to respond to you, you start by introducing the role. Then you describe the formatting and response style, then describe the reasoning process that you want Chagipt to go through. That's actually the place for your favorite prompt engineering techniques that you can also actually get from this course. And there's a whole dedicated section to prompt engineering techniques in this course. Feel free to add those techniques into this part of customer instructions, especially if there's one that particularly works for you most of the time. For me that's chain of thought. Then you can explain what you want in each case. For example, you can do if conditions, if this, then do this. But don't go too deep with it because that might introduce a contradiction or conflict. When you ask for things that contradict each other, then last but not least, limitations. For me that's the most important thing because my flow with age pet consists of more removing stuff and clarifying stuff then adding things. I want chat to refrain from its cliches. In the copy that it often uses, there are words that you knew that are sometimes used. But then with the release of the popularity of Chagpt, you see those words and phrases in every single post, and it's such a marker of AI generated content. In the end, we'll discuss a couple of useful add ons that you can add on top of your customer instructions. But I actually removed it from the preset that I use. There is a reason for that as well because I don't want to waste the tokens, but I sometimes go back and paste those depending on what I'm working on. Now let's talk about making custom instructions while custom specific use for yourself. In the first section, you want to describe your current situation, your profession, your product industry goals, background target audience, your activity by activity goals that you're working on, The typical tasks and maybe even KPI's if they can be understood by PT, maybe without numbers because you don't want to ruin the confidentiality as an option. You can add what types of answers you like. The second section would go the same pattern, but you just want to customize it all to yourself. The role, the response style, the reasoning, the conditions, the limitations, the structure stays the same, the details would differ. I recommend to focus well more on your needs. In this case, something more specific. For example, create customer instructions for PPC or for SEO work that you're doing. So keep in mind this structure, but just tailor it to more specific parts of the job that you're doing. All right, in the next video, we'll have a look at a demo of how our general purpose customer instructions work. I'll share my prompt actually taken as a preset from this course, we'll go to PT, use GPT four and have a look at how this works with customer instructions and without customer instructions. 16. 14 Custom Instructions for ChatGPT Demo: Here we are in the prompt book of this course, the resource that you can always refer to. Go back here, Let's go to the custom instructions for marketing specialists here. The custom instructions that I have here, important part, we can describe the tasks that we typically give to describe the answers that we want and our ability to understand information at a certain level. I don't want GPT to be forced to provide all the information in a beginner friendly form. And I also describe what I'm going to be doing now. The most fun begins in how you would like judge to respond. I give a role, then positive rules, the formatting, I want AP format, for example, For better visual and psychological perception of my content. I want the responses to be organized and marked up visually, more for my perception. Because I understand whether the answer is relevant or not, or good or bad for me, and I can skim it really fast. I want the breakpoints between the paragraphs and so on and so on. You can pass the course and read it, or you can actually go and copy and paste and play around with these custom instructions, which I really encourage you to do. I added the chain of thought prompting. I think it's very adaptable to many use cases that we can have and when it's not possible and not needed chat won't trigger this custom instruction, the negative rules, I also use them because these words have become such a cliche of chat. I instantly see that it's a chat post. Uses these words in the post simply because, well, there's nothing wrong with this vocabulary, but it's just over used by Cha PT recently. Every single answer that I asked for, it just puts one of these words maybe as a signature or maybe because it's just so popular. I don't know. But it just happens and I don't want to sound like everyone else who generates texts with PT. Let's move on now. Let's zoom in and have a look at the prompt. This is actually a prompt from the copyrighting section of the course, but I included the placement, the topic, the audience, and a couple of arguments to the prompt can also find the prompt in the prompt book. In the copywriting section right here. There are actually a lot of prompts. Here's the result that we received. Feel free to pass the course, to read through the response. I can already see some of the things that are stopper for me and a signature that the copy hasn't been even edited after being generated by ChagPt. First, the Magi everywhere. Rocket log, which is such a cliche in 2023 and we're heading into 2024 already, right? Unleash your true leadership potential. That's the cliche that I don't want here. All right. Understand the difference. Again, unlock these are things that sound a lot like GPT to me. Let's have a look at the actionable advice that Chechipt provided us with. I don't see that really actionable, because delegate effectively, okay, delegate might be, but empower your team and maximize productivity is definitely not actionable. It's very abstract. Develop strong communication. Again, very abstract. Sounds good. And paper, but in real world, it's easier than done. I'm trying to save this post a little bit with my follow up, I want the desire section to be more detailed by adding a recommended read and to rewrite the copy so that the reader can instantly apply these techniques. A lot of words here, but let's have a look. Delegate effectively empower your team to take ownership of projects. It's still not very actionable. It's not something I will do tomorrow when I come to work. Free up your time for strategic initiatives. Again, it sounds very motivational, but not as contentful and actionable as I wanted it to be. One more word from the list of words that are used in every chip Response. Ensure your vision objectives resonate with every team member. All right, Learning adaptability. At least we have one recommended read here, which is already something. It's not a bad read. Okay, now I want a short post caption to get the audience into swiping the carousel. It's not short, it duplicates what we have in the beginning of the carousel here. It's a pure duplication, word for word. I wouldn't do it myself because it definitely doesn't catch attention. We have one call to action, second call to action, that's the same. We have the words from the stop list again again. All right, this is great that we have such a vivid example here, but this for some reason, happens to me all the time. I really don't want the same words to happen in every single post. I'm not saying this response is bad. This is actually something that you can improve later by adding your inputs. But let's have a look at the base. Can we make a better base so that we avoid fixing the same things? Let's go and have a look at the example with custom instructions straightway. I can see that the sentences are sharper. Transition from a to B, that's a classical move. Do last, not more. I can see that these sentences are catchier. They're shorter as headlines. They are better. I don't see that cliche and those words from my stop list everywhere, I'm sure that somewhere they will pop up interest. Okay. Know the difference. Marketing disciplines and managing people are not the same. I like that it put the caps to keep the attention right. The meat of the post. Let's have a look at the meat of the post. Three pro techniques for future leaders. Actionable insights to hone your management skills. Delegate, don't dictate. It's basically the same thing it's create because it's comparable. Assign tasks that align with the team members strengths. Well, this is already much more actionable. I can see that It can be done tomorrow. Okay, And there's a metaphor, and I can see why. Because we have this in our custom instruction. Let's have a look here in our custom instruction, that's this part triggered, all right. Emotional intelligence, recognize and manage emotions in yourself and others. All right? Again, easier said than done, but it's very specific because managing and recognizing motions in yourself and others is already an actionable pattern that you can take. Number three says, regular feedback loops create an environment where feedback is welcomed and acted upon the action. There's the word from the list. I would definitely edit it, but it's just one word from my whole list. In the previous, in the previous option, there were a lot of them less editing for me. Now I want to, but I think we can still a better answer here. I'm going to do the same follow up here. Make desire section more detailed by adding a recommended read and re write so that the reader can instantly apply these techniques. Master these three pro techniques today. Immediate steps to elevate your marketing skills. No weight required. I would remove that. I like the M from APA formatting. Love it. I can see that it's followed. Quick tip. Identify your team's unique skills today. Assign tasks accordingly. Tomorrow, it already uses this rhetorical pattern with today and tomorrow we have the recommended read here and it's relevant and there's this instant application. That's what I was missing in the previous response without custom instructions. During your next team meeting, open a discussion on individual strength. Assign one small project or a task aligned with the strength that is really valuable. And I think that could actually make it to one of my posts in the future. Yeah, there's this analogy from my custom instructions that I didn't ask for right here, but you can see it from custom instructions. I think it's quite usable and it's easy to memorize such content actually. Now, emotional intelligence. Quick tip started Journal already very actionable. Great read and I actually recommend to follow Travis Bradbury on Linked in Writes great stuff, Take note of situations today that trigger negative emotions discussed with a mentor or coach to identify proactive coping strategies. Great, love it and feedback loops. Quick tip introduced by weekly feedback Friday. Thanks for the feedback by Douglas Stone and Sheila Heen. Schedule recurring biweekly meeting dedicated to open feedback. It's something you can instantly do when you come to work. Love, it. Makes such a better post. Now let's write a short caption to get the audience into swiping the carousel. And you can see that this one is actually short. There's a hook that's different from the one that's on the carousel. There's one call to action without repetition. I don't see anything from my stop list. Love it guys. How do you find these custom instructions? I think this does a much better job with custom instructions, especially when you use PT and GPT for model. These custom instructions work seamlessly with the prompts that we have in our prompt book. I'm excited to hear your feedback on how that worked for you see in the couple of minutes in the next lectures. 17. 15 Custom Instructions Customization, Addons and summary: That you have to general use custom instructions. How do you transform the same structure into something more specific? In the first action, you want to describe your current situation, your profession, product industry goals, background target audience. You also want to specify your activity, your goals, your typical tasks. For example, if you're an SEO, your typical tasks would be very specific, not just as you're doing CEO, but your tasks would include writing messages for link building, generating lists of keywords, analyzing keywords, grouping keywords and so on. Enumerate your typical tasks and maybe even KPIs as an option, you can mention which answers you like. Section two would have the same pattern, the same structure, but again, be more specific about that use case that you want for yourself and then save the custom instruction tested. Now there's one more thing that we should talk about when we discuss custom instructions. The addons allowed to modify the response in the end of it, here are a couple of cool things that I want. Here are a couple of cool addons that I found useful for myself. First, in the end, ask two critical questions from the point of view of my CM O Chief Marketing Officer. But that could be any of your stakeholders. For example, that could be your customer in automotive industry or the product owner of a video editing app. Next, you can ask Chajibti to suggest three smart follow up requests to develop the conversation and deliver on the task so that you don't have to think on the refinement that much in order to evaluate your response more critically. You may want Chagpti to respond with pros and cons in the end of its response. Or you can implement self ask technique by asking Chip to ask you three questions in the end and request some information to improve the response even further. One more useful add on is to add a brief summary of the response. It's especially useful when you like to receive long response or you ask questions that result in longer response by Gipet and you want to quickly understand whether it's relevant or not. All right, that's it. Let's quickly summarize. Custom instructions can save time, but until the feature performs consistently, don't expect too much of it. Focus on automating things that you repetitively fix in P responses or you repetitively write in your prompts. It's not actually necessary to use custom instructions all of the time. In cases when your context changes really fast. Sometimes even during one conversation, you might really just turn the customer instructions off while creating custom instructions. Think in terms of context, writing style, formatting, reasoning and limitations. Last but not least, test your ideas one thing at a time. If you don't see that idea applied consistently, just remove it from customer instructions. Because in custom instructions less is actually more. I hope that the general use custom instructions and some of the other examples that we have in the resource section of this course will be really useful to you and you will enjoy using them and that would help you be more efficient. But I also hope that now you're able to construct your own custom instructions in age. Pete, thank you so much for sticking with the course and I'm excited to see the rest of the lectures. Cheer spy. 18. 16 A Better Way to Complete Research Based Marketing Tasks: Hi, I'm just here with a quick update on running research based tasks, like analyzing the marketing landscape or finding some best practices in marketing for different formats, strategies, audiences, and campaigns. Well, first, the Chat GPT already has the web browsing functionality, but there's a problem with it. It doesn't access many websites. It takes a lot of time and it still hallucinates a lot. It's a good option, but it's not always perfect. It's possible that by the time you are watching this, open AI and Microsoft have improved it, but that's just not the case yet. That's why we'll be reviewing the second option. The second option to collect data is to find a plugin that accesses external links, for example, like a link reader. But these plugins are still at the early stage and often produce errors and don't always work the way you expect them to. There's the third option so far, it's my favorite one. It's actually the reason why I'm here with this lecture. It's called Perplexity AI. So it's basically a research based AI tool. And it's like search in GPT with a co pilot, which is also known as auto GPT. I suggest that as you go along the course, you try our research based tasks with perplexity instead of GPT or being. And just have a look and decipher yourself how it works and bad way it's free. Functionality will be good enough for most of us here. If you switch on the co pilot button, it will ask you additional questions and follow ups that are always very relevant. And it will actually make your results so much better in the end. I love how perplexity provides accurate results and references. It also searches videos, which is awesome. So I think it's a much better solution to collect data for analyzing the marketing landscape, for example, or audience information. If for some reason you can't use perplexity, I consider.com which is an alternative AI search solution. I hope that you find it useful and it helps you find the insights that you are looking for much, much faster. Oh, and by the way, I just wanted to ask you something. Please consider reviewing or at least ranking this course. It takes a lot of time to put it all together, to research, to record and edit those videos. And your feedback helps me record more updates like this one, and get more motivation to produce more content for you. Well anyway, I'm excited to see you in the next lectures. And until then, cheers. 19. 17 Quick Editing Tips To Sound More Natural: Today's 1 minute ChagipT tip is about proofreading and editing your copy. For example, for a social media post, ChagPti was taught that it's important to grab attention in the beginning of the post. And usually ChagPet starts writing social media posts with something like attention and audience or exciting news, something, remove that nonsense. I mean, it's too generic and it's too obvious that it's Chagpt second tip. Watch out for sentence constructions with from something to something as it's very typical for Chip. One more construction is not only something but also something else and whether you want to do something or you want to do something else. Yeah, one more thing. While editing those posts and sending them to stakeholders, pay special attention to the hash text that ChagPT suggests and be really cautious about those imogies. Maybe try using imoges for navigation purpose. Not just spread them, whatever Jgpty suggests. So be extra cautious as this might give out a tax generated by AI and it might look not very trustworthy. Yeah, that's it. I guess I'm a bit over a minute, but I hope these were quite useful for the time that you spent with those tips. Cheers. Bye. 20. Final Words: Congrats on completing this training on prompt engineering. You've learned the foundations of writing effective prompt for large language models. Throughout these lessons, we covered the prompt design principles, advanced techniques like fu shot and chain of thought prompting. And we learned how to apply these principles across multiple practical tasks. Stay curious, keep experimenting, and find new ways to save your time with prompting. But also recognize the current limitations and risks. These models require your guidance and critical thinking and most importantly, real life experience. After all, you'll be the one responsible for a real world decision making and consequences. We now have a powerful tool to augment your intelligence and experience. Use it responsibly. The future of prompt engineering is bright, and I'm excited to see what you've built. So share it with me if you want to. Thank you for joining this journey. Happy prompting. 21. Bonus: LinkedIn Outreach: Remember, avoid excessive activity that might be perceived as spam as Linton doesn't approve it. Manufactors, other than the text of a connection note, can signal a SPM. Only connect with people you genuinely want to connect with and interact and actually speak with.