Prompt Engineering Full Course 2026 [FREE] | Prompt Engineering Tutorial For Beginners | Simplilearn

And welcome to this comprehensive video on the essentials of generative AI, prompt engineering, and ChatGPT. Today, we are going from the absolute basics to advanced techniques that the top AI engineers use every day. The world is changing faster than ever, and there is one skill that is separating the leaders from the rest, which is generative AI. If you have been using ChatGPT just to write simple emails, you're only seeing the tip of the iceberg. Whether you're a student, a professional, or a tech enthusiast, understanding how to talk to AI is no longer optional. It's a superpower. Right now, every industry from healthcare to coding is looking for people who can master these tools.

This video is unique because we are not just only talking about theory. We are looking under the hood of large language models, understanding how tokens work, and mastering the science of prompt engineering to get exact results you want every single time. To make sure you get the most out of your session, I've broken it down into a clear step-by-step journey. First, we'll discuss the foundations: what generative AI actually is and how it evolved from basic machine learning to a powerful LLM as we see today. Second, we'll dive into the art of prompting. You'll learn professional techniques like few-shot prompting and also chain-of-thought to make AI think more logically. Third, we'll explore the full power of ChatGPT, including the difference between the versions like GPT 4.0 and also how to use multilingual features like voice and image analysis. Fourth, we'll go beyond ChatGPT to look at largest cutting-edge tools like Google Gemini, Sora for video generation, and also Grok for lightning-fast processing. And finally, we will wrap up with must-knows of AI ethics, biases, and how you can start building your own AI-powered projects. By the end of this video, you won't just be using AI, you'll be mastering it. Let's get started.

Also, just a quick information: if you're interested in diving deep into the world of generative AI, then this course is perfect for you. The Applied Generative AI Specialization by Michigan Engineering Professional Education will guide you through building and deploying GenAI-enabled applications with hands-on training tools like OpenAI, LangChain, and Stable Diffusion. In this course, you'll not only learn the theory behind generative AI but also gain practical skills by working on real-world projects. You'll master the art of creating AI agents that think, plan, and act autonomously, bringing innovation and automation to any business. With industry experts guiding you through the learning process, you'll have all the support you need to succeed and stay ahead in this rapidly evolving field.

One of the key benefits of completing this course is earning a certificate of completion from Michigan Engineering Professional Education. This certificate will showcase your expertise in generative AI, enhancing your resume and making you stand out in the job market. Not only will you gain the skills to develop AI-driven applications and work with modern AI tools, but you'll also be able to demonstrate your commitment to staying at the forefront of AI innovation. By the end of this course, you'll have the knowledge to create powerful AI applications that can actually transform how businesses operate, and you'll have a recognized certificate to prove your expertise. So, are you ready to take your career to the next level with generative AI? This course will give you the tools, the knowledge, and also certification to do that. The link is given in the description box below and in the pinned comments.

Before we get started, here's a quick quiz question for you to answer: What is a primary function of generative AI? To create new content like text, images, or code? To provide hardware support for computers? To act as a physical robot for household chores? Or is it to strictly store and search through databases? Let me know your answers in the comment section below.

The topics for today's discussion are your generative AI. So we have been referring to this term quite lately, right? And every day, you could see generative AI, ChatGPT. Okay. So I asked this question to you guys. So what do you think? Like, if I try to ask you a question like, in your mind, what is generative AI? What is ChatGPT? And in terms of a layman, how could you explain to me that if I ask you that what is generative AI, what is ChatGPT, what are large language models? So what, what, what do you think like, what is the answer?

Okay. So let me elaborate in terms of like, what is generative AI? Okay. Or basically, the term generative AI, what it stands for. Okay. So let us try to explore the, so in the terms of machine learning, so this is your generative AI. Okay. So then, then we have your deep learning. Okay. Then, okay. Then we have your machine learning. Then comes the artificial intelligence. Right? So basically, in the GenAI, so here we could have your large language models. So we could say that generative AI is a subset of your AI, right? Because AI is the bigger picture, and AI involves your machine learning, your deep learning, your generative AI, right? And we all know that deep learning is used to handle your complex tasks, okay, by the use of your CNN, RNNs, right? Yes, hi Radha, how are you? Okay. So, what those who have joined late, so I was discussing that, what is generative AI? So generative AI is basically, it is a subset of AI which aims to have deep learning, machine learning, and AI terms all involved in this part. Okay. And deep learning, if you talk about in layman terms, is that it is to handle your complex data, which could be your image, could be your text, right? And it could be served by using your filters, kernels, and your CNN convolutions, right? And your RNN, which could be used for analyzing your time series data, okay, and your textual data, right, text data, right? And on the top of that, there could be transformer models, like which could be, uh, uh, uh, here, which would lie in between of your deep learning and generative AI, right? So those who know about the transformer, okay, so basically, it is an architecture, which is your deep learning architecture, right? Which involves your encoder-decoder process, okay? So if you don't understand, so forget it, right? Okay. So I'm not going into the depth of the transformers or all this thing. So here, we'll focus on the generative AI and the applications involved in terms of generative, and we will talk about what is generative AI, right? Okay. And ML, we all know that it is a statistical analysis, the models which we have already done, right? So trying to use pandas, your NumPy, your Seaborn, right? So these were the libraries which were used to process your data. And then we could apply any of the machine learning, such as logistic regression or your, uh, uh, linear regression, right? Or your XGBoost, or your bagging, boosting, or your random forest. So these are the algorithms which could be used for classification or your regression task, right? So to predict samples, so these were to predict samples. Okay. And in the deep learning also, we have classification. Okay. We have regression. Okay. And there's another term involved in which deep learning could solve, like your text summarization, okay? And your sentiment analysis. So these were the tasks which deep learning could solve. But the GenAI introduction of the GenAI or the generative AI, it changed the overall picture. Okay.

So I try to give a, this thing that, let's say you have an input text, and you feed it to some GenAI model, right, generative AI, right? And it will try to solve multiple tasks. It could be your classification, it could be your summarization, okay? It could be your sentiment analysis, okay? It could be your code generation, right? Code fixing, right? So these tasks could be solved by using your generative AI model, right? So this is basically a model. It is a predictive model, which aims to give you the next token, next token generation, right? The term generative, generative is to generate new or the next token, right? So that is why it is being termed as generative. It is trying to generate a new text or the new token. Okay? So you try to put an input text. Let's say, I want to go to, okay, dash, and you feed this input text to a generative model, and it will give you a long list of suggestions. You can go to a school, you can go to a college, you can go to an office, you can go to a market, right? And whichever the probability of these tokens. So these could be called as tokens. So this could have your probability of, let's say, 96. This could have a probability of 0.5. This could be your 40. This could be 30. And out of this whole list of tokens which you are getting, this has the highest probability. So that is why this could be your predictive model, which is trying to give you the next text or the next occurrence. That is why it is called a probabilistic model. So this is basically a model which will try to give you the next token occurrence or the next token generation, and all this text is being called as token. Okay? So don't try to confuse the text with the token, right? So text is something that I is a text, and token is a numerical representation of that particular text. Let's say, "I" could be represented in terms of 56, because 56 is a number, and computers understand this language, not this language. So internally, what we do is when we try to talk in terms of coding, so internally, this text is being converted into a token, which is being fed into a generative AI model. Okay? And on the basis of that, these tokens have a probability, and the highest probability, it will try to give me that particular, like, result or the output that this is highest out of this whole corpus of the token. This is having the highest value. Okay? And basically, when you are trying to give more text, it will try to generate it, and it could be a summarization task, it could be a classification task, or a code generation task, right? Based on those applications, it will try to give. Got it, got it. A general idea, because like this is a very broad and a wide topic to discuss, right? But I have, I've been trying to give you a basic list that what is a generative model, generative AI, or a generative, like, how does it work basically? When we will also firstly try to see the applications, right? ChatGPT applications, that what is a prompt and how do you try to write different prompts and the outputs which you are trying to get, okay? Fine. But in the overall, that you are trying to see that you give an input, okay? And input in the terms of a prompt, okay? You are trying to give an input in terms of prompt, and you feed it to a generative model, which could serve many tasks or many activities which could be performed. Okay? So this was the general idea, and we will be discussing more about it, right? Okay. I hope this is clear. So, so basically, like, like this was the overall picture I'm trying to give you, but in the coming sessions, like I will be discussing more about it. Okay. Okay. So it will be clear to it, but but the overall picture I have given you, and I will be giving you more about it that how does your generative AI work and how, what is the theory behind it, right? So, okay, okay. So, by the end of this session, we will identify applications of GenAI in solving your real-world problems, okay? And also, we will try to create some effective prompts. Okay? So prompts is also basically a very effective manner in which we could clearly tell the GenAI what exactly do I want. Okay? So let's say, this prompt is basically, you are trying to tell, like, a two-year-old kid. Okay? So basically, we want to make our prompts crystal clear, and with a clear set of instructions, if you are trying to work with ChatGPT or some other GenAI model. So the instructions should be crystal clear. So in your mind, so you have a lot of things. Okay? But when you are trying to give it to a GenAI model, so basically, you need to write the precise terms that what exactly you want. Okay? So let's say you are trying to write a prompt, and there could be spacing. Okay? Okay. So that also is being considered a dot. Okay? Some special characters could be there. Okay? Right? Or it could be a JSON, could be your XML. Right? So how you are trying to give the information to a particular tool, that is also important, the structure. Okay? Simple text, okay? Or a JSON, or an XML. Okay? So what is preferred? Okay? So that is also a very effective way to write or communicate with the generator. So that we will discuss. Okay? Then we will also try to create some personalized documents, such as your resumes, or conduct some research. Okay? So that also we will try to see, and then we will try to see some applications where your ChatGPT has been built. Let's say, Write For Me, Can Yi, or Designer GPT, or Concess, right? So these are basically various tools which are given by your GPT for handling specific tasks. Let's say, for My Yi, so this is basically for designing, and the Designer GPT is basically for your designing your websites, and Concess, and your Universal Primer is basically for education, research, or something like that, right? So we will try to see the practical implementations of all these utilities and tools, and we will try to write a prompt, and there I will try to tell you that how you can interact with these systems.

Let's try to see the evolution of AI to generative AI. Okay. So, in the 1950s, okay, so there was a Turing Test. So basically, Turing Test, like this was a test which was performed, like, let's say, right in a room, okay? And there was a person inside a room, and there was a person outside it, okay? And this person will try to generate a sound, okay, from, from human, as well as from the machine, okay? And this sound is being generated outside, and this person has to identify whether the sound is from the machine, or it is from the human. Okay? If it will try to recognize that this is the sound from the machine, okay, then, then it, it, it correctly identifies that the sound is from the, from the human, right? So then it will try to see that, okay, the Turing Test has failed. Okay? So basically, the idea is that, like, this sound is being trying to mimic, whether this sound is from a human or from a machine, and this person outside, which is outside the room, has to identify it correctly. If it identifies correctly, then the Turing Test has been failed. But if it doesn't identify correctly, then the Turing Test has been passed. So Turing Test is basically a machine which is trying to mimic the humans, and it will try to fake this human being, right, the true human being. Okay? So that is what the Turing Test was there. Okay? So then, in 1966, there was ELIZA, a chatbot. Okay? So it was a chatbot using your large language, uh, uh, your language processing patterns, right? So these were there were some kind of patterns which were processing the language, and it mimicked human conversations or the, uh, using your pattern matching or your pattern substitution. Okay? So, so basically, it gives the users an illusion of understanding, but it does not have the contextual information. Okay? So that chatbot is basically developed by using your language processing and pattern matching. Okay?

Then, in the late 1980s, there came the era of neural networks. Okay? Okay. So neural networks, basically, the concept came from the neurons, right, in the brain, right? And which is basically, it is trying to solve your complex problems. Let's say, we have an image, and we wanted to recognize that image, that whether this image is of a dog or a cat, right? Okay. So can a human identify this image and try to recognize? So this was the task, and the first challenge breakthrough in the neural network was the AlexNet, right? Where it will try to recognize the numbers, right? So these numbers we are trying to recognize by using a CNN or an or an neural network layer, which correctly identifies that a machine can recognize a pattern or a number if we try to feed an image to a neural network. So the basic idea was that that we have a neural network, okay, which is AlexNet, and we will try to feed an image, and it will try to classify that if the image is displayed a number five on it, and it will try to recognize five. Okay? So this was the task given to, right? And this task was a supervision task, or or basically a supervised task. Basically, we have lot and lot of images which were prior fed to it, and we trained the neural network model, which will try to recognize this particular image. Uh, right, okay? So this was the neural network.

Then came the LSTM. So LSTM, basically, or your RNN or LSTM, right? So based on the challenges or the drawbacks on this CNN, so RNN was built. So basically, it could only try to process fixed-length data. So in the image, we could have a fixed length of data, in which it could be a 5x5 image, right? 3x3 image, or 256x256, right? Okay. So these were the limitations. But, but here in the RNN, we could have a text, we could have a time series, right? And we could have states, okay? So where our information could be passed from one state to another, and it will try to give me a classification or a sequence-to-sequence generation, all these tasks were given to, okay, right? So this is what was basically your LSTM. Okay. Long state, uh, uh, memory, uh, long state LSTM. I forgot the definition. Long, long short-term memory, right? Long short-term memory, right? So networks, these were the LSTMs. Then, in the year 2014, so there came the GRU and the attention mechanism, right? Okay. So there, by the use of this attention mechanism, so which was introduced in the transformer model lately in the year 2017. Okay. So basically, RNN was doing great. Okay. So basically, RNN and had a job in which, let's say, I wanted to convert this text, "I want to go to market," okay? So this text I wanted to convert to a Hindi language. "मैं बाज़ार जाना चाहता हूँ।" Right? Right. So this I wanted to convert into Hindi. So this is basically an example of language-to-language translation. So this is an RNN, and this is an RNN block. So where one of the inputs is being passed to an RNN block, right? We have the states, and that state is being processed by another RNN, which will try to convert it. Fine. So just I'm trying to give a list that what is an RNN. So here, you try to understand that this is some kind of a block, which is called as an encoder block. It will try to import the data, and the output is being passed to a decoder block, and it will try to decode the information, and then it will try to recognize some patterns or some processing it will try to do it, and it will try to convert this text to a Hindi text. Right? So this is basically a neural network. So this is your neural network, and this is your neural network. Right? So based on the concepts that "I" has to be converted into "मैं", "want" to "चाहता हूँ", "to" "को", "go" "जाना", "market" "बाज़ार". Right? So this is basically a machine-to-machine translation, in which every character has to be converted into another language based on the context, right? Okay. So basically, the thing was that there were some challenges in this block, which is your encoder and decoder. So the translations were not effective because the thing was that it doesn't lay the stress on which type of statement or which types of words are really important to that particular, uh, uh, translation. Let's say, "want" is important, or "market" is important, which would have the higher weight. So basically, this problem was solved with the help of your attention mechanism, where every word has been assigned some kind of an to it. Okay? So let's say, "I" would be important, and in the next statement also, that "he comes". So "I" and "he" are interrelated. Okay? So then we know that these two terms are talking about the same person. So then we assign some weights to it. Okay? Weights, we try to assign some higher weightage to it. Here, this is important. "to" could not be important. Okay? So "to" has the less weightage as compared to "I". "Market" could also be important. Okay? It could also have some weights. So as compared to "to" and "market", so "to" could be a lower weightage, and "market" could be a higher weightage. "go" could be important, right? And "want" could be important. So this way, you can just try to recognize that by use of some attention mechanism, we can just try to focus that in a particular sentence, which words are really important, and we could pass to this decoder mechanism. Okay? So let's say, consider in this part, let's say I wanted to give you some, actually, we are trying to read some document, let's say, okay? So while reading some document, if if we are trying to read some sentence, right? So we try to highlight, highlight this sentence, okay? So this part is important, this part could come in the exam, okay? So this term is important, I could read and memorize. So this is how exactly humans think. So similarly, for the machines also, we, if we try to provide some kind of a weighting mechanism that, okay, so let's say these were the very important, you assign some weights to it, you assign some higher weightage to it, so that when I try to pass to another platform or another system, so I could see that according to this or this, so this had a higher weight, I could only focus on this text, not this text. Understood? So that is why the translation could be effective. Instead of seeing the whole document, I could only focus on those which have a higher rate. So this was the list of your attention mechanism. Okay? So I'm not going into the math or the equations of the attention mechanism. So the basic idea is that when you try to feed a, uh, uh, sentence, and the words which correspond to that sentence, the words which have higher weightage or which are of importance, so is being treated, uh, uh, like, treated with more attention, basically, that works. Okay? If you are trying to pass it to the decoder, decoder mechanism, and for that part, okay, right? So if you are trying to understand this partly also, then it is also correct, okay? Because because this thing is being, like, if we try to explain the whole process, it will go like, we will deviate from our original topic, right? So this topic is not recommended, right? So I was just wanted to give you the generative AI and the tools, okay? But the theory part to understand. So this is your attention mechanism, right? Fine. Okay.

So now, uh, uh, this after the attention mechanism came, okay, after the attention mechanism came, then it was a transformer. So transformers came into existence with your attention mechanism. Okay? Right? So there was a paper, "Attention Is All You Need." So that was a major breakthrough in your transformer model, and from there, okay, so we have emerged in the transformer where we have an encoder and a decoder architecture. Okay? So this was the encoder-decoder architecture, right? So basically, you feed this text to an input text to the encoder, and it will try to produce a translation. Okay? Right? So this was the task which was given to an encoder-decoder mechanism. Okay? So the next part, like, which was there, this was encoder-decoder. Encoder-decoder. Another part could be your encoder-only part, in which you could pass the text, and it will generate some classification for you. Okay? Which could be a BERT, right? So this was the paper or the model which was released by Google, which is Bidirectional Encoder Representations from Transformers, right? So then, this is your BERT, which could be your encoder-decoder, and your T5, right? Which could handle multiple tasks, not only your, this thing, it could handle, right, translation, it could also handle multiple tasks, right? So T5 was the kind of architecture which was released by Google, that is possible to handle multiple tasks, and this is your encoder-decoder, which could be only used for classification, and we could have your decoder, in which we could pass the input, and it could be text generation, which is your GPT or your LLM. So guys, LLM is basically your transformers, okay? Which is inside your transformers, we have three categories, which is your encoder-decoder, decoder-only, or encoder-only, and that part is your LLM is decoder. So LLM is only your decoder-only part. Okay? In terms of your transformers, this you need to understand that it is only for the text generation, that it will try to, a probabilistic model in which we are trying to generate some text. Okay? You try to feed some input, and it will try to give you the next token, or the next sentence, or the next word, right? The next word, sentence generation. Though this GPT model is capable of. So that is why we have clearly evolved from Turing Test to neural network, then we have the RNN, then the LSTM, then the GRU, then the attention mechanism was the breakthrough in the transformers, where we have three kinds of transformers, which is your encoder-only, your decoder-only, and encoder-decoder both. Okay? Right? So this is your LLM, which is your decoder-only part. So all the Gen, ChatGPT, your Gemini, here, your Grok, right? And like, there are more, right? Claude, okay? All these are decoder-only with a little variation, okay? And trained on millions and millions of data, okay? Right? So this is your ChatGPT series. So on the basis of your LLM, which is your decoder-only part, so this is being evolved: Chat 1, then 2, then 3, then 4, then 4.0, zero, and so on. And still, it is being continuing as we talk. So there is some ChatGPT series which is being evolving by some other company, and over the time, it will be just released. Okay? So, so, so pardon me if I have not updated to the ChatGPT series or or some other tools, right? But, but the race is never ending, because every company is is going and making their own custom-made GPTs, right? Okay. Okay. So, up till now, any questions, any doubts, any, any anything you don't understand, let me know. Okay? So basically, this was the evolution from AI to generative AI. Okay? Right? Okay.

So now, as I already told you, that AI could be, as, as one of the students pointed out, right? That AI is basically generating new content, such as images, text, music, or other forms of creative output. Okay? So here, now, in terms of this, so this is your LLM. Now I will call it as a decoder model, a transformer model. Okay? And this is the architecture which you are trying to feed some input, and it will try to generate some output. And this LLM model is being trained on lot of lot of data. Right? So there are two things which comes under your trained model, right? Which is your parameters, model size, would be say that, and your training data, right? So ChatGPT 2 was having 1.7 billion parameters, right? Close to 1.7 billion parameters, and ChatGPT was having 175 billion parameters, ChatGPT 3, right? So you can see the difference that how many parameters are there. Parameters are just the weights, okay, of any model. If the machine learning model has some trained weights or trained weights, so these are called as parameters, and you can see that if you have so much weights, the model size will also be large. So it, the model size was 16 GB, the ChatGPT 2, but here it is like, I, I guess, like it is more than 16 GB. Okay? And the dataset which is being trained on these ChatGPT is like around, like, 500 GBs and so on, right? And more, more, more, right? So, so it is being trained on huge amount of data, which will have huge number of parameters. Okay? And that is why these ChatGPT, you cannot train on your system, first of all. Okay? So this I will try to make you clear. ChatGPT 2, they have released the model weights. You can download these model weights on your system, which is of 16 GB. Okay? And you can try to run the inference on your system. But ChatGPT 3 onwards, they do not, they haven't given the model weights. So they have given in the form of an APIs. Okay? So that is what we are trying to do is, we are trying to call the OpenAI or some Gemini or a Google API to write the prompt and get the output. Okay? So we are calling their servers, where these models have been posted on their servers, not ours. We cannot download this. Right? But this we can do it, but it will require at least 16 GB RAM, because when you are trying to work on a model, so or train it, or do some inference on it, so it will, that model gets loaded into your RAM. So at least 16 GB of RAM is particularly required for loading your ChatGPT model into your local system. Okay? Fine. So this was the thing that these generative AI models are so large that you cannot run on your systems or trying to train on your system. Okay? Right? Earlier, like if we are trying to do a deep learning or a machine learning, we could clearly do some training based on the data, but the data was so small, let's say 10,000 samples or 1 GB data, 10 GB data, right? So this we can do it, right? This was not an issue. But here for the generative AI, so it is being trained on huge and huge volumes of data, that accumulation of huge volume of data, you cannot, firstly, cannot do it. Secondly, it requires hardware also to run and perform the training. Right? After the training, so you try to make a model. So that model, they have already provided to us in the form of APIs, which we are trying to use as. So this is the whole process where your APIs, your ChatGPT is being trying to work with, and then these models are capable of trying to solve multiple tasks, whether it is an text generation, right? It is your email classification, sentiment analysis, right? Voice recognition, right? Or or your messaging or a chatbot, right? So all these tasks would be effectively done by your generative AI.

So what are the objectives? So the objective of generative AI is to independently generate new content. Okay? So content generation is some probabilistic model which will try to generate new content by deriving a structure from the recognized pattern. So recognized patterns, that we have a large amount of data, from that large amount of data, it will try to recognize the patterns and identify that how I can generate that content. And the another thing is that it assists us in some complex problem-solving, right? So let's say, like, you try to have a large amount of data, right? Which is being already this model is being trained on, and you try to write some problem, okay? This problem, it tries to identify and see that this similar problem hasn't been solved or by some other, right? Based on that data, it will try to recognize and it will try to give you some variety of outcomes, okay? So that is also we are trying to see nowadays. We wanted to write an email, you will quickly tell the context that I wanted to write an email that I wanted to apply for leave, I wanted to write to a professor about the project submission, and so and so, right? It will try to effectively generate that application or that email, which is being nicely written and according to the context we provide to it, it will try to generate, right? And seeing these, these patterns, and it will try to give us the output. Okay? And this, in fact, has boosted our creativity. Okay? Because it will try to produce unique outcomes. Okay? And also time saver, also, and new ideas and possibilities. So nowadays, there are some tools effectively in which there could be a no-code, right? You don't need to learn coding, effectively, you can ask, tell the tool to effectively create the website, to write the code for me, and it will try to write the code effectively for you. All we have been doing this, right? But in sometimes, there the human intervention is also needed, because these generative AIs, right, hallucinate a lot. So what is hallucination? Hallucination means that, yes, yes, Puja. Yes. It is basically, yeah. So the human intelligence or the human creativity, we would be dependent on these tasks or in these, these, the chatbots or these AI systems, right? So all those sci-fi movies which we try to see that in the 80s, they try to send us a message that these humanoids or these chatbots or these AI technology will will try to outcast us and it will try to rule us. So that, that will we try to see in the incoming 10 years or or or this. But the thing is that, like, we can just try to control this technology, okay? And try to be productive in that sense. So first, try to see that how we can leverage our brain to the most effective task, and then we can just have these these chatbots or these effective systems. Okay? But in terms of creativity, yes, it, it, it has boosted our creativity. Okay? And now, there it is being need for every individual to learn these tools. Okay? Because every organization, every company, right, is is asking to implement the AI solutions or the AI tools and and start using these tools, because if you don't use these tools, so you will be left behind. Okay? And using these tools, so use these tools in such a manner that it will try to increase your productivity, and it could be a time saver for you. Okay? In that time, you can perform other tasks and just be more productive in that. So all I can say that, like, if you are trying to use these AI tools, so use with a caution that instead of getting your brains damaged or being fully dependent upon these tools, so use your some ideas or some suggestions that how you can outsmart these technology, right, right? Suja, is that answer your question?

Now comes the large language models. So we will start talking about these large language models. So these are pre-trained models. Okay? So what I was trying to tell you, so these have been already been pre-trained. Okay? So you don't need to train on your system. So training it is impossible to train. Why? Data, hardware, right? These are the two things in which you have a constraint. So there is a huge amount of data, so which you cannot collect, and the hardware specifications, which is your, your, your, with multiple GPUs or high-processing GPUs, which is very costly, and like, individually, you cannot make this happen. And it is being big organizations which are using these these technologies to leverage and release these pre-trained models to the customers. Okay? Right? Okay. So these pre-trained models are capable of understanding and generating human-like text. Okay? Across various tasks and domains, and as we are trying to see that. Okay? So, so in every domain, these technologies are being ranging. If you talk about in your images, your music, right? And and and your other things, right? Advanced models. So, so these, these terms are being widely been used. Okay?

So now let's, let's break down this term into three parts: Large, Language, and Model. So, Large, why is this term referred? Because of the significant size, like, as I told you, 176 billion parameters, right? If once it's 76 billion parameters, and you need to require, like, around, like, if I try to see that 300 GBs of your RAM to run your inference on a single system, so close to then trying to say that how many parameters, just multiply by two, and you get a GPU. So let's say, if your ChatGPT was having 1.5, right, 1, 1.4, right? Let's say 1.4, uh, uh, billion parameters, right? Close to like 16 GB or close to 20 GB was required, uh, right, to do the inference. Okay? So nowadays, these billion, I'm talking about Chat 3. Chat like this is much larger, according to the size and the complexity also. By Language, so primary function is to understand and generate human-like language. Right? So this is basically a large language, a probabilistic model. Okay? And, uh, a generative model, which is trying to generate, and it is a transformer-based architecture, which is having a decoder-based architecture. Right? So this was the base with your attention mechanism. Okay? Right? So this is your large language and model. It is your mathematical representation which captures the patterns and the structure of a language data. So the model contains some of the architecture. What is the architecture behind? So that we are talking about your decoder architecture, and it could vary, right? It could vary from GPT to another GPT, or could vary from companies to companies. So the basic decoder model is is basic. They have used the basic decoder model, and then on the top of that, they have modified that basic decoder model, and the newer architectures are are releasing and coming. Okay? Right? So the overall architecture is your decoder vanilla decoder model, and on the basis of that, we are trying to change some of the architecture of that to just try and release some, uh, what we call, state-of-the-art models in in future or or in in in these days. Okay? Yes.

So now we'll talk about widely used LLMs. So ChatGPT 3.5, 3.4, 4, okay, 3.5, 4, 40, 4, right? Gemini, right? These were released by OpenAI. Then we talked about Gemini, which is being released by Google. Llama, right? Meta. Then Falcon, okay, Technology Innovation Institute. Pi, then we have Claude, Claude v2, Claude v2 more, v1, v2, okay? So these are philanthropic. Okay? So these are the various companies which we have these models, and they have released their keys. Keys means your API keys. Okay? So you cannot access these models. Uh, okay. So they have their their chat interface. Okay? So which we'll try to see ChatGPT, and in terms of your coding, if you wanted to use these models, so they have the APIs also. So these APIs are free at some level, but they are chargeable. Okay? So they charge, right? So when, when you try to use their APIs in your code or wanted to build some applications, so using your Python or your front-end or any application, so all these companies are given their API keys. So you can just create, create your, create an account on these particular, uh, uh, portals, and then request for their API keys. So you have an API key, which is free of cost for some time, for some initial tokens, and once, once these tokens have expired, and then they start giving you notifications that you need to pay to continue the service. Okay? Okay. Some of the tools or some of the models are open source also. Okay? It is not that, like, these things are are are chargeable. Okay? So these are chargeable. Okay? But initial level, we can have a limit that that we can use up to, up till, like, 20 tokens or 30 tokens, or based on their portals, they keep on updating. But there are some tools such as Olama. Okay? So this is your free. Okay? So you can use this, you can use this tool. Okay? So like, I will be talking about Olama also, right? Maybe tomorrow or or day after tomorrow, or or maybe in the next session. Okay? Right? So so that you can use it, and then you can just explore with it, or you wanted to understand it. Right? But there, these, these free tools come with some challenges. So either they are slow enough, because it is being your downloaded on your system. Okay? And they come with some challenges. So, so just trying to just educate you about the free tools as well as your paid tools. Okay?

So now the features: it will try to generate content, text summarization, sentiment analysis, data extract, and language translations. Okay? So these are some of the features. If you wanted to do the text summarization, so let's say I have a text, and I wanted to generate the summary out of it. Okay? So that it will try to give you the summary. Let's say you are having a chapter, a book, okay, in which you don't want to read the whole chapter, you just try to give feed all the data and just ask a prompt that I wanted to give a summary in 2,000 words or 200 words, and it will try to give you the summary. Okay? But it will occur with the token cost, which is, which is this key, key API, right? Which I'm trying to tell you that when you are trying to feed the input data, that is being divided into tokens, and the output data which has been generated, that is also having some tokens. So this token plus this token, it will try to add it, add up, and let's say this is your 200 or 2,000 tokens, and this is your 200 tokens. So total is your 2,200 tokens. So now you see that 2,200 tokens have been, 2,200 tokens have been expired. Okay? So just try to make sure that when you are trying to write the input and generate the output, so you don't, uh, uh, so they are trying to make the mechanism of costing with the help of this token. Okay? Okay. So this is the gist of GPT, like they try to give you the APIs, which is free, and they say that, okay, up till this many tokens, you are free, but if you exceed your tokens, then I will charge you. Okay? So that is what it says, and the cost, how they calculate, by the input text, every word is being converted into token. Okay? So maybe that word would be divided into some tokens, and the number of tokens would be dependent upon that word. Okay? So there are various strategies or various methods they use it to convert your word into tokens, and after that, they will try to calculate the cost. Okay? Right? Fine. So this is how your ChatGPT works.

Now, coming back to Gemini. So it aids in complex problem-solving and code generation. Okay? So if I try to see or tell you that, okay, which of these models is best? Okay? Either OpenAI is best, or Google's is best, or Llama is best, or Falcon, or Anthropic? Until unless you don't use it, you will never know that what is best suited for you. Okay? So maybe in one problem statement, like in my case, like, let's say I'm trying to work on a solution that in my part, that I work with Gemini or I work with Claude, right? Okay. So it has given me excellent results, and ChatGPT hasn't given me the excellent results. Okay? Like in another problem situation, ChatGPT works fine for me, and these will not give some solution to, right? So it depends, okay? So you can try any of these these models. Okay? Every company is trying to tell you that I'm trying to work on a complex problem and code generation, and I do this. I try to work on language translation or content creation. Okay? So basically, it is up to you. Okay? So which part you wanted to use? But OpenAI is is aggressively changing its models, is changing its technology, is changing its architecture to give you better results. Okay? And Google is also a big organization which is also releasing it. Okay? And day-to-day, with some of the researchers, some of the scholars, so they are trying to work on complex problem-solving solutions. Okay? So Ashish, I will try to give you some ChatGPT live demonstration tools also. So there we will

Try to discuss this. So for the every like in the beginning also I said that for the designer I have a separate chat executed tool for making the post. There's another thing, right? So that we will try to see in a few, uh, or maybe in the second half, we will try to see it.

So Falcon is also another, so which, uh, tells you that common reasoning and sentimental analysis, right? And Anthropic. So Anthropic, I've been using Anthropic, Llama also, your ChatGPT also, right? Into day-to-day activities and whichever like is giving me, uh, the results. And Grok is also very good, right? So Grok gives you a detailed explanation, as I already seen that. And in it also supports language, multiple language. If I talk in Hindi and try to say that like, like my parents are old enough, and they try to tell the Grok in Hindi, and it supports Hindi language also, and it will try to give me the answers in Hindi also. Okay, right?

So, so basically, it is up to you guys, okay? So these tools are already available in the market and, uh, you can use and experiment with this. And until and unless you don't experiment and don't use it, so then you don't, uh, make a choice, okay? Okay. But the idea is that like these all tools are charging. Okay. At one point of a time, maybe like they have a daily limit. Okay. Or they have a token limit. Okay. In a day, you can spend these money tokens to use it. And over the time, they would decrease their charge because these all companies are coming and like they would be a part and parcel of our life in the coming, uh, months or years, and they would be free of cost because their cost would be minimum. So, uh, like as for the, uh, tech enthusiasts and, and other tech-heavy people, they are trying to predict it. Okay.

So now, let us try to see that how this top GenAI tools are transforming various industries. Okay. So now, so this is your GenAI in the terms of code. We have the Copilot. Okay. So this is an excellent tool. Okay. So majorly companies are using this Code Copilot, okay, to be integrated with your IDEs. Okay. So when you are trying to write the code and you have a problem statement, when you are building some, uh, application or writing some logic. So if you want to clearly describe that logic, so Copilot does the work for you, and it will try to generate the code. And based on that, like you can just run that code, and this has solved a lot many, uh, uh, what we call time, right? And also there are new things or new, uh, patterns which come, right, which you don't know basically, right? So this is also a very effective tool in terms of your writing your effective code and helping you. It basically acts as a copilot for which you drive to it, and then you write your logic, and it will try to correct you, and it understands you, and, uh, gives you the output. Okay.

So then, Postquare.ai, Mutable AI, and these are also, uh, uh, some of the other tools which are being used for other companies, and you can just, uh, refer to these tools and then use it. Okay. Then conversational and text, we have the Cloud AI, right? Chatsonic, okay? Then PaLM 2, ChatGPT, Gemini, Grok, Notion AI, right? So, so this is basically we are trying to make a conversation. Like every organization is following this Copilot, okay, right? In terms of whether you are trying to integrate or work in your VS Code or your, uh, PyCharm, right? So it gets integrated. Right? Even if in your teams also, as just stated, so it it it is getting integrated with other applications. Okay. I will also, uh, teach you one tool where, uh, this is a ChatGPT tool where you can integrate with your browser. Okay. I will, uh, we will have a demo, uh, in the second half. Right. Okay.

So then we have the design tools also. So DIY, right? Is a very effective tool which is being used by OpenAI, Canva. Right? So we will see Design AI, Logo AI, right? And Midjourney, right? So, so these tools you can just, uh, work with your designs, right? In which you can create your, uh, creative designs or creative pictures by feeding them. Okay. So this, this is wonderful. Stable Diffusion is also a very good tool, right? Okay. For the sound, right? If you wanted to, uh, do something, uh, in sound. Okay. So then we have the AudioCraft and, uh, I, then 11 Labs, right? So, so these you can just explore it. For the videos also, right? So Smooth Area, Victory, right? Absence or AI, right? So these are also some tools in which if you are working with your video, um, so you can just explore it. Okay.

So, so some tools are free, some are chargeable, some give you a monthly subscription. Okay. But in the end, so they are all commercialized, guys. Okay. So there are no freelances. Okay. Everyone is asking to pay and use these tools. Okay. According to your convenience, and you can just select whatever tool you like and is best suited for your productivity, you can just go with it. Okay. Some of the tools I have personally used. Okay. So Copilot, I'm trying to use it for a quite a long time, and in the terms of coding and this thing, this this is proven effective. Okay. Right. So other tools I, I like use it like like Grok or ChatGPT or Jasper. Okay. So on the basis of your free access that I use it, and if, by changing the emails, so you can use it. Okay. And then one month you get a subscription, and then you can try it out. And if you think that your work is not, uh, uh, you are not, you cannot survive with this tool, so then you can just pay and use that tools. Okay. Okay.

So let us try to see, uh, uh, resume cover and using your ChatGPT. Okay. So, so let's see that if you have a resume and, uh, but we are not unsure that it matches the job description. Okay. So we can just align or quickly effectively update the resume with the job with the job description and save time. Okay. Because so this is a, uh, common, uh, basic, uh, requirement we have if we are trying to apply to different companies and, and different companies have different job requirements. And what it does, let's say, uh, the company has listed a job profile, and according to your resume, that is not meeting the criteria, and you are not trying to get, get the shortlisting done. Okay. Okay. So in that case, uh, let's use the ChatGPT and where we can just, uh, uh, see that how you can use it. Yes. Yes. Torups, that is also good. They are also using giving the subscription. So basically, it is for the, uh, free version. So if you wanted to go, you can click on this upgrade and then you can have it, right? So if you wanted to have the upgradation. Okay. So this is for $20 per month. Okay. So here, what we are trying to get? We are trying to get the deep research, multiple reasoning models, 04 mini, mini high, 03, and, uh, deep research low of ChatGPT 4.5. Okay. Right. And limited access to store video generation because each has, what we call, sub tools or sub, uh, uh, sub, uh, applications, right? Which is being used for specific areas. Let's say for for for Canva for for writing text, okay? So the these are the things which we can use, okay? So like if you, if you feel like this is like, uh, needed for my, uh, day-to-day activity, so you can just go and just, uh, get the, uh, this workbook. Okay. Okay.

So now I've shared your full content. Okay. So now let me try to explain you what I have shared it. Okay. So these are the details. The candidate profile is name Alex Johnson. Experience is five years. Industry is Information Technology, and these are the skill sets. Okay. So basically, what I did is like this is your resume. Okay. Basically, this is your position, Fullstack Company, Innovative Tech Solutions, and San Francisco, California. And this is the job responsibility, right? Okay. So now, up till here, it is your basic prompt, uh, basic, uh, this thing. We, we wanted to create a resume for Alex. Okay. Showing showcase his five-year journey in the IT IT industry as a Fullstack Developer. Okay. So it should highlight his expertise in Python, Java, JavaScript, and key frameworks like Django, React. Okay. As well as skills in database and cloud technologies like PostgreSQL, MongoDB, and AWS. The resume must emphasize his Alex's leadership in product, project development, team collaboration, passion for technology, and continuous learning. Okay. Right. So there, it is basically, it could be Alex, who could be you also. Okay. So here, in the first part, we have been trying to give the, uh, the resume of a particular person. And in the next part, we are trying to tell a large language model. Okay. So you wanted to create the resume for me that highlighting these these expertise areas. And if some company has some job requirements, then we can also add it here that this company is asking for the job requirements. Can you modify my above, uh, resume to match that specifications? That also you can do it. Okay. So basically, it is writing an effective prompt. Okay. So just try to do it and hit enter, and let's see what it will try to generate it. Okay.

So it has given you that Alex Francisco, this is your email ID, and this is your LinkedIn or this is your GitHub. And this is your professional summary. Okay. So earlier, like it was not basically a summarized version, but here now it is being a looks like a professional, uh, resume. The core competencies, like if I try to see as an HR or a or a recruiter, so I could see that this is a professional summary where the like Python, Java, JavaScript, and your various technologies are being listed down. And I could also parse these technologies because like like these these technologies what we are trying to require in our current company, right? Okay. So this is the core competencies. We know React, you know NodeJS, Python, or Fullstack, right? So these we are trying to have a requirement. Okay. That's good. Docker, Jenkins. Then we have professional experience here. What projects we are trying to do it, and what we are trying to do it, right? So this you can just see. Okay. So this is basically taken from your resume itself only. Okay. So education and your certifications and your project highlights and your soft skills and technologies, right? Okay. And then also it is trying to see that, uh, that if you try to have a Word or a PDF version, or you can create a customized cover letter for this role. Okay. So you can just see that what what matches your requirement. Okay.

So basically, what you are trying to do is you are trying to, you have fed some data. Based on that data, you are trying to ask something, right? So basically, if I talk about this, this is basically the instructions, right? Given to your LLM. Okay. And the above part, what we have used here, this above part, so this is your resume. This is your context. Using this context, using this your general information or your knowledge, you are trying to give me some output which is more refined, and it is more attractive, and it is much more better than the original one. Okay. So this is the task which we intend to, uh, we intend to do, and we are trying to do that, right? So, so basically, like based on the prompt, you are trying to tell that it must emphasize the leadership in product development, team collaboration, passion, and technology, and this these terms. Right. Okay. Right. So, so this is what, uh, like I want you guys. If you have your your resume also, you can just effectively change this and try to work on it. And if you, you wanted to ask some other information that you want the formatted part in a PDF or on a Word part, okay, so that also you can do. But make sure that like your account doesn't get exhausted, uh, because because this is basically your tokens. Okay. And the output is also your tokens which are getting generated. And, uh, right, it may have a daily limit of using this. Okay. Okay. For the general day-to-day task, or it is effective for us. Everybody, uh, fine with this? Yes. No. Okay.

So guys, uh, so let me tell you. So there is no such, uh, hard and fast tool, okay, that which, uh, LLM you can use. Okay. So it could be a ChatGPT, could be a Gemini. Okay. So it could be basically like the accuracy and, and the output which is being generated from this various LLMs, uh, could have evaluation criteria. How do you evaluate these systems? So that is a big challenge. Okay. Whether the data, uh, which you have given, or the question, or the prompt you have asked, on the basis of that, is it giving you the correct option? Okay. So sometimes it happens like when you are trying to, uh, write some logic, okay, and you try to tell that, okay, this is the problem statement, I want this code, and, and it will give you the exact code, and maybe that code is not running it properly, it is giving you the errors. Okay. So maybe one one task it has performed better, and another task it has failed miserably. And on the other sense, these these, uh, models, you can just, uh, try with other models and check, check that accuracy. But there is no, uh, hard and fast rule of checking the valuations. Okay. However, there are some metrics which is being, uh, denoted. Okay. So perplexity is a metric. Okay. So this is not a tool. Okay. So I'm not talking about the tool. I'm talking about the general term. So perplexity is a, is a valuation criteria also, right? In which I wanted to, uh, uh, evaluate my language model that on what basis that language model has a score which is better than ChatGPT, or if a same question could be asked to ChatGPT, or a Gemini, or a Falcon, or a Cloud AI, which one is performing better. Okay. So that evaluation tool is, is basically this thing. And another evaluation is your human evaluation. Okay. So you know best that according to your problem statement, this is, this is working fine, and, uh, this, this works better. Okay. So that human evaluators are best in this prompt. Okay. So I hope everybody is back. Okay.

So now, let's quickly dive into the prompt. So can anyone tell me what is prompt? Yes. Input we provide. Okay. Okay. So let us try to understand. Yes. Command, the message, the right to these two. Okay. Fine. Great. Great. Okay. So let us try to understand these prompts. So prompt, what is the prompt? It is a natural language text that instructs the generative AI models to perform a specific task. We have seen an example where we have provided some context. Okay. That resume, and in the later part, we have written some set of instructions that on the basis of this resume, I want to modify my resume in such and such a manner. Okay. Right. So a well-crafted prompt guides the general GenAI to understand the context and the specific requirements of the task in hand. Okay. So well-crafted. So the prompt what you write should be a well-crafted prompt. What does that mean? It should have proper spacing, proper instructions to be given. Okay. Okay. Right. And it should be clear and precise. It should not be some kind of an, uh, thing that you try to just tell the LLM model. Okay. Uh, he will try or she will try to understand it on your own. Okay. So just don't think it in such a manner that whether you have written, uh, some kind of an instruction which is vague, which is incomplete, and it will automatically try to understand it. Maybe it can recognize, but the chances are very less because it, because they have well-crafted prompt, a proper instructions prompt will give you the best answer also. The second thing, the cost, the tokens, as I already told you, as as one of the learners has already expired his tokens, the daily limit, okay, maybe that you must be a cost-directed also, in which we are trying to use these tokens in such a manner that that it should clearly describe the task which you wanted to, uh, tell the LLM to perform. Okay. Right. So precise prompt can lead to a more focused and useful output, and it has acted as a bridge between the user intent and the AI execution. Okay. So basically, what is my intent, and how the AI is is getting that intent executed, that is basically a prompt is. Okay. So prompt is basically a set of instructions which we are trying to give it to the LLM to perform our desired operation. That prompt should be clear, should be precise, should be well-crafted, should not include, let's say, you are trying to include some spaces, additional spaces. I want to, then you put a space. This is also counted as a token, right? Okay. So just make sure that, uh, it also recognizes some pattern that you have written something like this, and maybe on that pattern, it will try to recognize it. Okay. So make a proper English language understanding that that it will try to give you the exact answer. Okay. Okay. Fine.

So prompt could be, uh, a basic straightforward question, right? And the language model provides an answer based on the training data, right? Okay. So let's say, what is the capital of France? So in this particular thing, you have asked some question, and based on the training data, it will try to see and it will try to understand this context that in this, the capital of France, right? So this is basically a more weightage, this part with what I was talking to France, capital, right? So this is basically the keyword terms or the terms which it tries to recognize, and on basis of that, it will try to give you the answer, right? Fine. So prompts could be more complex and used for various tasks such as text completion, translation, summarization, or even creative writing, as we already seen that, okay? So we have seen a prompt in which we are trying to provide some information to it. On the basis of that information, we wanted to modify that, uh, particular, uh, resume or set of instructions, right? Okay. So, effectiveness of the response depends upon how well the prompt is crafted. Okay. So it is not that like, uh, uh, maybe if I try to write the prompt, or on on your systems, it has given you some other outputs, on my system, it has given me the other output. Okay. So that based on the prompt, it may vary. Okay. The basics, basis is that that, uh, these are probabilistic models. Maybe if you have written the prompt, it may not generate the exact thing. Maybe in terms of iteration, if you try to, uh, write the same set of instructions and try to give it the answer. Maybe in some some part, it will try to hallucinate. Okay. Right. Okay.

Now, the components of your prompt. When you are trying to write a prompt, what is the exact thing you need to do it? So let's say, let's take an hypothetical example. Okay. So you wanted to perform some task. Okay. You are trying to tell that that act as a user, which would be your medical practitioner. Right? So you can say to an LLM that you are a medical practitioner. Okay. Or a medical, uh, student, or a doctor, or a judge, or, or a lawyer. Right? Okay. So this you can have that particular user in user set. So user set means that as an, you are trying to tell that LLM, you act as a judge, right? You act as a medical professional, or a medical doctor, or something like that, based on your profession, or based on your understanding. Okay. So these are some rules you need to follow, and I'm providing you some information about some patient, and, and these are the patient text, images, or data set, and you give me the output. Okay. Right. So this could be one of the example in which you have a user, then you have your instructions, then you have a context, and then you have an output. Okay. So this is one of the prompt which you can effectively make according to your use case. What prompt we have made just now is according to our use case. What we have done is in the context and in the input, we have provided the text, that is the resume. Okay. Or you can say that, right? So input or the text, we have provided as a resume, which which act as a context, and we have given as the instructions for a task. This we have combined it. We can break down into simpler steps also. Instructions could be that give me, uh, uh, two-liner pages in the Word format. Okay. In the JSON format, I want that information to be given. Okay. So it should not exceed to, uh, uh, 200 characters, something like that. So these are the instructions which can I give it as a bullet point. Okay. Remember, if you are trying to give me the resume, so just give me two-liner page, give me in a Word format, give me in a JSON document. Okay. Right. Okay. So these are some kind of instructions which we can give additionally in which we can just ask or tell the LLM that perform, you need to, uh, work on this instructions, and the task is that you need to generate the result. Okay. So we have simplified that version. So in in one task only, we have given the instructions, also we have given the task also, but we can divide it individually when the situation is complex, or the problem statement is large enough. Okay. Right. Okay.

So let's say you act as a judge. Okay. And you wanted to, uh, uh, one, one, this is your case file. Okay. So, so based on the, and, and you pass some context of the court filings or some PDF documents, right? And you give it as a input, and you try to tell some instructions that whatever you know about that case, that these are the instructions which we already know that, and you wanted to perform on B this part. Okay. Okay. Or you can just say that, give me that information from this particular case files. Okay. Using this context, just extract me the party one, party two, you can extract me the judge's name, you can extract me some information. Okay. So something like that, you can just ask it. Okay. Another part in the medical, I already told you. In the legal, I already told you. There could be another industry also, okay, e-commerce industry. So based on the input data, data could be in the JSON, it could be in the Word format, it could be in the external word also, external site also. Okay. From the external word, which is giving you the context, and you could have the images or data set or the task, and on the basis of that, you wanted to retrieve all the information, and then you can just ask them that go to that site, perform that operation. From this images, I want this particular images, uh, ratings or your amount to be compared, and then, so based on what particular use case, you can just ask this. Okay. So, basic prompt has these many, uh, instructions, or these many, uh, components, you can say that. Okay. Understood? Yes. No. Okay.

Now, another thing which I wanted to tell you that there are frameworks. Okay. So for the prompts, there are prompt frameworks. Okay. So ReAct is one framework, Chain of Thought is one framework, right? And so on. So basically, what you do is according to these frameworks. So these are some of the frameworks which are widely used. Chain of Thought. Chain of Thought is saying that you, you try to tell some, uh, instructions or give some instructions to the LLM, and then you can tell them, okay, so this reason, this instruction, reason this particular statement, or think on this statement, and then give me step-by-step analysis of this. Right? So these are some of the terms you can just ask it in this, like after reasoning, just apply the step analysis, and then give it to give me the answer. Okay. So these frameworks could be used in that framework. So you have some, uh, task defined, you have some conclusion, you wanted to have an output. Okay. So these are basically, uh, one part I have clearly defined you that you have a user, then you provide a context, then you provide input, then instructions, and task. Another could be some operation you wanted to ask, that what is the task, what is the conclusion, what is the output. So, and, and, and you have a context based on that. Okay. So these are the some important parts or components of using a prompt. Okay. And make sure that prompt is your, uh, precise and your clear, and it should have proper spacing, that also matters. Okay. And if given with the JSON, or, or text, or XML, that would also be a, uh, a nicely, uh, in nicely given instructions because it has been seen in some of the papers that JSON, text, and your XML is a well-structured language. Okay. If you try to feed their data in this manner, and the chance of error or understanding this is better instead of your text. Let's say, name, you can see, give it, let's say, John. And then instructions, or text, or images, or resume, uh, headings, right? Instead of bullet points, you can just omit that bullet points, you can give it as a JSON structure. Okay. So that will also reduce your cost, and also it is an effective measure in which how you are trying to give a prompt to an LLM, because this is a query language, okay? And these input, or these large language models work on this query language itself only. Okay. So if you are trying to work as a, or think as a developer point of view, then it would be fine, and it will give you the best suited instruction and output. Okay. Fine. Okay.

So now, let's say when you are trying to give a, uh, prompt, okay, so what are the important things? Input, context, instructions, and task. So let us try to, uh, like divide this particular thing. Create a brief narrative. So this is the task, right? Which we wanted to perform. Okay. About a wizard quest. Okay. So this wizard quest is a instructions which we are trying to give it. In a magical forest. This is the context. Okay. So where this physical, where this, uh, like wizard quest is there, in a magical forest, right? Okay. And we wanted to, like, this is also instruction, you can say that, surprise ending, right? So this is also instruction, right? So, and create, uh, descriptions. Okay. Provide and load for create creature descriptions. So this is the basically the input which we have provided. Okay. So these all things you can just give it as a JSON format also, that we have a task, we have a instruction, we have a context, and we have the input, something like that. Okay. And here you can also specify a role. Act as a judge, act as a, uh, medical practitioner, act as a sales agent, right? Act as a CEO of the company, act as a database engineer, something like that, right? So this task uses instruction is that. So in that way, it will behave or perform better. So this is basically how you are trying to break down your task into simpler and easy steps. This is how you are trying to tell the two-year-old kid. So basically, we are trying to spoon-feed it and, and giving it in the exact manner with the short and the prescribed information. So that is the whole point in in in your prompt, uh, designing or your prompt analysis. Got got my point? Yes. No. Understood? Any questions? Any doubts? Okay.

So quick check. Which component should be included in a prompt for a language model to translate a technical document from English to French? A. Specify the source and the target language in the instructions. B. Use a data set unrelated to the technical field of the document. C. We could have included any background information about the document content. D. Include technical jargon without explanation. What do you think guys? A, B, C, or D? Okay, is saying A. What about others? We know A. Yes. So now A is the answer. Okay. So we need to specify the source of the target language. So that is the prerequisite. Okay. So I think all are on the same page. Okay. So let's try to do the hands-on. Okay. So day-to-day activities, like we can have a language translation or a LinkedIn profile and sentiment analysis. Okay. So this we will try to see in practice. So what we will do is we will try to have some text. Okay. And there we will try to do some language translation, and we will try to see that this text could be converted into some other languages also. Okay. Then we will try to create some LinkedIn profile, and then, uh, if my tokens, uh, is still remaining, and then we will try to see the sentiment analysis also. Okay. Fine. Okay.

So let's say we are trying to operate an online bookstore, and, uh, we offer books in multiple languages. Okay. And, uh, so we wanted to, uh, use the ChatGPT for real-time book translation. So let's say we have, uh, a book, okay, and some guy happened to, uh, see this book, and from that particular text, they wanted to, uh, create a, uh, language, uh, translation of this book to another language, let's say Spanish, German, or English, or let's say Hindi, some other, right? So we will try to see that like if this text be passed into the GPT with a set of instructions, and it will try to convert into in some other languages or not. So because what we learned from the theory part is that, so ChatGPT could be used as a language translation. Okay. So we have seen generating text, we have already seen, and we have generated the resume. Right. Now we are trying to see the language translation, whether it will try to do the language translation task for us or not. Okay. Work with the prompt. Okay. So let I try to Okay. So this is the scenario. Imagine an online book offers a variety of titles in multiple languages to cater the global audience. Okay. So we want to translate to a broader audience. Okay. So this scenario is no longer needed because we only want the prompt. Okay. So instead of using the prompt, we could also remove this and we can use this: "Translate the following book excerpt into multiple languages to make it accessible to a broader audience." Okay. So this is the text which I have given it. Okay. So let us try to run it. I will also give this in chat. So in Hindi, uh, in Spanish, in French, in Chinese, in Arabic, German, Russian. Let me stop it.

So now the next part is that we will try to create an impressive LinkedIn profile using your ChatGPT. Okay. So imagine you are trying to create your LinkedIn profile, but you are trying to, uh, aim it for something more exciting than a personal brand. Okay. So what I will do is I will give you try to give you a sample chat, uh, prompt in which you can modify. Okay. So based on every, uh, person has a different LinkedIn profile. So based on your LinkedIn profile, you can just, uh, give some instructions and try some, uh, uh, creative ways in which you can make your LinkedIn profile more, uh, uh, stand out and and out of the box. Okay. So let me share you, uh, that, uh, so it has given me the conversation started message also that, "Hi, this is, uh, something like I just wanted to, uh, do it. We can exchange insights or even collaborate on a project. Looking forward to connecting." So if you wanted to do some networking over the email or over the LinkedIn messages, so this is how you can do it. Okay. So, so this is how you can make a LinkedIn post also with the SEO hashtag also. Okay. So on the basis of your resume, on your creativity, it has given it. Okay. So this resume is not mine. This is someone else's. Okay. Uh, I just, uh, downloaded from the internet and then I'm trying to do some, uh, like, uh, asking your internet for a chat to do it, right? Okay. Similarly, you can also do it on your resume. You create a LinkedIn post, and according to that post, you can generate the ideas, and, uh, you can be more, uh, uh, like they could be, uh, some, uh, uh, text or, or some prompts which could be modified according to your use case. Okay. So it says that, uh, in the, uh, experience session, provide me detailed entries that showcase my contributions and achievements using action verbs and quantifying results. Also suggest content for starting conversations with new connections, focusing on recent industry news, within 100 words. So there you can see that we are trying to limit the words because it will try to generate and generate and generate, and, and it will try to generate the max token. Okay. How many tokens it has been intended to, but you can limit it and, uh, up to how many words you are trying to restrict it. Okay. So this is what I wanted to, uh, like showcase to you. You can upload your document, and on the basis of that document, you can make some nice LinkedIn profiles or LinkedIn headlines, and, and emails, or, or, or write some, uh, beautiful text in which you can just, uh, effectively use this, uh, thing. Okay. Okay. Great. So let's move forward. Okay. So this activity you can also, uh, perform later. Okay. So let me show you another activity. So this is how we have done, uh, we have created our impressive LinkedIn profile using our ChatGPT. Okay. So we have written effective prompt, so which will help you to make our LinkedIn profile stand out of the crowd based on some set of instructions. Okay. And within seconds, uh, it has clearly given me the instructions, then how you can just upload it and, uh, go to the LinkedIn profile, and, uh, you can get leverage of it. Okay.

So then the next task is your sentiment analysis. Okay. So in the beginning, I have taught you that it is for threat generation. There is part task we have already seen it. Then now the second task was your, uh, uh, LinkedIn profile creation. Right. Another is your sentiment analysis. Right? Okay. So this we are trying to, uh, give it, uh, basically, uh, set of instructions that how my sentiments are are being rated, either positive rated, or negative rated, or it could be a neutral, and based on the context of the input text, we need to, uh, rate those sentiments. Okay. So let us try to, uh, do this, this also. Okay. I'll provide you this, uh, Okay. So this is the prompt. Okay. And, uh, reviews, uh, let me provide you the reviews. So I will not, okay. So this is your only one review which I have given you. Okay. Because the thing is that there could be more reviews which we can give it, but we only wanted to restrict our reviews to only one part. So, "Categorize the provided review based on the following input: Positive, negative, neutral, mixed, or question or appreciative. Then these are the reviews, right? So provide the sentiment for the following reviews." So this is one. So likewise, you can have two, three, four. Okay. So based on these reviews, so we wanted to rate this, do positive, negative, neutral, mixed, and based on this, we have given some instructions also. Okay. So this act of instruction, you can just include some keywords also that how that some good, happy, or positive comments, right? Negative could be negation, could be no, could be some, uh, some, what we call, like disagreement, right? So something like that, so you can just act some keywords also, or some give some instructions also to rate this, right? And on the basis of that, you can just give your sentiment. Okay. Read this particular text based on the sentiment. So it says that, "The reviewer expressed both positive and negative opinions." Okay. So negatives and, uh, it has given the, uh, mixed sentiment. Right. Okay. Right. So you can just, uh, fiddle with some other options also. I will just share the other text also with you. Multiple data processing kind of. Okay. Okay.

So, let me explain you what is multimodal. Okay. So, okay. So it comes from the term modalities. Okay. So there are different types of, uh, no, not modules. Okay. So modalities, modalities, like we have a touch, okay, we have hearing, we have your, uh, seeing, that is your vision, right? Okay. Okay. So this is basically with the, there are five different senses which we can use it, right? So which is your taste, right? So these are basically modalities which we are trying to use it. And in terms of your LLMs or in this thing, So, uh, we have sound, okay, in which we have vision, which is your, uh, images. Okay. Right. So taste and touch, like, uh, this we don't have. Right. So right now, what, what modalities we have? We have the text, we have the sound, we have the images, right? So these, these touch and touch and taste, we are not using because this is, uh, on another level. Okay. Hearing and vision, we are using. So these are the basic modalities that we have in human beings. Right. Okay. So this comes from the term multimodal, and this multimodal, these are the modalities which your LLM can process. Okay. So text, we already seen it, right? Like we are trying to write a text, and this has clearly processed or, or, or given as a input as a text, and this has given us the output as a text also. Okay. So multimodal means that input could be your text plus image. Okay. And output also, you can ask a text or an image. Okay. It depends. So this is basically an example of your multimodal capabilities that these are the multimodal capabilities where your LLM can handle. Okay. Fine. So you wanted to give as a input as a text along with that image, and you, uh, giving your reference to an image, you will ask that text, uh, based on the prompt, you will try to ask that, that this image, from that image, as an context, give me some output. Okay. Or you can ask from the voice also, or, or you can give your voice also. We give you a sound also. From that sound, you can analyze that sound and give some text that bases on that, this this recordings. Give me some output. What they are talking about? What is the, uh, uh, reference or summary or something like that. Okay. So these are basically our multimodal. Okay. Understood? Yes. No.

So integration of multiple modalities combines text, images, videos, and audios to more to understand and generate response. And, uh, like we can integrate, as I already told you. Okay. So multiple inputs, we can combine. We can combine text, image. We will also see the practical demonstration how we can combine image and text, and then we can get the output. Okay. So it is like a dynamic and interactive response. Okay. So basically, you wanted to give a context in an image, and from that image, you can ask a question, and this is on reading that image. Let's say, your, your PET scans, your MRI scans, or your X-ray scans, beed to an, on the basis of that, you can ask some question. According to this scan, what do you think? Okay. So you can just cross-verify the judgment of a doctor with an, okay, is my doctor being correct or not? Okay. So this could be the thing. So maybe like you try to give your tax, income tax, uh, what do we call, Form 16, and on the basis of that, you are trying to tell them, okay, based on this form, like just try to create the best possible income tax return. Maybe could be a bill, could be your, your invoice. Okay. And you can do a lot of things. Okay. So maybe could be your, your voice could be also be there. Okay. In which the speech could be there, and on the basis of this speech, you wanted to understand, maybe it is not in English language, maybe it is a Spanish or a Japanese language, and you wanted to understand the context that what they are trying to speak. Okay. There also it really helps. Okay. On a video analysis also, so from the basis of that video, you wanted to try to understand that what this video is trying to, uh, uh, uh, say or content is there, you wanted to write that video. From that video information, you wanted to write some textual description, that also it will try to do. So these are the capabilities in which ChatGPT has been emerged as a leader, in which they are trying to give the multimodal capabilities, which not only integrates your text, but images, as well as your sound and your videos also. Right. Okay.

So let us try to see the practical again, and, uh, we are trying to explore the multimodal capabilities. So we will try to, uh, see that, uh, how visual and textual inputs they, we are trying to input it. Okay. So let me, okay. So we started with the, uh, getting started with GenAI and, uh, introduction. So then we worked upon that various evolution, uh, from since 1950 to present. Okay. So in the 1950s, there was a Turing test. The Turing test is basically a test which is used to, uh, mimic, uh, humans, right? So there was a room in which there was a person sitting, and you will try to generate a sound. So that sound would be a machine sound or a human sound. Uh, that would be identified by the person sitting outside. If this is being identified correctly, then it failed the Turing test. Okay. And if it is, uh, identified, uh, like it identified that it is not able to identify, then we say that Turing test, it is passed that Turing test. Okay. So then there was ELIZA chatbot, right? So this chatbot, uh, was basically, uh, this chatbot was basically made some kind of an instructions to the, uh, what we call, uh, using your language, uh, regage pattern, right? Okay. Then came the neural networks. So in the 1980s, neural networks became very popular by the evolution of AlexNet. Okay. That was the major breakthrough. Okay. So, so then it helped to recognize the images, and from the images, we could recognize the patterns. Okay. So using your, uh, uh, neural networks, or we call as Convolutional Neural Networks, right? Or CNN, right? So in that case, like if the image is being input to a convolution layer, it will try to, uh, analyze the patterns. Okay. From that patterns, it will try to identify that whether this image is of a dog or a cat. Okay. And based on the classification, we can do various classes. So that was this, uh, neural networks, uh, era, uh, around 1980s. Then came the LSTMs. LSTMs by LSTMs, right? And we have also RNNs, right? So the major, uh, disadvantage of neural network was that like it was a fixed length. Okay. So we were trying to have a fixed length, uh, input, right? And the time series and your NLP or your text was not, uh, passed into your Convolutional Neural Network because like it cannot process a sequential data. Okay. So for the sequential data processing, we need some kind of an architecture which will try to input your sequence of, uh, data to the neural networks, that will try to give you the output. Okay. So whether it could be your sequence to sequence, or a machine translation, or a language translation task. So that, uh, RNNs or LSTMs have to detect it. So GRU was another, uh, another mechanism, right? So which have gated, uh, mechanisms and, uh, gates was involved into that, also helped in your sequential data and was helpful. However, there were some challenges of RNNs using your gradient descent and optimization, so that was taken care of. Then, uh, the major breakthrough came with the help of your attention mechanism. Right? So attention mechanism, what does attention mechanism? Uh, what is attention mechanism? Attention mechanism is basically we wanted to focus the importance of a particular word. So this is a word. So this word in this particular word, how much importance is there with respect to other words? Let's say, I, with respect to other words, how much importance is there it is having? Want, how much important it is having with respect to others, right? So this is how we are trying to provide some kind of an attention, or some kind of an, right? In which we could decide that this word, or this word, or this word is having, uh, higher importance. Okay. So that was the purpose because of the attention.

mechanism that would give some importance to a particular word. Uh earlier, all these uh sequential data was having same weightage. Okay. Then it doesn't have any context. Okay. Now, the context we could have a context in which we have to say that that "want" is important, "to" is important, "is" important, "to" or "go" is not that much importance. So, with respect to the other words, okay? So that is the crux of the attention mechanism, that what word is having some importance with respect to the other words.

So, this attention mechanism was later introduced in the Transformer architecture. So, Transformer architecture, what it does? It has an encoder, okay, and a decoder. So, encoder-decoder, these are basically an uh um uh what we call uh complex architecture in which we were having some neural networks, layer normalization, uh then uh skip connections, right? So, and multi-headed attention mechanism where the output could be passed into the decoder, which would be having a mark attention, and we have the output. Okay. This kind of an architecture was proposed post uh in 2017 by Vaswani, right? "Attention All Is All You Need" was the paper, okay? So, there, like it was the state-of-the-art paper in which uh the attention mechanism was used in the Transformer architecture and and it addresses the issues what we have in the sequence-to-sequence modeling in the RNN, right?

So, then with the help of this Transformer architecture, so then the LLMs came into existence. Okay. And then uh now we are having your GPT series, which is a decoder-only architecture of a Transformer. Okay. However, encoder, we can still use it. So, we can independently use encoder, we can independently use decoder, or we can use both. So, there are three kinds of architecture which could be used. Okay. One is the encoder, decoder, and your both. So, particularly your LLM is using your decoder architecture. So, which is helpful in generating your text, right? So, as we have clearly seen that we have generated a summary of your resume, right? And uh it could be your sentimental analysis, that also we have seen, right? Then code generation, right? So, these are the techniques uh we could see uh is very helpful in your GPT series or your LLMs, okay? So, this is the theory behind that how GPT series have been evolved, okay? Fine. Okay.

So, then we saw that what is generative AI? Is then category of AI system that would help in generating new content, images, text, music, or other forms of creative output. So, basically, this generative AI is a subset of your AI, right? Which could be having deep learning and your uh Transformers architecture, right? So, objectives, we could uh recognize patterns and offer a wide variety of options and uh boost your creativity, and we can do multiple tasks, right? So, basically, it not only helps you in your productivity but also into your day-to-day life. It is very useful, and you can use and start working with your uh these uh gen AI tools.

LLM, so it is called as Large Language Models, that is your foundations, and it helps humans to uh do various uh human-like text or generate human-like text across various tasks and domains, right? Okay. Okay. So, then why we call it large? Large refers to a significant amount of size and complexity of these models. So, as I already told you, CH GPT-3 has 175 billion parameters. So, parameters is basically your uh weights, right? We call it, right? In which these LLMs have been trained on. Okay. So, these LLMs have been trained on large amount of data. So, data like in in terms of your GBs or petabytes or TBs, you could call it. Okay. So, it is impossible to train these large language models onto your system. Okay. Right. So, you need high computing power. You need high hardware resources and much powerful GPU machines to train these uh high powerful TPUs or or GPUs or or your uh TBs, right? Right. And then uh also the machine capability also uh you need it. Okay.

So, language, so language because like the primary uh motive is that it is it is your Natural Language Processing. So, NLP, right? So, this is in the NLP domain where you have your uh data which is fed as a text and the model. So, model which we use is basically your Transformers model, and is your decoder-based architecture, right? So, which helps in uh generating the sequence of text. So, these are some of the widely used LLMs like GPT-3.5 or 4, that we have been widely using it. Okay. So, that we already discussed that it has multimodal capabilities, including power to process different formats like text, images, or videos. Okay. And it helps to generate the content, text summarization, sentimental analysis, or data extraction, or language translations. Okay. So, these all examples we have seen yesterday. We have seen how we can convert one language, one text to another multiple languages, right? Or sentimental analysis, or text summarization, or data extraction. So, data session today, we will see that how we can leverage this part.

So, these are the different uh models uh built by different companies, right? And uh every company offers somehow the services to its clients, okay? And and and they say that we are the best, and there is no hard and fast rule that which LLM is best. So, it depends upon your use case and uh on to use that how you are trying to use it, and and and these models are keep on updating uh themselves, right? So, as we speak, so there could be another version which will be released shortly, and this could be much better than your GPT-5 or your Gemini or your Claude, right? Okay. So, these are some of the uh free and open-source tools, so that we have discussed, right? And uh you can just explore these tools, uh like so some of the tools are free, and some of these are chargeable, and some comes with one month subscription or a 15-day subscription. Okay. So, these you need to uh understand.

So, then we did an activity that how we can just uh uh create our resume using our ChatGPT and update with our requirement of a job alignment, right? Then, what is a prompt? So, prompt is the natural language text that helps to perform a specific task, and uh prompt which will lead to a focused and a useful output, and it helps the bridge between the user intent and the AI, right? So, yesterday, I told you clearly that prompt should be precise, should be accurate, should be correct, and it's something like that you are trying to tell something to your two-year-old child, explaining things, and that should be clearly and well-constructed. Okay. It would be structured enough. So, without the proper structuring and the proper uh um uh proper length, so the model wouldn't understand, and it will it can give you vague results.

So, prompt, how does the prompt works? So, it's a uh like it it's a like it's a straightforward question. So, when you are trying to put a straightforward question, the language model provides an answer based on the trained data. Okay. So, how does it work? Let's say you are trying to ask a question, and it has been, this question is being asked to a model which has been trained on huge amounts of data, right? So, the data, uh data, if I talk about, so data is your Common Crawl, if you try to search on the internet. So, this was the C4 or the Common Crawl dataset which was used, and along with that, this is your Wikipedia data is also used, and your internet sites scraping is also used, right? And your news feeds and other things, right? So, lots of lot of data has been used for the training purpose, okay? So, so sometimes it happens like when you are trying to ask some question to the LLM, okay, the recent one. So, let's say, uh like the Indian Prime, the Indian Prime Minister went through some country, right? Okay. So, so this is the recent news, okay? So, this recent news, if you tell to LLM, so maybe it wouldn't respond because he wouldn't know, uh that data is not fed here, okay? The recent one, okay? So, data must be up till some certain point. Let's say, today is 26th or 27th, right? Okay. So, 27th of July. Okay. So, maybe the data is being kept up till 27th of June, right? Last month, one month has been updated. So, this up till 27th to 27th July, the data is been not fed into this uh LLMs. Okay. If the data is not fed, it cannot tell you the events uh happened uh in this course of time. The recent one cannot uh because the data is not available, so he cannot predict, and it might give you vague answer. Okay. Fine.

So, this you need to understand. If you are trying to uh write some prompt and ask some question which uh the training data hasn't been fed or it has been the recent one. So, it will try to generate some answer that is for sure, but the answer wouldn't be correct. Okay. So, think it in this manner. If I try to tell a two-year-old kid that uh uh like what what subjects did you read in school? Okay. So, she might do a tantrum and say that like I didn't uh uh eat the food, I didn't uh go to the school. Something like that. But the she will try to give the answer, but the answer is not correct. Okay. In the same manner, you are trying to ask a question to the LLM, and the LLM doesn't know the answer, but they will try to produce some answer because it's a generative model, it has to produce something, and then you are the judge to judge that like whether the answer given by an LLM is correct or wrong. Okay. So, on your basis of human intelligence, you can just see that the answer which the LLM has produced is is vague. So, he's not giving me the right answer. So, that is what we call it hallucinations, right? So, it tries to give you the answers which are not contextually correct or factually correct. Factually, like it should be factually correct, means it should have the correct facts. Okay. In in in in that sense, like if I try to tell you that what does Prime Minister Modi uh discussed in his current state, right? So, so he doesn't know about the facts, and the data isn't correct, so he's unable to answer that question. So, maybe conceptually, like he may give the right answer, but factually, it is not correct. So, that is what the hallucination. Okay.

So, then we discussed about what are the components that make a great prompt. We have a specific task or an action or of a process. Then we do some set of instructions which we give to the LLM that these are the instructions you need to perform or stick to these instructions, and this is the context. Okay. So, context, and this is your input. So, input could be in the terms of image, text, data set, or context could be anything. It could be it could be information about the text, or could be image, or something like that, right? So, these things are interrelated, the input and the context which you are trying to feed. So, this we have clearly seen an example. So, this was the task which we wanted to create a brief narrative. Okay. And instructions like about the uh wizard quest, and this is the context in a magical forest. Okay. And the input we take from a provided and uh nature descriptions. Okay. So, this is basically how we try to differentiate the particular uh components of a task, the instruction, the context, and the input. Okay.

So, then we did three activities of language translation. Okay. That uh we try to convert a text to a different multiple languages. LinkedIn profile creation. Okay. I guess all of you have created your LinkedIn profile and and and updated on your uh LinkedIn, and then we also did your sentiment analysis. So, this you can see that like a single LLM is capable of handling multiple tasks. Okay. So, we can also do a classification. We can do a text generation. Okay. And here we could do a machine-to-machine or language translation. Okay. So, this is these are the tasks which we intend to do uh in our day-to-day activities or day-to-day uh life. Okay. So, this was the LinkedIn profile and sentiment analysis.

Then we discussed about the multimodal capabilities. So, what are multimodal? So, so there are five kinds of a modalities which a human being has. Touch, we have. Then hearing, then we have your speech, right? Uh, then we have text, or then we have your uh image, right? Okay. Then out of that, uh, this hearing, speech, and your text and image. So, these are the modalities which your uh LLM can process. You can give a text document. You can give an image. You can also give a speech. This could be converted to text, or speech to text. So, this is the power of your right. Okay. So, then we also see one activity where we are trying to give an image of the US, and then we are trying to ask some questions that what are the possible uh tourist locations and how we can plan our stay in in that country. Okay. Yes.

So, then we also see the negatives or or your sentiments, customer feedback. Okay. So, this activity uh we need to do it. So, let's try to do this activity, guys. Okay. So, we will use a customer feedback. Okay. So, this feedback is in the CSV, right? So, we have a customer's feedback CSV. So, that we wanted to put it into positive, negative, or constructive. Okay. So, So, let's try to uh analyze the customer feedback, uh collected after a major sales event, right? So, the objective is to identify areas of success and opportunities for improvement, thereby enhancing is future sales strategies. Okay. Right. So, in the previous sessions of Pandas and your NumPy, right? So, we did this activity quite often, right? So, we have your customer sales data, and we wanted to increase the customer sales, right? And we wanted to predict the future sales also. Okay. So, instead of doing all these operations, so we can just uh put this uh CSV into the LLM, and we can just ask them, okay, give visualize the results, give me some pre-cour steps, and and everything. Okay. It will try to understand the context from the CSV and try to produce the results. Yes.

Uh, so Puja, your limit has been reached. Your daily limit has been reached. Okay, because the thing was that like that is what I was trying to tell you that if you are trying to use the ChatGPT, okay, so it has a daily limit set. Okay. So, so that is why even like I don't know like maybe like this uh would continue uh in this particular session or not, right? So, my my limit could also expire, right? Uh, they you haven't used yet. Okay. So, um, I don't know like because because they have the limit of the tokens, right? Already I told you. Okay. So, that is why they are trying to uh tell you. Okay. So, you can just uh uh write again tomorrow. Okay. Right. So, that is what I can tell you. Okay.

So, let us try to see what the output they have given us. Like the output, the dataset contains customer feedback with element columns like review title, name, rating, date, category, comment, and useful. And in the comments field, they are trying to uh take the positive, negative, constructive, and neutral. And here is the value counts of each of the uh comments, right? So, then it says that it seems like I can do more advanced data analysis right now. Please try again. Okay. So, this this is something you can get more often when you are trying to use this, because you are trying to hit their servers, and their servers could be busy right now, and they they could say that like we cannot do uh some advanced analysis, okay, because it is a free version, guys. So, understand this thing. So, they are trying to market this and trying to uh ask everyone to pay for it so that they can use their service. Okay. So, sometimes like it can do the analysis, sometimes it won't do the analysis. Sometimes like it will tell that your limit, your your your daily limit limit has been reached. Okay. So, I used the chat a very long time ago when it was into what we say, into a beta version, right? So, now Indie is saying that I've got the pie chart. Yes. Okay. So, I didn't get the pie chart, but I like got the pie step-by-step Python script. Okay. This is your Python script in which you can run this Python script. You can copy this Python script, and then you can just see that how what is the pie chart, and accordingly, you can just see that. Okay. So, it is very easy. So, we are trying to write some code, and uh this is how it says. Okay. So, Ash is also given some pie chart, and this is, yes, so, so you can just ask Indrajit, like, "Can you give me a Python code snippet uh to run locally generated pie chart for the uh customers feedback categories or something like that? Can you generate the step-by-step code?" So, by asking this, something that you need to uh give me the code, so it will try to generate the code for you, okay? Right. Okay. So, to make sure your Python packages have been installed, and uh try to uh run this and and see for yourself. Okay. So, this is how the code generation it does for you. Okay.

So, that we were trying to discuss uh in the theory part that not only sentimental analysis, uh your summarization task, or your text generation, right? And multimodalities also, the code completion and your code generation also, it will try to do it. Okay. So, this is very powerful, but use it with caution, and use your brain also, because whatever they are trying to tell you is is some is is right, but you you are the judge, okay? Because as I told you, don't take it blindly that always your ChatGPT and this LL models may be right, okay? So, you just need to verify your data also sometimes, okay? So, that is why your human evaluator or a human expert is needed to evaluate this ChatGPTs or your large language models, right? So, sometimes the other way around, I've been seeing some news that somebody was using this Chalm model for a quite long into their code bases, and accidentally, by like someone told GPT to run some script, and accidentally it tried to delete all the entries, and it popped a message that "Oops, I deleted the employees." Okay. So, that could also be done because it is a generative model, and it can interpret your query. Okay. Uh, but the query which you are trying to write should be crystal clear and precise, so that it may not interpret it wrong, and if the interpretation is wrong, and it may lead to some disappointing results, as I was trying to say that it deleted all the database entries, and it would have a huge impact. So, if you are using something like that, uh try it first, and then use it on the production or some environment. Don't try to blindly use it, because what these guys I understood was using uh in every IDE, these tools get integrated. Okay. So, when they are trying to write their code in some integrated environment, and in in your in your IDE, so, so they are trying to ask the LLM to do the job for them. Okay. In that process, the database entry is called related. Okay. So, maybe you can just uh uh listen also, because uh this thing is limited, right? Okay. So, okay.

So, now the next session is about your powerful explore powerful GPTs, okay? So, this we are trying to uh uh compare the difference between different types of GPTs like get GPT-3.5, 4, and your Turbo, right? The context length and the tokens has been increased, right? So, so yesterday I explained you about the tokens. Can someone tell me that what what are tokens? Yes. The word space is denoted on different colors. Okay. Keywords, number of characters and spaces. Yes. So, so the tokens is basically uh like uh what I tried to show you that was a representation. Basically, it was a representation. Okay. But the actual tokens is that you have a word. Okay. And that word is being converted into some numerical representation. Okay. Numerical representation. So, let's say this feature is a word. This could be have some kind of a numerical representation. This chat GPT-3.5, right? So, chat is let's say 15, it is like 45, this is your dash is your 20, and this is like let's say 113, right? So, these are some kind of a representations we call it as tokens, okay? Right. So, when you are trying to pass like that feature, chat, GPT, something like that, so you have an array in which you pass all these. It will pass like this: 56, 15, 45, 20, and in turn, it is your feature, RGBT, right? And your 3.5, something like this is how it is being represented. So, this you can see that there are six tokens, right? "I want to go to college," right? And these tokens would be represented in the form of token IDs. So, 40 is being represented by "i." 16, 82 is represented by "want," 316 is represented by "to," 810, "go," 2, and this is "to," then this is "college." So, this is what they are trying to represent it by using this thing, right? Let's say I further say that, yes, so there is some logic, Puja, okay? So, these these these how how these are been allocated, okay? So, these are not allocated as such, okay? So, these are allocated by some kind of an algorithms. Okay. So, this algorithm is Byte Pair Encoding. Okay. So, Byte Pair Encoding, what it does is, let's say, uh, so this is the algorithm which is used being your Chat GPT-3. Okay. So, I don't know about the four and four meaning. So, different uh versions of this Byte Pair Encoding has evolved, right? With time. Okay. So, what it does is like, what does Byte Pair Encoding does is, let's say, I, if I try to write here, okay? So, now you can see that "i" is one token, "am" is one token, "going" is one token, "to" is one token, but "call" and "g" is is separately it has been divided, okay? So, every color you are trying to represent is of different token, and this is what your answer is, your question is that like, is there any logic? Yes, there is some logic. What they do is they try to divide the numbers, right? Let's say "call" and "g" or "L" and "LM," right? In future also, if we try to see "LM" or "L," they try to break it down. Okay. So, let's say, uh, uh, uh, "come" and "coming," maybe like they have a different strategy in which they will try to break down "come" and "ing." So, this these are represented as two tokens. This is represented as one token. Okay. So, there are different strategies or algorithms in which they try to uh use this tokenization strategies. So, there is they they, right? So, general thumb rule is that if we take around 100 tokens, so 100 tokens is approximately 75 words, okay? If you are trying to write a 75-word document, okay, that will close to have 100 tokens. So, you may argue that there are 75 words, according to one word per token, they could have 75 tokens. No, that is not the case. So, here in this case, let's say "i," "want," "to," "go," "to," "college." So, there are six words. Okay. Right. So, there are six words. But now here there are 11 tokens they have tried to identify. Fine. So, this is how uh this is your OpenAI tokenization for calculating that how many tokens are being used or or how many tokens this text corresponds to. Okay. So, this is not something like uh uh representation into different colors. Okay. So, this is representation into different token IDs. You can see that the token ID. So, these are your token IDs. So, this is being passed on to your LLM. The what you try to see is words, they are converted into tokens, and then the token is being passed, and in the output also, they generate the tokens, and then it gets decoded, and you get the text value. So, this is the whole process, right? So, I write it down. So, you have your input text, you convert it to tokens. Tokens means this array of numbers, you pass it to the LLM, right? LLM generates these tokens. Okay. And this tokens in turn gets back or decoded into a text which you see on the output. Fine. So, this is how it it is it is working. Understood? Yes. No.

So, how many tokens we can avail on daily basis for a free ChatGPT? Okay. So, that you can go there into that uh particular uh uh in in your ChatGPT. So, there there could be a daily limit uh like I I don't know like basically because you need to check. Okay. So, that is the uh case because they keep on changing because this uh ChatGPT which I have used, this is your uh beta version, right? So, when they try to release some 100 or 2,000 customers, so there I have I have used them used it, right? I I put an email, and then they try to send me the link. Will the number of tokens be different in different versions for the same text? Yes. So, it depends upon the algorithms, basically. In the G, maybe in in your uh some 3.5 version, okay? So, it may be different. Okay. So, let's say this is having 11 tokens, and in the Chat, so it may be different, right? So, so based on the algorithm, so this Chat GPT-3.5 may be using different algorithms or different strategies, right? So, maybe it won't be different, but for the ChatGPT-3, I as far I know that they are using the Byte Pair Encoding. So, there is a sentence pair encoding also, the Byte Pair Encoding. Why the name suggests? Because every byte, every word is being converted into a byte, and then we they are trying to use it. But uh uh there is character-wise encoding also, sentence-wise encoding, and these are different algorithms which could be used. So, it depends like uh that what algorithms this Chat GPT-4.0 is being using, and for the GPT-3, I'm sure that they are using Python imported, and now it has been legacy, right? So, then they no longer use this DP-3, okay? So, right.

Then we were trying to discuss that about the speed and all these things, right? Okay. Right. So, ChatGPT-3.5, Chat GPT-3 is legacy now, and now we have 3.5, 4, and 4, and 4.0 or Turbo. Okay. So, we have faster response, and this is slower, but it will try to handle longer context and more cohesive summary, and it also handles very long context, right? So, ChatGPT-4, as far I know that it can handle a 300 pages uh uh uh uh book, right? 300 pages PDF book, right? In in in its context, okay? Right. So, different prompts or different responses you can get based on your different versions, and also it is not something like that I have asked a question, and you are trying to ask the same question, the response could be the same. So, it must have a different response, right? As already we have seen that uh the pie chart haven't been generated on my system. So, it got generated on one of your systems, right? So, the results may vary. So, it depends upon how you write the prompt. But in sometimes, if there are consistent results, let's say, "What is the capital of France?" Okay. "What is what is who is the Prime Minister of India?" So, there the results might not change. Okay. So, that is consistent. But if you wanted to generate a summary or or a text summarization, right? So, there the results may vary depending upon that how it it perform, and based on the version also, it it may be different. Okay. Right. Okay.

So, now uh we stopped here uh in in your yesterday's session. Okay. So, now we'll be focusing on writing uh such as content generation, writing assistant, or editing. Okay. So, "Write for me" is also a very good tool of your ChatGPT, which gets your personal writing assistant. Let's say you wanted to write an email. Okay. So, there you have some your context, and you wanted to craft it very precisely and in a humble way. You wanted to tell your manager that you are trying to apply for a leave, you are not well, or something like that, or you have made a project presentation, and and and this is the outcomes, and this is the findings. Okay. So, this is very good and useful tool in which it will try to write and uh uh and beautifully craft your mail, right? Okay. So, let us try to see this in practice. Okay. I will try to If we copy the contents of output, will that also count in tokens? Uh no, copy the content means I didn't get, Puja. Uh, you are trying to write one input from okay, and send it to the uh return uh recent time, right? So, these are the tokens, right? Copied my resume today. Copied in the sense, you copied your resume and uh put it into the chat GPT. This is what you are trying to say. So, let's say you are trying to copy your resume and put it it here. This is what you are trying to say. Okay. So, so I'll tell you. Okay. So, what is the uh Okay. So, rest. Yes. Yes. So, now uh yeah, I got your point. What you're trying to say is, okay, so the thing is that let's say this is your resume, right? So, this is your prompt, basically, okay? This is your prompt, okay? This is your input prompt, input prompt, okay? So, these are having some, let's say, uh 200 tokens, let's say, let's assume it, right? Okay. And this is your output, right? Okay. So, this is the output, and this is also having, let's say, like 150 tokens. Okay. So, together they are being added, and this counts that that how many tokens have you used. So, your 350 tokens has been expired. Okay. Right. So, the moment you try to write the input, so these tokens have been calculated, and how much you are trying to ask, okay? And the output which we are trying to generate, that is also calculated. So, this is how it works, understood. So, maybe you have written or copied your uh resume or given your resume, and also you have given the prompt. So, this resume plus this prompt and the output. So, these three things are being held accountable for your token. Okay. So, this is your uh ChatGPT writer. Okay. So, let me clear up the screen. Okay. Go to this site, chatgptwriter.ai.app. Okay. And here. So, here you can see the interface. So, interface looks pretty much good, right? So, here you can write an email asking a satisfied client for a referral or testimonial. And here you have your models, right? So, these are all the models. So, GPT Mini, Gemini, right? DeepSeek, right? So, DeepSeek Reasoning, right? So, so, so, basically like you can also fiddle with other models also, and you can have a writing style, how professional, casual, straightforward, forward, persuasive, friendly, and you can also have a length that give me a short length, medium length, or long length, right? Okay. So, you can just use any of this. So, let us try to uh do one activity. So, I'm trying to write a blog post for our new eco-friendly reusable water bottle, water bottle product, right? The blog post should be should introduce the product and its unique features. Okay. So, first, I am trying to give some instructions that the blog post should give the product, its unique features. Discuss the importance of reducing single-use plastic in our daily lives. And third is highlight the health and environmental benefits of choosing our sustainable water bottle. Okay. These are some of the instructions which I'm trying to get it. Okay. Okay. How do you open this writer app? So, just click on this. I'll just which is informative, engaging, and persuasive style. We can also set this tone to connect with the health-conscious and environmental environmentally aware individuals, right? So, here we can just ask that we can have uh regular uh ChatGPT or Gemini, right? So, this is intelligence rank one, and speed rank is also one. There is a rank two and rank something like that, right? Okay. So, these are all paid versions. You can also use DeepSeek also, right? So, just you can just fiddle with it, any of it, right? So, let us try to uh use it and see what the responses we are trying to. So, you can see and you can try it out. Okay. Let me share the prompt with you guys. Okay. There's an app also for this uh install ChatGPT writer also. You can download the app also and install it, right? Okay. So, whenever you are trying to write an email, so this also gets populated uh there. Okay. So, you can just try it out and then you can just uh uh experiment with it. Okay. So, that ChatGPT which we were trying to use it, uh the uh this this ChatGPT. So, that was your normal ChatGPT, right? Okay. So, it will it can take the task uh like uh uh your normal task, right? But this was specific to when you are trying to write something. Okay. So, the more enhanced version, okay? So, this is your general GPT in which you are trying to ask it. Okay. Okay.

So, then we will also uh use your Canva AI, which you are trying to use it for the designing purpose. Okay. For the specific task, there are specific ChatGPTs have been built up on the top of that, where you can ask questions, and then it would be very useful for it. Okay. So, earlier we were using some Grammarly tool, right? To to correct the errors of our writing style. So, now this strategy is like you can write an email, you can collect that email, you can just ask them, so that it will try to generate an email for you, and which is very uh uh tone friendly, or your uh nice looking, or or formatted, and with correct grammar, right? So, these are the things which you can do it now. You can see that, right? So, here it is coming out, two out of P 15 free responses used. So, out of the 15 free responses, so only two have used, and rest uh you can see. Okay. So, that is like then it will try to ask me for the upgrade. So, basically like there are no freelances, and these CPTs have been uh you need to be cautious while trying to handle it, right? Okay. Okay.

So, let's move on uh further. So, this was "Write for me," right? Which you are trying to focus on your personal writing assistant, right? Your coach, in which you wanted to write some journal, you wanted to ask them that whether this grammatically is this correct, you are trying to write some essay, you are trying to write some uh uh some summary, right? So, an idea, right? And you wanted to uh tell the GPT to be your personal started writing assistant coach. Okay. Right. So, that's the next uh tool is your Canva AI. So, in it also helps in your increasing efficiency in tasks or workflows. So, Canva makes the designing easy by creating designs, okay, from your description. So, let's say you wanted to have a uh design created, and you have a uh some idea in your mind, and you write that idea, you note down that idea, and try to give that in the form of a description, and then you can just uh make that description into a creative design. Okay. So, that's let's try to see this in action. Okay. So, let me open this uh Canva AI also, not there. Okay. Okay. So, go to the internet, and you can just uh view the link. Okay. So, let's open. So, what I'm trying to give it a prompt: "Create an image uh for this social media post to announce the launch of Sunrise Brew." Okay. So, there's the launch for the sunrise view, and I wanted to create a social media post. So, the image should be show the warmth of the morning sun and the rich flavors of the caramel and a chocolate. So, I wanted to generate an image because I have a launch of my brand, and I wanted to use the elements such as sunrise or morning light, and a steaming cup of coffee to suggest a fresh start of the day. Okay. So, now uh the reason why I have given you this prompt is that so you can see the wording of that prompt with the precise clear instructions, telling the ChatGPT that you need to do this. Okay. Use these elements such as sunrise or morning light, right? To and a steaming coffee to suggest the some information, right? The instructions which I'm trying to get give you. The design should be welcoming, welcoming, and energizing, and include the text. So, I wanted to include the text: "Introducing Sunrise Brew. Start your day with a slab adventure." in a clear font. Okay. You can also specify the font, spacing, sizing, something like that. The style should be modern and clean, and targeting your young professionals and coffee enthusiasts. Let us try to hit this and see what that happens and what it is generating. Okay. So, this like what it does is that this ChatGPT Canva is is is trying to load an external plugin, right? Which is your Canva plugin, right? So, it goes into the Canva site, and it talks to that, it it pushes this query to the uh another platform, and take that response. Okay. Okay. So, this is basically they are using some external tools to divert the information or divert the uh input, and from that input, they have been using this information to get the output. Right? Now, you can see that, right? These are the four options they have already checked. Okay. So, looks pretty much good to me, right? If I wanted to start a coffee brand, then it is it is something I could use this one, right? Or this one, because it has a text also. I could edit this text, and then I could see that. Okay. So, here also I could edit some text. Okay. Fine. So, this is also very good tool in which uh so here you can see that "Talk to ChatGPT plugin Canva, canva.com," right? So, from this, it is trying to deviate this information to another tool which will get uh the input and process it and takes the output and back, it is being displayed here. This is how the whole process done. So, it depends like maybe like your screen is getting different uh or maybe it is there it is available in mobile apps too. Uh, I haven't checked it. You can go and check it also, because I don't know like I haven't checked it. BT is being available in mobile also, and if it is having the mobile version, so you can just uh check it yourself. Yes. So, options are coming different. So, yes, maybe you are getting different options, as I already told you that when you are trying to ask some questions, so the questions uh outputs the the out generated outputs may be different for for different persons, and also depends upon the version you are trying to use for the chat. So, tool for uh floor plan interior design creation. Okay. So, uh this tool could also be best because Canva and this thing. Okay. So, you can explore because I I have not uh gone in this direction that which tool is best for floor plan. Maybe I could search it down, and then probably in the next session, I could tell you. Okay. Because there are multiple tools, right? I shared a PPT uh this, right? So, these are like your productivity tools. Okay. You can see okay for the design. Okay. So, maybe for the design, you have these many tools. So, Designs, Hashful, Logo AI, Stocking, right? Then your Daily Stable Diffusion. Okay. And and much more. Okay. So, maybe there's some startup which is working on designing your own house floor plan and interior design creation, that also you can use it. Okay. Right. So, so you can just explore these tools and just uh see that what is best fitted for your requirement. Okay. Right. So, Canva is some tool which we are trying to use, and this thing that is widely used, right? For for your uh template creation, of your website creation. Okay. So, that is what we are trying to use it. Okay. Yeah. You can ask calendar the same question to ChatGPT also. Okay. But these tools, what it does, it will try to redirect to uh like this is basically for the designing purpose. Okay. So, it is not uh something you cannot ask to ChatGPT. They will try to give you different answers also. But but if you are trying to ask uh some designer guy, right? Uh like designer guy, you are trying to ask some uh coding questions. Okay. Okay. So, maybe he will not respond correctly to it. But if you are trying to ask a designer, the design-based question, so he may give you the correct answers. This is what we are trying to do. This Canva tool is basically for the designing purpose only. If you have wanted to do some designing uh uh thing, so then this tool is the best. Okay. If you go for the journal, so journal, your ChatGPT is best. Okay. And here, then "Write me," if you wanted to go and uh uh elaborate or expand in in that writing terms, so that you can uh use it. Okay.

So, now this is another tool uh this which helps in coding, which is your uh automatic code generation, code reviews, or even educational tool for learning programming. Right. So, Designer GPT is a specialized AI designed to instantly create and host beautiful, responsive HTML web pages tailored to user requests. So, for example, you have something uh website in mind, right? And you wanted to uh do some coding, okay? And you want to generate the code for you, and then you can use this code to build beautiful websites uh of functionality. Okay. So, this is the tool which you can use. So, this is your Designer GPT. So, let us try to use this. Okay. So, "Provide a design concept for a website or a product landing page of an online store. The design should include a header with the store logo, right? A navigation bar, a hero section featuring the main product, sections for the product details and testimonials, and a footer. The design should be visually appealing with a balanced colors and scheme and a clear typography." Right? So, just try to hit this, and we will see what it will do. Okay. So, basically like if you are trying to Okay, just see what the uh because it is not giving me the uh give me the code. Okay. It is trying to ask that which image so you would like me uh to use for the hero section and the product images, right? Uh, so I will say DALL-E. Let's try to see also, guys, let me know what uh output you are getting. So, let's try to see uh different outputs. Yes, the PUA is getting nice designs, right? So, it looks impressive. Mine is still downloading. So, such realistic designs it is trying to produce. Okay. So, it failed right on my system because I've been using it quite often, and I can proceed with the HTML layout and provide an image. Yes. So, I could say that yes, proceed, get the HTML layout. Okay. Nice. So, now let's say it is giving you the logo, hero section, product list, all the code it has been giving up. Okay. This is your main content. This is your product features. This is your footer. Okay. So, this Replit is also another tool. Okay. So, you can explore it. So, in the tools that the LLM tools which I showed you. So, this is also a very good tool. So, placeholder is used for the image until the DL image is ready. So, there could be a placeholder in which uh this could be your uh link. Uh, this the image source. This is the image source, right? So, this is the placeholder. You can just put your image there, and then you can just see. Yes, is also getting some nice uh designs. So, it's looks so realistic, right? Okay. So, you can just change the name of your store and the product details, right? You can just uh write it down. Okay. And uh right. So, it will also save lot of time and effort, right? When you are trying to build on that side, and the coding also it will try to give it to you. Yeah, this this design is also good. Okay. So, now you can see that like like there are like 10 different designs which you can get by a single prompt, which 10 people are writing simultaneously, right? Okay. Right. And the kind of

An image which you are trying to get is very impressive. Okay. So these designs, uh, look realistic to me, right? And based on this, you can create your website and you can just, uh, uh, tell them, right? Okay. So by simple prompt, right? So you can just try to ask some questions and, uh, from that, like, it will try to give you some realistic designs and then give you the output which is for the research and the analysis, right? So let's say you are trying to do some research and, uh, this is a consensus tool which is a scientific research assistant, leverage in AI to synthesize insights for academic papers, class.

Okay. So, what it does is, let's say you wanted to do some research and, uh, and, uh, like you wanted to, uh, solve some kind of an, uh, problem, right? For that, you need some research to be done. So that is basically your consensus, which is a scientific research platform, and it will give you concise, clear, summarized summaries and evidence based on the response of your to your inquiries. Okay.

So let us see this in practice. Okay. So I have given you the scenario as well as the prompt. So let us try to see what is the scenario. So scenario is that a researcher is, uh, conducting a literature survey on the impact of large language models on the natural language understanding, that is your NLU task. So they aim to gather and analyze recent studies that explore the advancements in LM. Also, recent advantage of branding, particularly in the areas such as question answering, text summarization, or sentiment analysis, and the researcher is intended to identify key trends, challenges, right? And potential future directions. So based on this, they plan to highlight the role of fine-tuning and transfer learning in enhancing the performance of LLM on various N and new tasks. This is the scenario in which the searcher is trying to give it, right?

So what prompt? So this is the context or the scenario which we are trying to provide. Based on this context which we are trying to, which we intend to solve it, we are trying to conduct a literature survey, right? On the impact on the large language models and, uh, focusing on the recent studies, uh, that involves your question answering, text, and summarization, or sentiment analysis. Analyze the, analyze the key trends and challenging things. Okay. And give a summary of your findings, highlighting the potential future directions for this research in this area. Right. Let us try to see what it will be the answer. Okay.

So now, uh, it has clearly defined me this summary. Okay. So, and the summary, it will try to tell me on the basis of sentimental analysis, or question answering, or the text summarization, according to the task has been separated out. So beautifully, it has given me the links of that papers also, which I can just, uh, see and check, right? So if you are trying to do some research, and the first and the most important thing is that you need to gather information that what is basically, uh, uh, what basically it is being developed or, or the research is going on in that area. Right? Okay. So you just need to do some market analysis of what all papers have been released and based on that, uh, in what direction you can, uh, head to. Okay. So, so you can see that advancements in question answering, right? And these are the papers. Performance in text summarization. So, these are the papers. And then multilingual low-resource adaptation. Okay. This is also a very, uh, paper which is be recent, 2025, 2023, right? Okay. Challenges and the gaps. Performance gaps in low-resource settings, or hallucinations and robustness issues, right? And future directions could include your bias mitigation, explainability, right? Okay. So this is, uh, something like if you are trying to do some research work, so this is basically your consensus, uh, platform in which, uh, it will try to give you this thing. Okay.

So the next, uh, tool is your Universal Primer. So this is also a very good tool in which the fastest way to learn everything about anything, right? Okay. So this we also going to see. Okay. So let us try to see this in action. Okay. Okay. So you can try it out on your own also, right? If you have a particular scenario. So this is a scenario in which, uh, what we are trying to do is we are trying to clarify the concept of chemical equilibrium for high school students. Right? Okay. So, so the scenario is that Alex is a working professional with a background in engineering, preparing an important project presentation that involves understanding of chemical equilibrium principles. Right? So despite this experience, it struggles to grasp how changes in pressure and temperature and concentration impact on chemical equilibria according to the Le Chatelier's principle. So basically, he just wanted to understand the Le Chatelier's principle and, uh, how the changes in pressure, temperature, and concentration impact your chemical. So he wanted to seek straightforward explanation with real-world examples to deepen his comprehension and facilitate practical application in his project work. Okay.

So now the prompt is that offer a practical explanation of Le Chatelier's principle and its relevance in chemical equilibrium, uh, tailored for working professionals. Describe the influence of pressure, temperature, and concentration in non-chemical equilibrium. Right? Okay. So this is the prompt. Okay. So we can directly put the prompt and then we can check. Okay. So it has given me the high-level overview with an analogy that how, like, we can use this, uh, Le Chatelier's principle that offers technically request and immediately, uh, in the near industrial or, or engineering context. Okay. Okay. So you can also come up with your own, uh, uh, basic, uh, prompts, right? And if you wanted to learn something, if you wanted to be curious about how does it work, then you can just use this. So basically, what we have done is we have used four different tools, which is your ChatGPT writer, then we used Canva, then Designer GPT, then Consensus, and your Universal Primer, right? So according to your use case, you can ask to your normality also, but these are specific because it will try to direct that site in which they can access that information. Okay. Right. And then it can use that. Okay.

So, make sure that you are, uh, like when you are trying to use it. Okay. So, use with caution and, uh, try to clearly express your views or ideas in a much, uh, clear language. Okay. So, that would help. Okay. So, for this, like, we had come to the conclusion of, uh, this, not this session, but this slide, that GenAI is a subset of artificial intelligence that leverages machine learning and deep learning techniques to generate data, right? So LLMs could be classified into several ways depending upon your functions and the users. Okay. Functions, like, let's say we have different types of tools which we are trying to use it, okay? So that is how it is being classified in several use cases or several different domains, we can call. What are prompts? Prompts guide the GenAI to understand the context and specific requirements for the task. We all know, right? So we talk to these LLMs or the GenAI by using our prompts. So it also performs various complexities such as creating resumes, LinkedIn profiles, email classification, and sentimental analysis, and day-to-day tasks, which we have clearly demonstrated how it is very effective in creating our LinkedIn profiles or creating our, uh, posts or Instagram posts or messaging, right? Or our blogs, right? Okay. Then the multimodal capabilities, right? Which we have the images, videos, or could be text, can simultaneously process and generate various types of data such as your text and images. Okay. Right. So this was the key takeaways. Right.

So let's try to, so optimizing GenAI models. So this is the topic we are trying to do. So GenAI, uh, designs various types of creative content, right? So that we already seen, including your images, text, and music, emphasizing its role in creative and. Okay. So LLMs like ChatGPT are advanced AI models that go beyond the understanding of human language. So there are other than ChatGPT also that we have not explored, but in the coming sessions, we will be exploring those as well, right? Like your Gemini and your Cloud, or your, right? And, and, and your DSE, right? So these are some of the ChatGPTs which offer some free tools, are there some paid tools are there, okay? So that will also, uh, do it. We will also try to explain you that the practical APIs implementation, that how we can use it in our code. Let's say we are trying to build some, uh, code or, or a platform, right? Which uses this GenAI tool. So that also I will try to make sure that that you have an APIs. Okay. So you can, uh, get these APIs from OpenAI and you can write your Python code in which you can build your own GenAI or ChatGPTs. Okay. Right. So what you are trying to do is you are trying to build a platform where you can write something and, uh, internally your API, it would be called to your OpenAI and then it will feed the response and you will provide that response. So that also we can do it, right? So that also I will try to, uh, if time permits, I will try to, uh, have a session in which the API processing, I will try to teach you. Okay. Right.

So now, uh, so we have seen that, like, how the summarization happens, right? Uh, if you are trying to provide the prompts, and in this, end of the session, what we will do is we will try to generate effective prompts and optimize interactions with GenAI models. Okay. Now the only thing challenge what we have is that how to write the prompt. How to write the prompt. Okay. What are the ways in which we can write effective prompts? Right? Because all these things matter a lot, right? The prompts which we have written has given us right kind of an information what we are trying to achieve. But what are the ways, what are the frameworks in which we can write effective prompts? Okay. So that we will try to, uh, see advanced prompting techniques such as zero-shot, few-shot, chain-of-thought, self-consistency, and tree-of-thought, uh, generate the desired output from the GenAI. So these are some of the techniques which we are, uh, learning now in this particular, uh, slide, that how we can leverage this thing and, uh, improve the, uh, output according to our use case. Okay. So this is what is called as fine-tuning. We are trying to fine-tune our prompts. Okay. Right.

So let's say how, how it will help. Right? Let's say you have an LLM, okay? And you write one prompt, P1 is the prompt, and you feed this to the LLM and it gives you output, let's say O1. Okay. So now, as a human, you will see that this output is not perfect according to my specifications. What you will do? You will have a feedback mechanism in which you will try to update this prompt, okay? Right. So you will modify this prompt to a little bit that it would have some clear instructions. And if the instructions are missing, then you feed again, again to the LLM and generate the output O2. Okay. And then you will check whether this output is what I need. If not, then again, you need to refine this prompt. Okay. So this is what is called as instruction fine-tuning. So the instructions or the prompts which you are trying to use, you are trying to fine-tune that according to your use case and you try to do up till now that that when, when you get the desired output. Okay. In this case, like it could hallucinate and it can provide you vague answers, then you need to change your prompt and then try it again and again. It's a basically a trial and error and basically on the right side of instructions or, or some, uh, structured output or structured input can give you some nice output, as we have already seen in our case that that the, we have provided the, uh, instructions very clearly and precise, and it gave us the nice templates or the UIs designs, right? In, in, in our, in our case. Okay.

So now we'll begin to the instruction to fine-tuning prompt engineering. Prompt engineering involves, uh, conversational AI models such as ChatGPT to generate the desired output. Okay. So this we all know. These all chatbots are your conversational AI models or AI tools. So a well-built prompt must have clarity. So clarity in terms of your text. Okay. So you have a clarity in your mind that I wanted to make this application, I want to generate this kind of an output, but from that mind, you just need to have a textual information. Right? What kind of a text you want? Let's take an example. If I try to tell my, uh, like, two-year-old kid to visit, go to the shop and, and buy some sweets from that shop, right? So how I would direct that child, right? So it is basically as you are trying to write an effective prompt to tell the, tell the LLM to give me the, generate me the output. So how what would I say? I would say that, okay, uh, you come out of your house, okay? And you drive or you walk around 50 meters, okay? And then you see a left, uh, turn and you again go 10 meters and then you see two shops, okay? And from one of the shops, which named, let's say, grocery store, so you go to that shop, buy some, uh, sweets and come back the same distance as you have walked. Okay. So this is a set of clear instructions which we am trying to give it. Exact information that you walk this much space, you go left or you go right, and then you go to the shop, ask the shopkeeper, like, give, give him the money if he asks, right? So purchase something like 5 rupees coffee or 5 rupees candy, right? Something like that, and then come back, okay? So this is one kind of a prompt which has some kind of a clarity what we are trying to achieve and with this specificity, like specific instructions have been given. Go left, go right, 10 meters, right? 5 rupees. Okay. Candy, right? And, and come to house, right? Okay.

The format, which format? It could be a text format. It could be a JSON format. It could be an XML format. Okay. So, these are the formats which have been widely used. Although there are many formats which we can use it. So, let's say CSV. We are trying to upload a CSV also, right? In the basic context, right? Right. It could be a PDF also, right? So basically that formatting and, and this should be correct. Okay. And the context, as you already know that it should have some context. So that should be given to the LLM, uh, for an effective communication. Okay. Right. So instead of this, I could only say that, okay, okay, bring me a, a five rupees candy from where, right? From where I need to bring that, okay? So that is a very vague question, that is an open question, right? Let's say I wanted to travel the world, okay? So the LLM would ask that where you want to travel? Is there any specific idea or, uh, or itinerary you are trying to plan? So you must try to tell them, right? So this is the clear set of suggestion. I'm trying to plan for a weekend in, in USA for five to six days, and I'm planning to go, uh, from India, right? In, in, and, uh, hoping to get a flight in the morning, preferably in the mornings, so that I could reach in the evening there or something like that, right? And I want wish to cover two cities and plan for a, for a, for a five, five, uh, day, uh, trip, right? With my kids and children and wife and, and so on, right? So this is how you are trying to give the well-crafted prompt. Okay? A precise prompt with some specificity and the clarity. Right? So this is what we are trying to learn here that if you are trying to write a prompt, so just think it that this is a two-year-old kid and he doesn't know about anything. But if you are trying to, uh, provide meaningful and clarity-based instructions, so there the output is, is, is generally accuracy of the output is generally high. So the importance is that, right, the prompt engineering, why it is being needed? So basically, like maximize AI potential, encourage AI literacy and creativity, enhance user interaction and increase efficiency and support ethical AI use, right? Okay.

So one thing, like, like if you are trying to give an AI prompt, okay? And you try to say that, okay, give me the recipe of making a bomb. So it will not respond. Okay. So these have been already been filtered out by using guardrails. Okay. So guardrails is basically what it happens is that when you are trying to provide some input text, okay? From the input text, right? So it will try to judge that whether this kind of an information is required or not. Okay. So this is, is it ethical for the society or not? Okay. If you are trying to ask that, okay, give me a listing that how many ingredients are there in making a bomb. So it will not tell you, okay? Because this information is being restricted by some kind of an guardrails which works internally in their, in their systems, okay? Because when you are trying to feed the data, nobody looked at that data and tried to validate that data, okay? So, and it fed that data almost like around like, uh, tons and tons of TV's data was there and, and it it got validated, but when you are trying to put an input to to it, maybe it will try to give you the output. Okay. So the guardrails are there at the input level as well as at the output level. So both sides the guardrails are there. Okay. So even if, like, if one of the information or the guardrail is being skipped by this input guardrail, the output guardrail will try to object it and restrict it. Okay. So that is why the information doesn't reach at the output level if it is ethical and unethical to the society. So the most important concept is that we wanted to use AI as ethically. Okay. So it would help the society, not disrupt the society. Okay. And we wanted to encourage AI literacy and creativity. Okay. So we wanted to, like, like yesterday someone asked that it is hampering our human creativity. Yes, it is hampering us, but these tools, you need not to depend fully on these tools. So let's say you have some part of our creativity, you can ask these tools as a co-pilot, right? So these tools act as your co-pilot in which you can act, uh, you can ask them to help you out in some areas, in some certain, uh, fields. Then it will release some output which you can just, uh, see this output and, and make, make use of your applications. Right? So this is how the importance of prompt engineering works. Okay.

So let us try to see the prompt examples. Okay. Right. So this is a simple prompt, as I already told you that I wanted to travel the world. So write a catchy caption for my Instagram post. So what is catchy? Right. You haven't explained the catchy caption, right? So, so ChatGPT just gave the answer, right? What it thinks. Okay. Embracing the beauty of simplicity in a chaotic world. Simple joys. Find yourself. Okay. Right. So according to this knowledge or the data which is being fed to the ChatGPT, so it will try to, uh, uh, uh, label this term as catchy or, or the output is catchy. Right? We think that this, this term could be a catchy. Okay. According to the chat. But if you try to write some sort of, uh, uh, uh, precise or, or, or, or some set of instructions which are clear enough, right? Craft a concise and engaging caption for the Instagram post featuring our refreshment rate. So we are now targeted on refreshment rate. Okay. And I want you to have a concise and engaging caption. The target audience for this post are young, young adults, travelers, and party girls. Okay. So this post has to be kept in mind that which class of customers we are targeting. The class is your young adults, travelers, and your party. And also include five, five, uh, high, high-performing hashtags related to this refract. Okay. So now you can see that this was a, a simple prompt, and this is very engineering, prompt-engineered prompt in which you are trying to add some, uh, details to it. Okay. Fine. Right. So this is how you wanted to write, uh, not just random prompt, but some kind of an engineered prompt in which it will give you the desired outputs. Okay. Right.

So now we'll move to the advanced prompting techniques, right? So let's say there are some techniques which is called as zero-shot prompting, few-shot prompting, chain-of-thought prompting, self-consistency prompting, and tree-of-thought prompting. Right? So, so these are basic for you, uh, techniques, right? So there are other more techniques also. Let's say ReAct, reason and act. So it calls that reason. We wanted to reason the instructions and then we need to act on that instructions what we provide, right? So there's a CAR framework also, clue and reasoning, uh, framework, right? First, we deduce the clues and the reasonings, right? And then we try to feed it. And there are so many other, like, based on the papers which we try to see, so there are multiple frameworks the users have used in, in terms of this, uh, prompt technique. Right? The basic one is chain-of-thought and your tree-of-thought. Right. Okay. So we'll try to focus on this and we will see that that how it will help when your results are not optimized, and then you can do, uh, some sort of tweaking in your prompts and then it will give you the optimum results. Okay. Yes. So are you, are you following guys? Yes. Any questions? Any doubts? Yes. No. Yes. Anyone? Okay. Fine. Okay.

So, let's talk about zero-shot prompting. Okay. So, one is zero-shot, uh, short, or one-shot. Okay. Right. So let's try to understand what is this zero-shot, few-shot, and one-shot. Okay. So zero means zero. Few means there are few samples. One means one sample. Right? So let us try to understand this. So zero-shot. So we are trying to create a social media post for a brand. Okay. So, so we are trying to create a social media and here we are not providing any examples. Okay. So, let's say, uh, we could say this, uh, I wanted to generate a social media post. Right. Right. So this is your prompt with the zero, zero example. Example means that you are not providing any example, and example you can provide here. I wanted to generate a social media post. For example, you can give that social media post. So this is something I wanted to generate, right? So you can give one example, two examples, right? Let's say you come across some social media post, Instagram, and you find this post very intro and you find it very creative. Okay. So let's say post one, and you want to write or, uh, do, uh, your social media post similar to this post. Okay. So then you can put this post here. I wanted to generate a social media post based on the below post. Okay. This is like an example of one-shot. You are trying to provide some examples to the LLM. Okay. So this was the original prompt, but you are trying to provide some example. If you provide one example, then it is one-shot, and if you provide more examples, then it could be few-shot, and if you don't provide any example, then it would be your zero-shot. Fine. So this is how you zero-shot, one-shot, and two-shot, uh, prompting is done. Why it is being done? Let's say you are giving some additional context to your GPT, telling them, okay, I wanted to generate something, and I have seen something that based on this, uh, example, you generate some content. So what it will do? It will try to see this example, based on that example, it will try to generate it. And if you try to provide more examples, the more is the better. Okay, but this more comes with the cost of the tokens, as I already explained you. Okay. So you can give as many examples as you like, but that comes with the cost of the tokens, right? Okay. So, so generally, it is being seen that instead of providing zero-shot prompting, you can provide one-shot or few-shot. So that helps to give the output better because you are trying to provide, you are trying to tell something. Let's say there are two friends. Okay. Right. So, so they wanted to, uh, uh, shop some kind of a dress. Okay. Right. So one, one friend is trying to explain the other friend that that I wanted to, uh, wear that dress. He's trying to explain. So this friend will try to say that, do you have any example of some dress that I could see? He says that, yes, I have some dress. So this dress, some looks something like this. Then this guy, the guy two will explain, uh, will, will reason it better, and it will try to understand it better that, okay, so this guy is trying to look for a dress which looks similar to this pattern. Okay. Right. So this is what we are trying to do here also that when you are trying to feed the, uh, inputs to the LLMs. So you provide some example that let, let's say this is the example of some, some, uh, picture or design that I just want, then based on this design or example, you wanted to generate me some output. Fine. So this is your zero-shot prompting. Okay. So it provides the model with a limited number of specific examples in a definite context. Okay. Right. So it will not train the model. So this you need to understand. It will not train the model. Okay. So we are just trying to provide the context to the model, right? With examples, so that it could take that examples and try to understand from that example and try to give us the right kind of an output. So this could be called as in-context learning. We are trying to learn in context. We are trying to provide that context, okay, inside your prompt, and from there the LLM is trying to learn itself from using that examples. Example one, example two, right? So this is a kind of an data which you are trying to provide to the, and from this data, it is trying to learn in context. This is called as in-context learning. Right? Fine. So this is what it does. So generally, in-context learning or providing examples is a better way in which you can perform the accuracy of your output. Okay. Understood guys? Yes. No. Okay. So what, what you haven't, uh, got it? Okay. Okay. So I, I'll repeat it again. Okay. Right. So, Okay. So let's try to see one prompt. Okay. So this is a, uh, plumboxy. I couldn't pronounce it correctly. Okay. Right. So this is something a word. Okay. So this is a word. Okay. This is a word. Okay. So this word could be a fantastic creature that inhabits the enchanted forest of Orofland. An example of a sentence that uses the word, uh, plumboxy is. Right. So we are trying to, you give an example of this word. Okay. So we have tried to give that example that this word is a fantastic creature. This is the definition of that particular word, and we are trying to give that example. Now we are trying to put that another word and we are trying to say that, uh, this is a word, and you can, you give me an example of the sentence that uses the word quiz wizzler. Okay. So here we are trying to give one example of that particular word which we are trying to use it. So this is your one-shot example. Okay. It tries to learn from this first context and then based on this context, it will try to generate it. So this is what is called as console. And if you try to give multiple, uh, examples, then it would be called as. So when you are trying to, so this is your LLM, right? Okay. So you are trying to provide a prompt and it gives the output. Okay. So writing a prompt involves three processes, which is your zero-shot, one-shot, and few-shot. Okay. Zero-shot is basic. Okay. So let's say you are trying to write a prompt which you wanted to write an Instagram post. Okay. Based on some ideas which you have got into your mind. Okay. You are trying to write it. Okay. So this will not have any examples. Fine. So you feed to the prompt and it will give you the output. Now, another one-shot, you wanted to write an Instagram post. Okay. Based on some idea which you have on a mind, also you can give an example. Let's say this was the Instagram post of my friend depicting this, this information, and it shows this hashtags and so and so, right? So you also can give that this is the example, and this is the pattern which I want to get my ideas generated because Instagram post could have like, it will try to generate the pattern itself, but you are trying to restrict here that, okay, this is the example which I'm trying to give it. This could be the pattern which I'm thinking about, okay? So this is one-shot example in which you are trying to give one example to it. Few-shot is that you can give more examples to it. Okay. In one example, it could be a pattern. Another example, it could be another thing in which your, uh, some creative ideas or the creative text is written. Okay. So that way it comes up you or one-shot. Okay. So basically, this helps to the LLM to understand what you are trying to think in your mind. Okay. The examples and the context. So it tries to learn from these examples. It learns, not trains. Okay. So training, we are not trying to train, we are trying to learn from these examples which, uh, you are trying to feed it, and this is what is called as in-context learning. Okay.

So now the another method is your chain-of-thought prompting. Okay. So what is this? It, uh, generates a sequence of outputs rather than getting the entire output at once, right? So first, you need to do some, uh, sequential outputs, and then we pass it on the another, then pass it on another, and then it follows a chain, right? Okay. So it, uh, helps to, uh, get a detailed response, okay? Because we are trying to focus on one part first, from the explanation or the detailed output, we are trying to focus on the other, and so on. Right? So this is how the chaining of the thoughts is, is, is actually doing the process under the hood. Okay. So let's say, what are the main threats to global diversity? It has given me the answer. Then, given these threats, based on these threats, it is trying to ask which conservation strategies are proving to be more effective. So we are trying to break the part, break the statements or the prompts in parts. Okay. And this is a very effective manner because let's say you have written, what are the main threats to world diversity? Based on the threats, which conservative strategies are providing to the most effective, right? So you can write in, in, in a one single statement also, right? But the thing is that that if you try to, uh, work on it on the first part, maybe this, this is up to your requirement or not. So this you need to check it, whether it is, it is up to your requirement, then you can change or modify this prompt. Basically, this is the scenario. So one output becomes the input to another, right? Another prompt, and then you can just make a chain and then evaluate your response. So this is how your chain-of-thought prompting, uh, uh, works. So the second example, how do these strategies integrate with local human communities? Okay. So you can just see that, uh, the answer has given that by involving local communities in decision-making, providing alternate livelihoods, prompting education and awareness, right? And what are the emerging technologies are supporting conservation efforts, right? So one prompt, you make it, give the output, and then again and again, right? So this is how you deal with the chain of thought. Not you are trying to give the prompt as a whole, but break it into parts and then evaluate step by step. So this is the main motive behind chain-of-thought prompting. Understood? Yes. No. Any doubts?

Then the self-consistency, right? So self-consistency prompt is that it delivers reliable and similar responses when asked the same question repeatedly. What is the capital of France? Paris. So consistently, it will try to answer Paris, right? So this is self-consistency. We all know that it is based on the fact that the capital of France is Paris, and it will not change. Okay. So in terms of your text generation, when you are trying to generate something, right? It might give you some, uh, different answers, but in terms of your, uh, self-consistency prompting, where the output is fixed, so there the outputs won't change. So we can also check it ourselves also. So these all things we will do practically in the next session, right? So how we can do, uh, your, your chain-of-thought prompting or zero-shot or one-shot, right? So this part I have dedicatedly like focused on the theory part so that you can understand it, and even if you don't understand it, so in the next session, we will be doing our hands-on to it. Okay. I will be giving you some examples, and there we will be seeing the practical implementation of it. Okay. So the reason why I wanted to, uh, take the theory first is that so you get a double revision of it. Okay. Fine. So this is the prompt that what is the largest planet? Largest planet in our solar system? That is Jupiter. And if you try to change the, uh, uh, prompt with some other, uh, modifications that, can you tell me which planet is the biggest in our solar system? Still the answer remains the same, right? Because here it tries to deduce the context that what context we are trying to refer to, and what is basically here referring to, and the results are almost the same.

So the last part is your tree-of-thought prompting. So what it does, uh, it forms a tree with various thoughts or paths or steps resembling a branch of a tree. Right? So each part is basically a potential step or idea towards solving a problem. Okay? So these all these steps can contribute to the problem-solving. Right? So, so then the AI evaluates these steps that which one is the best solution, and it will try to give that best solution for, right? So it will write, try to, uh, uh, divide the task or the complex problem into multiple parts, and it will see that which part is the best for the problem solver, or it can combine these four steps and to generate one output. So this is your tree-of-thought form, right? So here you can see that find x into x + 3 = 1. Right? So it is trying to solve step by step. Step one, step two, step three, and step four. Right? So this we can also say that, uh, think and act, or reason. Right? So these are all the things which you can involve, and there it will try to break down the problem into smaller steps and then evaluate and arrive at the correct solution.

With that, we have come to the end of this course. If you have any doubts or questions, let me know in the comment section below, and our team of experts will get back to you as soon as possible. Thank you and keep learning with Simply.