A brief guide to prompt engineering

1. Introduction: Hello and welcome to this brief guide to prompt Engineering. My name is David argument, that is, and I'm a software developer n mathematician. So why from the engineering, AI is now a crucial component of bold or daily lives and our businesses and the fast development AI tools over the last few years has had an unavoidable impact on our daily lives and only keeps getting smarter and more functional. So this technology has altered how we communicate with people. A robot's, the communication between humans and machines will need to improve as this evolution progresses. It can get us one step closer to realizing the full potential of AI if we properly comprehend how to interact with it. And a theorist salt, we will be able to get fresh insights and extract better information, increasing our knowledge of a variety of topics. So understanding proud engineering is crucial to obtaining these benefits. 2. What Is Prompt Engineering?: So what is prompt engineering? The ability to communicate with AI effectively is crucial as we set. This involves writing Prout's that serve as commands for the AI. And prompt engineering is the process of creating inputs that remind the output of a language model, Lake Chad, TBT. And to achieve good results, it's important to provide high-quality inputs. On the other hand, only the fine prompts can lead to inaccurate or negative responses. Prompt engineering covers a broad range of applications, such as chatbots, content creation tools, language translation tools, and virtual assistance. However, you may be curious about techniques utilized by the AI technology to produce its responses. So let's learn how these models work. 3. What are LLMs?: So what are some going to cite the Microsoft here is tense were large language model. And this is a term that refers to AI models that can generate natural language texts from large amounts of data. Enlarge language models is deep neural networks such as transformers, to learn from billions or even trillions of words and to produce tax on any topic or domain. Large language models can also perform various ventral language tasks such as classification, summarization, translation, generation, and dialogue. And we have examples here. The most famous one is GPT-3, but there are others like bird eggs allow Excel net and a little, a little a. I'm not sure if I pronounced that correctly, but yeah, the L stands for large. In the case of LLM, it means really, really large, can be millions, billions, or even trillions out. The L stands for language, which refers to the fact that the word sentences by breath leave at the core of how these kind of semantic AI works is it stands for models. And they're high-growth dimensional mathematical representations of large amount of written information. So what does tend to be d have to do with LLM, okay, we already establish that connection there. The chat, typically the system is powered by an LLM AI model, invented Open AI based up on the GPT-3 model. So the model here is actually GPT-3. And right now we have GLUT4, Right? And chat GBD is just the application that Open AI built. So you can think of chat TBD as an application built on top of that LLM. That it has been specifically tuned to engage rural directors chats. 4. Supervised Vs Unsupervised learning: So diving a little deeper on machine learning and artificial intelligence in general, there are two primary learning methods based on the language model. Supervised and unsupervised learning. Supervised learning in bulbs using a labeled data set containing data with a right answer. While unsupervised learning is its own labeled data, requiring the model to analyze undetermined, accurate responses. So typically four or GPT-3 relies on unsupervised learning to generate responses. That's why they don't always have the correct data or the correct answer, because they are not trained with correct answers. Language modeling is a fundamental component of the various AI language applications that enables a model to create texts according to the given prompt. And here we have an image of classical machine learning. We have the supervised learning, classification and regression, ie the urine economies. Mathematician or engineer, you might have heard of linear regression. So that's actually as supervised learning. You are doing machine learning. We're in New York doing regression and unsupervised learning in the classical sense. You can see it this clustering association dimensionality reduction. This is actually a very simplified explanation of supervised versus unsupervised learning. 5. Information Seeking Prompts: So now let's take a look at the different categories of prompts we have. So the most basic one is the information seeking prompt. And these problems are specifically designed to gather information. And the proms mostly answered the questions what and how. It's like we are using Google. So we have some examples here. What are the most popular tourist attractions in Ecuador? How do I prepare for React job literature review. What are the most common types of cyber attacks and how can individuals and organizations protect against them? So you see, this prompt is a little bit more detail of the result we want and we're going to talk about that later. What is the history of the Olympic Games and how acid both of retirement. So all of these prompts are its output of information seeking prompts. 6. Instruction based prompts: Now we have instruction based prompts and you have been using them for quite a few years, is to give instructions to the model to perform specific tasks. A good example of such promises, the use of Siri Alexa or Google Assistant. So Chad DBT is very recent, but you have been using instruction based prompts quite a few years. E.g. when you tell a lexical that you are given structure-based prompt. And he viewed tell tap DPT called that. Obviously, it's not going to know who is that. And it's also not going to be able to make a phone call. The latest episode from my favorite TV show, again, Chad DBT won't be able to play the latest episodes from your favorite TV show because it's not connected to a TV. But Ted TBT can answer. The other examples like provide a step-by-step instructions for assembling a piece of furniture for a flat back kids such as ikea dresser. That's something that Chad GPT, can I give you the answer? Right? A tutorial on how to use popular software programs such as Adobe Photoshop or Microsoft Excel for a specific task or project. And F. But here for a specific task or project, because if, if you'd tell chat the beauty, give me a tutorial of Adobe Photoshop that's going to be too general. You need to provide. What do you want to learn? Specifically? Because learning Adobe Photoshop is very complex, right? The guide for practicing a relaxation techniques such as mindfulness meditation or deep breathing for reducing stress and promoting mental wellness. That's an also an example of instruction bits prompt. 7. Context providing prompts: Now we have contexts providing prompts, a given context and examples to the AI is pretty important. These prompts provide information to the AI to help you better understand what the user needs to serve. False. Here we have an example. If you're planning a party and need some decoration ideas and activities for attendees, you can structure your prompt like so, and planning a party for my child. What are some decoration ideas and activities that yet, and these might do to make it enjoyable and memorable. So what's the context here? The first sentence, I am planning a party for my child. That's the context. Now, the AI knows that you are having a party, right? And it can better give you a better response. 8. Comparative prompts: Now we have comparative prompts. These tools aid in comparing and evaluating various options presented to the model for assisting the user in making a fitting session. So this is very easy. Timber sale is compare a and B, which is a better investment as dogs or real estate in terms of long-term financial growth and stability. And actually, if you do this chat tip of tea or whatever is going to tell you, I'm not going to give you financial advice. These are the pros and the cons. So that's very good because having proximate cause is basically a way for you to make a better decision because you are the one who's going to take the decision at the end of the day. You can use pros and cons explicitly in the prompt, like in this example, what are the pros and cons of using a credit card versus a personal loan in terms of interest rates, fees, and credit score impact. Again, it's going to give you pros and cons. It's not going to tell you, Hey, this is the best thing to do. You are, at the end of the day, the one taking that decision. What advantages does a hybrid car have over gas-powered car in terms of fuel efficiency, environmental it back and cosine. Again, hybrid car has pros and cons. Gas-powered cars have pros and cons. At the end of the day, you're going to have your own opinion base and the pros and cons. So in this sense, tools like Ted to be terrible or impartial, so good for that. 9. 09 Opinion Seeking Prompts: Now here comes the interesting thing, opinion seeking prompts. So before I continue, I must tell you that the AI doesn't have any opinion. Remember, the AI is just trained with a lot of data and that data actually comes from the internet. So the answer is going to be based on someone else's opinion. The purpose of these prompts is to elicit the AIS perspective on a particular subject. One example is, what is your opinion on the use of social media by teenagers? You delete it, has positive or negative impact on your mental health as social development. By asking chat DBT, these exact question, look at the answer. As an AI language model, I don't have personal opinions, but I can provide information and insights on the topic based on research as studies. And then it will give you the answer. Got the question, right. So that's very important to know. The API doesn't have an opinion. Here we have another example. In your opinion, what are the most pressing environmental issues facing the world today? And what steps should be taken to address them. Or in your opinion, what are the most important qualities of effective leadership and whether you believe embodies this quality public or private sphere. Again, I have to remind you, the AI doesn't have an opinion. This is people from the Internet, something. So there can be bias, which we're going to talk later. 10. Role based prompts: Now let's talk about the role-based bombs because these are the most important prompts. There are so important that the official API for Chet GPT assumes you're going to use these kind of routes. So if you're into that, you can read the API documentation and you'll find that these role-based prompts are very important. And in general, if you do let role-based prompt every time that you are fine because it's very useful at treated has worked for this particular category, is making the use of the five Ws framework. Cool is the first one. Science or role you need. The models play a role like a teacher and of all birds have and so on. What that refers to the action you want the model to do when you're at this art timeline to complete a particular task where it refers to the location or context of a particular prompt. And the y refers to the reasons, motivation or goals for a particular prompt. And usually include information about why do you want to learn. We have to be specific about that, that duration of your learning period and your learning goals for the prompt, providing more details we will result in more personalized or rebel event. Make fan, please make sure to read this in English. This is not required, but it's better if you do it in movies. In general, LLMs will work better if you do it in English. You have Translate tools that can help you with that. Again. So let's see an example. The coup marketing manager. What create a new social media campaign when next quarter, July, September or whatever, where targeting North American market and the y increased brand awareness and drive sales. So the prompt can be something like this. As a marketing manager, create a new social media campaign targeting the North American market. In the next order to increase brand awareness and drive cells, you have the $5 there will be responsible for implementing the campaign. What platforms will be used? When will it launch? Where will it be targeted and why it is important for the company's goals. So you have to be very specific. Only if you ask this to chat, to pity, you will have this answer, which is pretty long and it's pretty detailed. And I think this can place a lot of jobs. So you have to learn how to prompt these kinds of effects. 11. Perplexity.ai: Now a little warning. It's important always in every category, in every prompt to verify the accuracy of the models responses. If you're uncertain about the subject matter at home, if you didn't know about the topic beforehand, you have to verify responses because it relies solely on the model's output may not delta correct information. Since the model isn't always accurate, be sure to cross-reference the information with other sources to bother. It's a curious. So how do we achieve this? Chat DBT doesn't give us the answers, right? So we have these tool called perplexity.ai. So you can go there to the website. And it's basically like tat typic d, but d of z sources. Also Google bar, which is like the competition for chat. Tbt, will do this, but it's not right now at the time I'm recording these videos. It's not Bartlett to everyone. So e.g. I. Asked Amy is create sharp, give me five websites to search worth psychology articles. And it gave me these things. And it's nice because you can click on these links as well, right? And if you click on these brackets here, It's going to give you the sources of where this thing collected the correlation, right? Because this perplexity tool can also connect to the Internet, right? And that's useful because it doesn't have that plenty 21 cutoff, knowledge, cutoff, like chat to pick a tree. 12. Stop using the Google Search pattern: Okay, now we have seen all of the categories prompts. Now, how do we make these prompts effective? So first of all, we were taught by Google because Google is a big company. It has been in the market a lot of years. And they have improved their search engine so that it works with less words, right? So the last iteration we get to Google, the better e.g. u. Dot as Google. When did the French Revolution take place? You ask French Revolution date and that's it. And you can also put quotes so that you can have exact matches and all that stuff. But we have to totally forget this way of searching information with chatty PT or other applications built on top of LLMs. Because it's the complete opposite. We want. Now as much information as we can provide AI to give contexts. Example, the five W's and all this stuff. So please forget about searching like you were doing it with Google. So the first thing, clarity. If you are in a relationship, you'll find that clarity is very important. And in every type of communication, actually, clarity is pretty important. So as it says here, clear communications crucial in any setting, including prompting nunnery. So to create an effective product, it's important to clearly define your objective, is will ensure that the AI can precise responses to your proud. 13. Context And Limitations: What other thing is active? Given context and examples, as we showed in the context providing prompts category, supplying additional information can see is the AI in comprehending the tenant goal of the problem, which made yield more presets for salts. What else? Set limitations. So the AI must be given boundaries to operate within his increases accuracy and avoids irrelevant the provision. And I went to give you a nice trick. You can set limitations while this is not the only way you can do it, right? But if you put t l colon, the semicolon, semicolon at the end of the prompt. Then you will have like the too long didn't read version of what Jabhat, of what you want. So give me a summary of the French Revolution. The French Revolution obviously is a big historic event. So a summary is very ambiguous. It can be ten pages of PDF and that will be a sub-array. But if you say the are too long, didn't read is going to give you a short paragraph of what the French Revolution is. 14. Break down queries, rephrase and iterate: What else is effective? Rake down queries. These binding queries into smaller, more manageable blocks. Canon has AI ability to handle the information. By doing so, the model is able to grasp each query bearer and generate improve responses. So what does this mean? You don't ask too many questions at once. The AI is going to work better if you ask one question at a time. Okay? Now, iterate and rephrase. If you're unhappy with an AR response, try rephrasing it and provide more context or Martin samples per improve results. If you want to do this, you can obviously copy the proud and paste it again. But actually in chat deputy, you're using that. You have these little button where you can edit your prompt as submitted again. Okay? Another thing is requests to step-by-step explanation. If you require in-depth details were a breakdown concerning a complicated topic, you can frame your prompt in a manner that directs the AI to provide comprehensive as answers by dividing inch state. And this actually pretty, pretty useful because Lee humans understand better when we are given instructions in an order matter. So the AI is capable of doing that. So don't hesitate in asking a step-by-step explanation of some procedure or some learning. 15. Prioritize important information: Another thing you can do is to prioritize import that deformation. Highlight the most important information the product. By doing this, you're telling the AI develop you some providing responses that are relevant to the heightened edit inflammation, e.g. here, and make it a list of the best soccer players. However, the best soccer players are massacres syndrome, Alto, Maradona ballet, you know. But I'm saying put at the top, the younger players. So it will switch the response. It will provide my response where the younger players are the top of policed and this only giving me ten, I can ask more. And probably it's not going to list a messy or persona and Aldo, because there are more younger players that are very good. So the first one is urban Callen, which right now he must have like 23 years old. And again, always check, double-check the answer. He doesn't play in Borussia Dortmund, n bar. So this answer is partially correct because Arlene Cowan is the best younger player, but she doesn't play in Portugal. Then you have killed and then by pair which is still planning periods syndrome. And at the time of recording this video and these other guys, I don't know too much about soccer, so excuse me. 16. Be careful of the bias: Before we finish this brief guide, I have to tell you the AI bit fulls and limitations that you have to take into consideration. The most important one is definitely bias. The accuracy of machine-learning algorithm depends on the data provided by his data can lead to biased output, highlighting the need to review that just print data for possible biases early in the process. So this image summarizes it. If you put this thing in, you're going to receive thing out. That's why you will hear in the news that all the AI is racist or it's discriminating some more. And that's because there are lot of things like this in the Internet. So we didn't have control about that. We didn't have control about what people put in the Internet, right? So we can't do anything about that. But there's another kind of bias, which is pretty, pretty color. It's important to keep in mind that when interacting with a given incorrect information may lead to the I agreed with you, even if you're wrong. And this has happened to me a lot of times where I insanely that I am correct, but really I am not. So it's recommended I have some understanding of the topic before asking the eye. Again, you have to double check the sources. If the AI provides an incorrect response, it could be helpful to rephrase the question and provide additional context. So again, we have control about this. Don't think you are the smartest kid if rho, if you are not certain about something, than try to rephrase your prompt so that Tad to pity or whatever is lot unconsciously agreeing with incorrect information, if that makes sense. 17. AI doesn't have feelings: Another thing that is obvious is filling the API doesn't have feelings. It might, you might think it has been because sometimes ted TBT tells you, Hey, I'm sorry, I apologize. It's not feeling anything. So I often struggled with complex language and in isolated human emotions since it lacks the ability to feel its decisions pertaining to typical human behavior may not always be accurate and reliable. So don't be surprised if it gives you some nonsense or very cold hazard problem because it's not going to take into account the fields. However, that doesn't mean that you cannot rewrite the tank with a certain tone. You can re-read the texts so that it seems that you are happy. It's syndrome, work professional, more sad. So eight eyes are capable of doing this. You can achieve this with this quilt bot, which is a Chrome extension. Or you can do it inside TO may app, which is actually the app I'm using. Procreate these slides. 18. Conclusion: Okay. So what is the conclusion of this brief guide? Well, I'm not good at running completions. So I asked chat TBT to write a conclusion for an online course about prompted generating. In conclusion, this online course has covered the essential aspects of prompt engineering, providing a comprehensive understanding of how to create effective prompts that the illicit desired responses from users. That's right. We have explored the key elements of prompts. He couldn't language, timing and context, and learn how to tailor prompts to different user groups as situations. Through practical examples and exercises, we have gain hands-on experience in design problems that engage user and facilitate the side behaviors. By completing this course, you now have a solid foundation in prompt engineering and are equipped with the tools and knowledge to create effective prompts of Dr. users engagement achieve your goals. Whether you are the sign-in prompt for a website or other digital platform. The principles and techniques covered in this course we will be valuable assets in your toolkit. With practice and experimentation, you can continue to refine their property engineering skills and create even more compelling prompts that delight your users and drive business success. Pretty nice. I hope you liked this brief guide. See you in the next course.