Hi guys, my name is Han. Today we are gonna talk about the prompt engineering. This is one of the most demanding skills for this ticket. So let's start what we have today. Okay, so today we are gonna talk about what is the prompt, what is the prompt engineering, how to write a good prompt what is the technique we can use? How to, how this prompt is evolved and where what is a trend for emerging trend for the prompt? Okay? And we'll see some of the cases, studies as well as ethics and responsibility of AI practice. We are going to cover everything in this today's presentation. So let's start guys. What is the prompt engineering? So prompt engineering, basically I use user give the inputs provided the requirement, prompt, document that or structure that instructions and provide this detail to the a, a l process, the data and generate the output based on the instruction of provided by the prompt and you will get the result. Based on the instruction you've given. So that is in simple way we can say it's a prompt engineering Now. What, let's understand more about the prompt engineering. Prompt engineering is not only the art, it's art and science to generate your desired results. Why we call art, because it give you the desired results and why we call it science, because if you have to have, provide the clear cut instruction structure and context of background behind that, we have to remember one thing better, the input and better the quality. Otherwise you'll not get a good response if do not have the good. Prompt. Prompt. Can be, prompt, can be for in any form, text, visual and audios can be provided. The input can be provided in any form. Okay. Visual, audio and codes. So let's see how this prompt engineering works. So we have to understand the concept for the prompt engineering. But before we go and talk about more about the prompt engineering, we need to understand what is the LLM Large language model, which is. The heart and key of the this one. So next couple of slides, we are going to see how this prompt has evolved and we're gonna talk about the LLM as well. So let's see how the prompt has evolved in 2020. There was a keyword prompt. There, is there, sorry, there is a keyword prompt. We just give the inputs through the prompts and just hope for the best results. Okay. Then we see the then we see the 20, 21, 22, a few short examples or come. So it means like you provided some of the example template to the. To the AI or prom based on that example. In template, it'll generate the response. Then the role based or chain of thoughts has been evolved in 20 23, 20 24. So AI can be act as a role act, act as a plays a role or persona act as a analyst, act as a manager, and and can provide the response accordingly. Then it's further announcement has happened, the chain of thought, thinking about the logical step-by-step process, and it'll provide you the, provide the outputs. And then multimodal prompts where we talk about the the text, audio and different types of, can be considered as an input. This oral, we are gonna discuss in next couple of, next couple of slides where we are gonna talk more in detail about that one. This is the, just a history of how the prompt is evolved. Okay, now let's go and talk about LLL as we discussed earlier. Large language model, it's basically, basically the data which can be provided in any form, text, images, speech. The large set of data has been pointed. Provided, and it's been trained, this clean data will be stored in the vector database. This is typically is the function of the large language model. So large language model is the source for your AI to generate the data. This is what the this is, this is the feature of the ai, which can be develop, developed through the deep learning method we are using to develop the AI and develop this large language model. And which, if we are going, which is used frequently used by for the prompt engineering. For the prompt, let's seek couple of examples. Are there, okay. Like we all knows about, or most of the works are using the chat g, pt, and other AI based prompt engineering tools. So chat, GPT is also used. The LLM large language model. It is a 2.5 trillion parameter, used two huge dataset to perform the different multiple types of functions. Similarly, the bot, it's another one LLM model. For prompt engineering, it used the three 40 millions of parameter to provide the desired results. So Google, Germany is another one example, which is developed by the Google mine. This all are the. All of the example of the state of the art will provide you the very good, accurate results based on the LLM has been developed. So now let's go further. There was one case study was AI powered Financial Reporting Chatbot. So have author one of the article AI Reporting Financial Chatbot, which you can find out on as per the below link. So we have implemented this for our clients or customers. So after implementing, we see the 65% productivity key in 94. 94% of reporting accuracy and the significant faster performance we saw. When you have time, let's go through this article and read it. You will get more insight about the article. Now, what is a prompt? We need to understand the prompt before we talk about the prompt engineering more. The prompt is nothing but it just a call of action. Call to action. What is a call to action is you tell to the AI how what to do, how to do, and sometimes we need, say what to avoid. To get the desired output, that is called a prompt. So prompt engineering, why the prompt engineer matters. The nowadays the AI is used by or prompt ai, use AI and prompt and used by almost in all the sectors, especially the, when you have to write a good detail or documents or research, paper, law, healthcare, business, or. Significantly using this one. But we have to remember, in order to get to write a good good documents and document, we should have, the good LLM or large language model has to be developed accordingly. So good data input will give you the good results. Okay, so better input is the better output. Not only the, so we have to see the quality of the prompt. Good quality prompt will give the good results. Bad quality prompt will not give the good results. It'll give you the give the bias information and it'll waste your time. This one, so we have to focus we'll see some of the examples in the see how the good and back prompt will impact the work. Okay, so this is what you have to remember this to talking to when you're talking about the prompt engineering. You have to remember when you like how to talk to the ai. So the language is, it is called a prompt engineering. It is the prompt, the language which you're using effectively communicating to the ai. So let's see. Why effective prompts are important. Effective prompts is important because if without the effective prompt there is it'll directly influence your quality with the AI response. If you do not have the good prompt, you're not get good response. Crafted prompt will give you the desired results what you're looking for. That one, that is why the importance of the prompt is there. Now, let's see some of the examples. Good versus bad prompt. You can see here if you go and just give the instruction, explain the ai, help ai, so it is a very weak the scope is very high. You'll not get the results, you'll get a very generic results or output of this one. But in order to give us a good instruction to the prompt, okay, you, if you mention explain AI in a hundred words using the simple language, it is a, it'll give you the desired results and with the specific limitation. So this is one of the example and another example. If you say, tell me about the cybersecurity. This is again, a little bit weak. You're not talking, you're not specific because cybersecurity could be the plenty of information available. What you're looking is not very clear. But if you want to make it the good prompt, you can say, list of three, common cybersecurity threat, or explain how this micro Microsoft defender mitigate them. Similarly, you can ask like. Prompt, if I can you write the email for me. It's very generic. But if you ask them to play the role or act as a project manager and draft the email repetitive response based on the highlighted uncertain points, you'll get a better results. So these are the example purpose of this example to show to all of you. Prompt see how you will give the input or instruction based on that. You'll get the response. So we have to learn about how do you need to provide the provide the the prompt or details to the ai. So in order to get the good results. So let's check it. See. So the best way to to give the input to this one, you should remember this. Four points. Okay. So we should provide the context or the background of the work, of work, what background, what you're doing context to the ai, and then the provide the task information, what you want. AI or prompt has to respond back to you and then provide the constant what are the limitations are there If you have limitations so that you should provide. And similarly, to provide any additional information which you want to do. See here the, one of the example is highlighted here. The marketing manager like suppose, okay, so you are the marketing manager for the few new product launch in FMCG sector. This example is talking about you are providing the background about the marketing manager belongs to the certain sector. So this will give us a will help us move to get a good results. Similarly, we. To perform that good ta we AI to perform the task more efficiently. So now the example is written here, the right is short paragraph that describe the target audience for the product and the key marketing messages that will be resonate with them. So this is the task, what you've given after providing. The background and the task. You give it with the example, and then you're putting the constant on that one. The response should not be more than one 50 character and should be focused on the most important marketing messages. So now you're limiting and providing the constant one 50 words and focus on the certain areas. And then additional information like the product is washing liquid for the washing machine, targeted for the urban and semi-urban households. So with the help of this type of structure once you provide to the prompt, you'll get a very accurate answer. So keep in mind this formula, the context towers, constant extend function. This is very helpful to provide the structure, form detail to get a better results from the prompt. Okay. You can remember a few of the works for the prompt engineer, which will help us to give us a more accurate results. Like Act S you can, this is a persona pattern, okay? So where you can ask LLM or ask AI to be, act as a manager, analyst, specialist. So we'll give you the better result. When you say ask for the analysis, he'll provide you the more detailed analysis of the certain. Certain detail which you're looking for, this specific field summarize when you asked to summarize today, I use this word summarize. So then here to summarize for you in the way you're looking for multiple information in in the summary statement, the way you want. So use this word, this words are very helpful. Generate. So it'll generate the output, the way you're looking, the based on the ideas or the details it's given. So it generate the new ideas as, as well as the concept for you. So use this so it'll help you predict. So based on the current place, the data level, he will predict. Or train for you. So you can use this word, this, these are the words which will help you to give a more record results, elaborate more. So you can have asked to AI to elaborate more in simple language. So on particular topic, on particular point, on a specific specific area. So it'll help you to give us a desired results simply. And the last one is nothing but more than like simply simplify. So it can provide the result in the simple language so everyone can understand easily. So these are the power works. Along with what we saw, how to write a prompt along with this power box, if you can use that will be very helpful for you to get the good results. So now let's go to the next one. So there are core, prompt techniques. There, there are many techniques which can be used to get the better results. There are a few techniques are listed here, like role prompting as we already talked about. We can ask AI to be act as a pm act as a lawyer, act as analyst. So these are the few examples there. Role-based prompting. So can we play the role and give the good results chain of thought. It's a step by step process the way human think. You can provide the instruction and you can expect that result as per the. As per the instruction given, so it can solve your complex problem, I think with the step by step process. So you should know how to instruct to the prompt. So this is the chain of thought. We can use the technique to get the better results You could have multiple time. You need to give the instruction like step by step process. First time you give certain instruction based on the output, you will go the second set of instruction. Then third set of mention. I will give you the good results based on that one. All together come by few short is just another one case where you can provide the example based on the example, AI or prompt will return you the the response. Zero short nothing but just a straightforward just to provide the instruction based on the instruction able will like a response back to you. So these are the core pro technique. Which you can use for to get a, get the better results. So we see so far the three way. Okay. You can use the provide the context, background information, more details to them, okay? And we can use the the power ws as well as we can use this prompt technique to get the good results Now. The prompt quality. Okay. When we are talking about the prompt quality, we have to remember one thing garbage in, garbage out. So good prompt will give the good results. Bad prompt will give the bad results. Before, before we like what, for what is the best way to provide the details to the prompt. Keep in mind the goal and audience first. Okay? Keep it. And then accordingly, you can provide the information, provide the more background and data point as much as possible as a part of the format, what we talked about, background context, this one, and to how to mention very specifically, you need the desire in which format. You need the bullet points summary, PPT or Word document. They, the AI will be generate all this for you. Add the restriction because you're not adding the restriction. It'll be the open in it. Either the length. Okay. Tone, perspective or any other thing, which you are looking a specific thing from as a part of your response. So these are some of the points which you need to be which you want to make sure you are using. This, keep in mind this point for using the prompt. Okay. For accuracy to avoid any biasness. No. We talked about prompt. We see how to write the prompt, but if you, we see some of the good example, good and bad example, how to write the prompt. So there is success story based on the prompt engineering after evolving of the prompt engineering. There are a few example, I'm, we are gonna talk today. There are plenty of them are available if you to. Go and search through the AI or go in Google, you'll find so many examples as there. One of the, one of the examples of most of the customer support or fortune fin aid companies using the AI assistance or chat bot for chat bot and a other AI for ticket resolution, we see the 14% increase on their average resolution. Similarly, the MIT able to with the help of structure form, they're able to produce the better research articles and documents, so the quality is significantly improved and save the time. Also. Similarly the PCG has like b, c, G mentioned, like even the non-technical people can solve the complex problem with, because of the prompt engineering is available because of the prompt, good prompt is available. So this is another one cases there. Okay. And we see the McKensey has developed the product library, which can help us to go code receptor and identify the root cause. It saves significant time as well as money for them. So these are the few success stories out there. Many more could be there. This is the like, how to use the prompt. So you will, you really get a good results of the prompt. Now let's review some of the story, which is. The failure cases. So as I mentioned earlier, okay, the garbage in, garbage out. If you do not have the good LLM okay, and you do not have the the good LLM, then you'll not get a disaster if you don't know how to use the prompt. Also, in that case also the, it could be the case of failure. So one of the airline, chatbot, give the example like, give the misleading information to the customer and the lawsuit has been filed and the airline paid the huge penalty. This is one of the information we are using the the chat bot, because might be the chat bot given the misleading information. It's not trained properly. With the LLM, this is one of the example. The another one, the sensitive data leak example is happened where someone is. Accidentally share the code to code confidential code with the, with open AI or with the with the prompting engineering. Okay. So that become the issue. And then later on the organization, block the chat GP within their own organization. One of the, one of the very, very interesting case with the terms of service agreement has been, updated by by by automated ai and they have mentioned on that. The the chatting and other information, customer information can be used to train to the AI without their concern. Consent. So this has, again, become a big issue. It was came in the news also. Article. So then later on they have automation corrected, that problem. The one, another one thing the, if you're using the AI to, for like example for hiring. Okay. So one of the case, what happen in hiring the the AI will. Was a bias and providing the information based on the age insects. Okay? And that could create the issue. So later on, this has been blocked by the Ian who are using that ai. So this one, so you to remember when we are getting the. The benefit of the AI that we could have the issues, also the ai. So how to use, what process we need to use. What is the ethical way of working? How to write the prompt to get the good results. We have to make sure every time we need to validate that results. Since you see some of the cases of failure and success, and we earlier saw some of his example of good and bad prompt. This could be the cause of the success and failure. Now let's go and see. So we so far we talked about lots of things about the prompt, how to write now, how to write the prompt and now see looks, how, where the trend is going now, future trend we are talking about. So ums. Like we need to, like in the next, next 24, next three, four year, you will see the more governance will be coming to the picture. Talking about the more about the standard audit more audit is going to happen because it, AI can be used. According to governance and audit has to be happened to make sure the accuracy of the AI to avoid any issues. AI going to further going to use, like that's a multi-agent going to closely going to work with the copilot that is q See it's in, it's. It's, we'll see in the next 20. You only going to see in the next four to five year it's going to more, you'll see the progress on that one multi-model. We already using text, images, and audio. Now you'll see the started combination of that one be it's coming in the future. Ai so. Can be given the instruction on multiple modes together and get the response in a different way, the way you're looking and the most important dynamic learning. So it can be, AI can be learned as on the fly adoption that you see this changes is going to happen in the next three to four years. This is the trend. Now let's take a pause here and because we talked a lot about the prompt, so what we covered so far. Okay. We talked about what is a prompt engineering. We talked about what is a prompt, how to write the quality prompt. We talked about provide the good detailed background context. This one. Pick the pick. We can. Use a different technique. We talked about like chain of thoughts, prompt and other techniques can be used for writing the good prompt, what tool and framework we need to use. And and most important thing about this one, the ethical part, which we're going to cover in the next couple of slides, like how to, what could be the how do we need to work and how do we human need to validate all these ones? So these were the, so far, the key takeaway. Let's go forward. So ethical, we are working, so we have to remember. AI will give you the lots of input. So you should be transparent with your client customer or anyone whom you're working. If it is if it is ai, AI contributed work, you need to mention clearly it is AI generated work. It could be edited by you after generation. Taking the help from the AI is not a bad thing, but you should be transparent about that one. Communicate accordingly. Avoid claiming the AI genetic content belongs to you. Clear. Please make it big. Be clear. Okay. And to your client, customer, or anyone? Like it is the age united contract. So you can, you, it should be good for everyone if you mention this one. So we talked about this one and. We should not use the unbiased wording while writing to the prompt, because that's very important. This is the basic ethics we have to follow while using the prompt for for ai. So we should, we all are responsible for following the ethics and develop the good community to how to use the ai. Okay. Now let's see. Governance and safety. We already talked a little bit about in the previous slide about the government governance and safety. We have to be very clear about how to handle the data. So some or and retention policy that has to be defined very clearly within the organization as well as outside work. So we should have rules and regulations should win. Place to avoid the misses of the policy. Train your team or train your peers colleague. On other members who is using this one ethical way of working. Okay. And make them responsible for the good, using the prompt accordingly. So output, AI generated output, it's, and make sure you are verifying that AI generated output it. Can be a bias based on the LLM is there. If you not human need to verify that output. Otherwise it'll be, so we have to make sure all the AI generated content has to be verified by human. So you have to crosscheck with this one. So this is what the most important thing for the governance and safety point of view. Now, the, all the data presented in this. Presentation are you can see the reference and data sources are available here. So it's all the data we are presented from the certified places we pull the data. So it's talking about that. Now I want like when to talk more about what, how the prompt engineering impacting your work. Okay. Or your current workflow. Okay. How your what is your experience about the prompt engineering? We'll talk about some time and we'll take some of the question about HL policy and what should be the perfect prompt for you and how is your experience about this one? So let's talk sometime during our meeting. We'll more deep dive about this one. So look like I have covered most of the stuff here. I think so. We are good for now. If you have any question, please post the message to me. We'll try to answer that question. Thanks for joining. Thanks everyone. Thank you guys.