AI Made Easy: Dive into the World of Generative AI for Beginners #mlwithakansha #generativeai #live

Hello. Hello.

Hello. Hello everyone, welcome to the webinar. In today's webinar, we are going to discuss about generative AI, what generative AI is, how it works. So myself, Akansha Yadav, I'm an engineer who's currently working closely with IBM as an SME in the domain of artificial intelligence, machine learning, Python, data science, and data analytics. So are you guys ready to dive into the world of generative AI? If yes, type Y in the chat box so that we can begin with our journey of generative AI. Till now we heard that artificial intelligence is the leader who is leading the entire technology, but nowadays we are hearing one word which is generative AI, which is booming a lot due to its capability to mimic the intelligence of humans. So today we are going to see what exactly the generative AI is, how it works, and where it comes in the picture of the technology. So generative AI is also the— due to its capability So let's begin and let's continue our journey of generative AI. So let's first see what exactly generative AI is. There might be some new learners who don't know about generative AI, so I wanted to give you a quick example that you guys might have used ChatGPT. OpenAI, which has built a ChatGPT for us through which we are searching for the content with— like, it's, it's been easy to search for the content. So generative AI, the realistic example or the trendy example I could say, is the ChatGPT. So generative describes a class of statistical model that contrast with the discriminative model. There are two types of model: one is generative, another one is discriminative. So generative Generative, which generates new things, and discriminative, which helps you to, you know, cluster the things or classify some things for you. If I say generative AI, it is a type of artificial intelligence, because in generative AI also there is a term called AI, which is— which stands for artificial intelligence, which can create new content, original content. It can be any type of a content like an image or a text or music, even a code. If you are a programmer, you can use generative AI applications, generative AI APIs to code something or to get some ideas about the new code that you wanted to generate. You can imagine an AI that can paint a picture for you that even doesn't exist, but it is in your imagination. You know how it could be, you know about what do you want in that picture exactly, and you can tell AI about that particular idea, and AI can generate the exact sample of that image. You can even write a song in the style like your favorite artist, maybe. So that's the power of generative AI. As I said, there are two types of model: generative model and discriminative model. Generative model generates new content new data. It doesn't take the reference— it takes the reference from the data which has been used to train it, but it generates the new data. It's gonna be the new words or the new data instance, basically. And discriminative model discriminates between different kind of a data instance. I will give you more examples about it, so stay tuned till the end of this webinar. So let's first discuss that where it stands in today's world. What are the things that we are using nowadays, a day-to-day life, and where we are experiencing basically generative AI? So the very trendy example would be the AI-powered friend suggestion on social media. Whenever we open our social media, it can be either Instagram or Facebook or any other like X, which is Twitter, So it gives us the friend suggestion, your known, your family members, your college friends, your school friends, like that. So how all these things are taking place is just because of this generative AI. Auto recommendation on social. Suppose I am trying to post one picture on Instagram, so what my Instagram would do is my Instagram would give me some suggestion to tag the picture with whom I have— suppose I have taken a picture with my friend and my friend is also on social media. So what, uh, the machine learning and the generative AI will do is generative AI will try to figure out that whose picture is this from the entire, uh, social world on internet, and it will recommend the name of that person to me that, okay, this person is this.

Would—

do you wanna tag this person along with you're like while you're posting the photo. So that is what auto recommendation on social stands for. Another example would be generative AI-powered email assistance. Email is the most useful thing these days. If you wanted to send an email to your boss or to colleagues, or you wanted to share some files to your friends, you are using emails for that purpose. So the example would be like, um, there is something called filtering mails, right? These These days you might have seen that in your mail section there are 4 different tabs: inbox, social, promotion, spam, and many others. Like updates is also one of them. So what mail is doing is— Gmail is doing is Gmail is able to filter out the things for you. This is gonna be in your inbox, this is the social thing, this you are getting from LinkedIn, maybe Instagram, maybe Facebook. This is the promotional thing, these websites are doing promotion and sending you such emails, so that will come under the category called Promotion. Another is the Spam, that this is not good for you, this is gonna harm something, this is not valid thing, this is illegal kind of things will come under Spam. Updates are the trendy thing which you will get under this category. So Gmail has filtered out these sections for us and makes our— make like our thing easier. So like regularly I don't want you to see what is being on the social thing, I just wanted to get the update if my boss has sent me something or my colleagues have sent me something like that. So I will go on to the directly to the update section and I can check it. If I am more onto the social person, I will go to the social tab and I will get the data from there. I will get the mails which is related to social only. Another example on Gmail is that when I am typing a mail to someone, when I am typing, I have written that I would request you to do this, I am looking forward to the positive response, something like that. So once I will write one or two word, it will suggest the third word or the complete sentence to me. This is how generative AI is making my thing easier. My day-to-day life has been easier just because of generative AI, because I have been using it in daily routine. I will tell you how it is, like what are the different application which you guys can also use. Another is the Alexa. If you have bought that Alexa product, or you have the Alexa in your mobile phone, or Google Assistant, something like that, like similar things I'm talking about, like Alexa, Siri, or Google Assistant. So how this helps is, and how, where the generative AI plays a role in this picture is, suppose I wanted to make a science project and I know the topic, I have the topic has been given by the teacher and I know the topic, but I'm not sure what I have to do with this, what I have to exactly create, what all do I need to make it. So what I will do is I will ask Alexa, create a shopping list for me. For this project. I wanted to create a science project, this create a shopping list for me, and it will create one list for you which includes everything. Another personalized experience that we— everyone, like everyone likes to have is on shopping places like e-commerce experience, Amazon, eBay, or Myntra. Which gives us different discounts. Suppose I am searching for something and it gives me the discount as per my suggestion, like whatever I'm trying to select, I'm trying to search for, it will give me the relatable discount on the same product. Even it does the another part of thing, like when I am buying a laptop, let's suppose, so it suggests me other things which I can— may— like I may forgot to buy while buying one thing, but it will suggest me so that I can, you know, complete my shopping at once. So these are the realistic example which we are experiencing. Now let's see how generative AI works. These are all the things which are done by generative AI, whether it is friend suggestions, auto recommendation, or email assistance, or Alexa, or e-commerce experience, and there are many more. These all things, how are these things works?

So this is how it works.

I have just created a simple blocks for you to make you guys understand it. So first what we do is we get some data. On the basis of that data, after getting that data, we train our model. We train our model, we test our model, and we deploy it. We train it, we test it, and we deploy it. Once we have done all three things, then comes our foundation model. It means our foundation model is ready. It can be any model. It can be CNN, it can be RNN, it can be transformer-based model, it can be any other model. How it works is foundation model will generate content. It can be any content— text, audio, video. Once we have our content generated, that is the output. But how the things works is based on the training, testing, and deployment and the foundation model. Foundation model is the main concept over here in generative AI because there are different types of neural architecture. How artificial intelligence is working is because it is trying to mimic everything that humans can do. So this is my original picture on my left-hand side, and the website which is aiavatar.generator.com, this helps me to create AI avatar. It means there are multiple, uh, images that can be created using the one picture which is mine picture, and I tried to create one AI avatar of mine. And these are the AI avatars which generative AI has created for me. So this is the AI avatar creation, or we can say that this is the example of generative AI. There are many more examples. If you wanted to create an animation, what you can do is you can move on to the site called Leonardo.ai, which will help you to create image And then for— once you have the image, whichever image you wanted to create, you can move on to the another website which is the convert.leapix.com/animation. This will help you to create animation. So you can use these websites like AI Avatar for building a social media page. Which is not even of human. You can create an AI avatar or for your display picture of in social media, or you can use animations to generate the content or to post something relatable on your social medias and grow it. Another example is you can even create a video. I will just show you which I have created using the studio.did.com. This is even though a paid one, but you can use the free version of D-ID, and this is how it will work.

It does have audio as well. My name is Akansha Yadav, and I'm passionate about technology.

How it works is basically I have given it an image. Once I have the image, I have added a script. After adding a image, I have added a script, and then I just pressed a convert button and it converted an image to video. That's how it works. That's how— that's easiest thing which you could ever do. For creating this video, I won't be able to do it manually because lip-sync is also the main part in the video. But AI, generative AI, can do this for you. If I talk about the importance of generative AI, then there are multiple things which can be covered in it, but I have highlighted 4 of them on my screen. Areas like creative industry, if I talk about art, design, music, or entertainment, these industries are— need some, you know, the songs which we are able to— there are some similar songs. But if you wanted to revolutionize some things, then you can use the generative AI and you can design something which is exceptional and which is gonna be the unique thing. In science and research, it is doing some— it's doing a lot of work basically, but which we can do is we can accelerate drug discovery and material science part also we can work on. In case of a product development, we can generate innovative ideas. Suppose, like, before something is published into the market, which are some— before something has been, you know, sent in the market to, you know, just to test it, there is something called prototype. So for making those prototype, what you can do is you can use generative AI in the product development and get your things done with the innovative, uh, creation of some things. There are something called, uh, in the ex— if I talk about the things that we are using day to day, like mobile Okay, so let's continue.

So I wanted to watch this video, but I don't have 24 minutes of time to watch this complete video. I just wanted to know what exactly has been shared by Microsoft in this video. So what I will do is I will move on to the another tab, and what you have to do is you have to search for the ChatGPT and This one, this is the Chrome extension which you can download, basically add on to your Chrome, and you can use this. This is not an ad, I'm just sharing my references to you to learn effectively or to get some important things. What you can do is you can get this, you can search directly onto the Google that YouTube summary with ChatGPT and Cloud AI, and there will be one box there will be one button which will say that Add to Chrome, and you just have to press that button. It is not, uh, for— it is not visible to me just because I have already added. So what you can do is you can click on it and this will be added. So into the extension section, you will find that extension. So what you have to do is, suppose this is the video which I wanted to see, that there comes one block which is— which says Transcript and Summary. So I will press the down button and it will show me the entire transcript of this video. In one go I got it, right? I don't have to watch the complete video.

This is not it.

What you can do next is there is one button called this one which will take you on to the OpenAI ChatGPT 3.5 version and it will give you some key points out of it. So it has taken the entire summary, entire summary of that video, and it says that Summarize the following in the 5 bullet points. Even if you want more points, what you can do is you can edit it and you can get 10 or 12, whichever, like totally depends upon you. So it gave me 5 points, basically 6 points.

Yeah, it basically give you the key points which are important from that video.

You can edit it to get more about it if you are interested to read it. So you can summarize the following things in as many bullet points or as like there's a title also given, transcript also given. So you can use this method to learn effectively if there is something related to history or Paul Sainz, or there is a news section, something like that. So you can use it effectively, I would say. This is the best use of generative AI for the learners who wanted to learn more in this time, I would say. Let's move on to the next section. There are many more applications of generative AI which you can use. One of them is using the chatbots which are created by multiple, like OpenAI ChatGPT, Bard one which is created by Google, Bing which is by Microsoft. If you wanted to create videos, you can use Runway, Pictory, or Descript. For productivity, if you wanted to create your portfolio, you can use Notion AI. Notion is one of the websites on which you can create a portfolio, or Taskade is one of them. MeetGreeK is also good one for your productivity.

You can—

in the Taskade, you can arrange your task, to-do task, or weekly task like that. For design, you can use Midjourney or Firely or Microsoft Durable websites. If you wanted to create website, you can use 10Web or Durable or Imagick. If you are a coder, yes, this is on the coding platform, so I would suggest you definitely about the coding as well. So Copilot X or Ask Codey or Code Whisperer is the trendy one which I have mentioned in this. You can use any of them for your code purpose. There are many more code AI, you can search directly onto the Google for that. For presentation, if you wanted to create, you can use Tome, Decktopus, or Gamma. Or automation, you can use these things. So you can even take a screenshot of this for your future reference and use it later. So let's see what is the difference between AI and generative AI. Yes, both the terms are different, AI and generative AI. There is a difference between generative AI and artificial intelligence. So let me give you one example. So let's suppose I have a book, and this is going to be generative AI. So let's suppose I have a book, and in this book I have added all the details about cat and dog. Specifically for cat and dog, right? So what AI can do is— AI, ML, or DL— what these things can do is this can classify. So with the help of this AI, ML, and DL, what I can do is I will give a new picture of a dog and I can ask that, what is it? Is it a cat or is it a dog? This is the capability of ML/DL, machine learning or deep learning. And if I talk about generative AI, what this can do is, suppose I have given a complete book, I will say that, go read this book. And later on, what I can do with the generative AI is I can say that, okay, so you have read this book, now tell me what is CAT. So what I'm doing over here is I'm looking forward to create new content. Which is gonna be in the category of text generation. So what, what generative AI can do? Generative AI can create new content, but ML and DL can do classification, clustering, regression part of thing. So that's the basic difference between generative AI and artificial intelligence. Let's see where it stands in this picture. So earlier you guys might have seen on Google or on social media or other creators who are talking about artificial intelligence, machine learning, and deep learning. So the basic picture was there.

So in the generate—

where when the generative AI came into the picture is after the deep learning. So let's understand each of the things. In depth. So what artificial intelligence is, it is basically, uh, a thing where— which has the ability to imitate human intelligence, basically to mimic the human intelligence, to use the human intelligence for the reference. It's using the human, human intelligence for the reference, and it's gonna create a same machine. And, uh, machine learning is another thing which is automatically learning from the data. In machine learning, we provide the data, it learns the pattern, it learns through the pattern of the data, and it can do multiple tasks like prediction or classification, clustering, regression, and many more.

Deep learning and machine learning are similar.

It, it totally, uh, differentiate on the basis of the data. If you have small data, it goes with machine learning. If you have complex data, it goes with deep learning. And deep learning is algorithm that mimics human brain into machine. And generative AI, as I told you from the beginning of this webinar, that it creates new things. It create new content, image, uh, or audio or other things. So as I said, AI is a broad category in the computer which aims to create a machine that is capable of doing— that is capable of doing the task which humans can do, that typically requires human intelligence. The task like reasoning, which is the toughest part, another is the problem solving, speech recognition, learning, or language understanding. These are the things which humans can do. This is done by the human intelligence. So what artificial intelligence is dealing with is artificial intelligence has aim to create a machine that is capable of performing all these tasks. In machine learning— machine learning is a subset of AI, just like in computer science, AI is a subset of computer science, and machine learning is a subset of AI— that focuses on developing algorithms and models that enables computer to learn from the data. Instead of being programmed to perform a task, machine learning learns from the patterns, make prediction based on the input data. I will tell you about the evolution in this industry, in this technology, how this is evolving from past 20 years, from past 30 years. We will discuss that thing as well. In machine learning, there are various techniques which has to be used— supervised learning, unsupervised supervised learning, reinforcement learning. It works in various domain as well. It is— its main task is to, you know, without programming, how the task can be completed without programming for that particular task. Programming you have to do obviously, but for the data, on the data, you have to train basically something.

You don't have to say that, okay, do this or do that.

No. Another thing is the deep learning, which is a subfield of machine learning. As I said, deep learning and machine learning are the same thing. Deep learning is a subfield of machine learning that evolves around the neural network. Neural network is because of our brain. In brains, we do have neurons and neural structure, basically. So deep learning is also following the same thing and try to create— trying to create multiple layers through which 3 things can be done. If there is an input layer, then there is a hidden layer, then there is an output layer. In the hidden layer, it tries to study the pattern. It's try— it tries to, you know, observe the things from the data and keep it in the brain so that further on in the output stage, it can make some prediction. It can do some several tasks which can be done using the human intelligence. These networks are also known as artificial neural network. This DL can perform multiple tasks like image recognition, speech recognition, natural language processing, and other complex pattern tasks as well. So this is going to be very interesting thing because I am going to relate all the three— all these three things with the realistic example. So let's suppose artificial intelligence— artificial intelligence is an entire kitchen which is filled with various tools. Every tool that we use in cooking, it's in the kitchen, right? So artificial intelligence— consider artificial intelligence a kitchen which is filled of tools and appliances. Moving ahead, in artificial intelligence there are two subsets: machine learning and deep learning. Till now it was, but the addition is the generative AI. We will discuss it But you know what you can say, suppose is like in the artificial intelligence we have two domain, V subset, uh, is the machine learning, one of this, and deep learning is the second one. So machine learning is the stove which is capable of cooking delicious meal based on recipes. Artificial intelligence, which is everything— machine learning, deep learning, generative AI, all the things— consider it as a kitchen. Then machine learning is something which is capable of cooking. Based on the recipes that you are going to provide. Recipe over here in technology is the data.

On the basis of the data, it is a—

machine learning is capable of doing multiple tasks based on the data. So it is like a stove which is capable of cooking delicious meal based on the recipes which I— you are— which you are going to provide. Then comes deep learning, which is a powerful oven which is capable of doing more complex cooking tasks with advanced features. Like cooking can also be done on the oven and even in the— like even on the stove. So both are same, but deep learning is capable of doing more complex cooking, just like deep fry— like baking and other tasks which cannot be done on the stove. And machine learning is able to do something based on the data. Then comes a generative AI, which is also a subsection or subpart of artificial intelligence. It's like a food processor which is capable of creating entirely new dish and flavors based on its knowledge of existing ingredients. There is a website called SuperCook which is gonna be easiest way for you guys to learn more about the generative AI. So what on this website, what you can do is you can select the ingredient which is available in your kitchen, and after selecting the ingredient, this will suggest you multiple websites based on the ingredient which you have. So generative AI is something which you can create new. You have the data, you have the things. So obviously this website has the dishes, uh, the recipes, entire things, but it is suggesting you something new, new dishes. There are different machine learning models. Generative AI is not limited to all these tasks. Like, there are multiple tasks like text-to-text generation, image-to-image generation, image-to-video generation. These are the multiple tasks to be done using generative AI. So there are multiple models which are working in the backend. One is the GAN model, autoregressive model, and variational autoencoder. These are the 3 models. Fourth one is also there, which is a diffusion model, but which is not in the picture right now just because the industry is still working on it. So as of now, I would, I would just say that you can consider 3 models: GAN model, autoregression model and variational autoencoder. Let's discuss the GAN model. GAN model is a neural network which is based on the deep learning, obviously, and it has two, two parts. One is a generator part, another one is a discriminator part. Generator part tries to create new data that is similar to the training data, while discriminator part tries to tell the difference between the real and the generated data. This competition between these two parts helps the generator to improve its ability to create realistic data. So let me give you one example of the same and how this entire thing works. So let's see, in— if I talk about the generative AI GAN model In the generative AI GAN model, there are certain things which we will be, uh, as I told you, there are two things. Main thing is one is the discriminator, another one is the generator. Generator model, and this is the discriminator model. Block. So I will just quickly recreate it after just removing everything.

Okay.

In the GAN model, there are two blocks. One is the generator block, another one is the discriminator block. Suppose there is one prompt given. Prompt is basically a sentence or the detailed information of what do you want to exactly create, and it's given to the generator. So generator will try to create one image. And backend, on the backend of the data generator, there is one image already provided which is in the training part, which is the original image. And this is gonna be the fake image which is generated by the generator. So what discriminator will do in this is discriminator will try to fake, try to, you know, uh evaluated whether the fake image is exactly the same like an image which has been asked by the user and which has been provided in the training. So on the basis of that, it will give the feedback whether it is positive or negative. If it is positive, it will move to the output block directly. This image will be the output block. If it is negative, in that case it will return back and the generator will create the new image again., and the process will continue until another, the competition, wins. Another model in the generative AI is the autoregressive model. Autoregressive model is also the transformer-based model. We also call it a transformer-based model, which is a type of a neural network that generates data one element at a time. Each element is generated based on the previously generated element and the training data. Auto-regressive models can be used to create a variety of, of data types like music, images, etc. Then there comes a variational encoder, which is also a type of a neural network that learns to compress the data into lower dimensional latent space. It works with a diverse range of data. Variational encoder works with a diverse range of the data and it rapidly reduces the dimensionality. It works basically when you wanted to recreate the image, when you are working on something called, uh, data compression, when you wanted to anomaly detect something. Like, suppose I wanted to detect a disease using the ECG signals, so that is the part which variational encoder takes. So, and image synthesis Basically, variational encoder is trained in the static environment. Where comes— like, in the variational encoder, there are 4 parts. One is the original image, there comes an encoder, then there is a latent space, and then the decoder and the reconstructed image. Just like I showed you in the GAN model, just like I told you in the GAN model, that I hope now my PPT screen is visible to you all. So in Encoder, let's picture this again. There are 4 parts. One is gonna be the original image. This is the original image. Then there is one encoder, then there is one latent space, and then there is one decoder. After decoder, the output is generated, which is the reconstructed or the new image reconstructed output. As I told you, variational encoder works on—

suppose there is one image and that image wants to be like—

I want that is of 100 KB and I want that image to be converted into 10 KB. It means I am regenerating the same image but by decompressing. Compressing it.

So that is the main task which is done using the variational encoder.

So this latent space, the entire thing is done using the static environment. And in the encoder, what will happen is it will work on 3 things: it will study, it will isolate, and then it will create. It studies the probability distribution, then it isolates the most useful data variable, most useful data variable. I think my whiteboard screen was not that visible, so I will Rediscuss it on the screen. So I was talking about the encoder. In the encoder, study, isolate, and create works. It studies the probability distribution, it isolates the most useful data, and then it creates the compressed representation of the data sample. This is how variational encoder work. And if I talk about the latent space, I will quickly add that picture in the screen. So yeah, this is what I created. And in the latent space, I was talking about mathematical space. This is basically a mathematical space which stores the large dimensional data in the compressed format. From the original data, it will take the useful data, it will study the probability, it will isolate the thing. And when, when working on the isolation after working and creating something, it will store the large dimensional data in the compressed format. And in the stage of the decoder, it will decompress the compressed representation which has been done by the encoder, and it will maximize the likelihood principle and minimize the differences between the original image and the reconstructed image. And then it will produce the output. That is how variational encoder works. It works with the image synthesis, which creates the new game maps or anime avatar. These are other things done using the variational encoder model of generative AI. And in case of data compression, it can help you— like, this is also the example of the variational encoder, which can help you to forecast the very, uh, variety of, you know, surface of the stocks when you have the stocks data and the graphs has been displayed. So there are certain data previous data, maybe of 10 years or 12 years. It will take that data and on the basis of the data it will do the data compression and will show you that specific data. That's how it works.

There was one more—

I told you about the GAN, that there are two models, generator and discriminator. And in variational encoder there is a latent space encoder encoder and decoder. There is one more which is autoregressive model. It is also known as the transformer-based model, which focuses on valuable text, filter unnecessary elements, and model will be, you know, long-term text dependency only.

So all these things are happening in the industry.

So let's see that how these things are evolving. So earlier, when the artificial intelligence was into the picture, it was a rule-based system. Rule-based system means which is programmed directly by the engineers. So programmed by the engineer means that the normal programming that we do— suppose I'm taking an example in the Python coding— that if temperature is greater than 20, then it is hot, and if temperature is less than 20, it's cold. So this is directly programmed by the engineers.

Programmers—

engineers have already mentioned that these are the parameters that you have to these are the conditions that you have, and you have to use these conditions only to suggest some things. And suggestion is also provided by the engineer. Then the next step was— after rule-based system, the next step was the machine learning, which was able to predict something. How it works is basically it has the data, it does the data collection, it trains the model, it tests the model. Once the model testing is done and the model meets the criteria that the accuracy of the model is good, the errors are come— like errors are less what we are expecting from the model, then we will deploy that model. And that's how machine learning was working.

It was based on the data.

Then when, when the deep learning was in picture, deep learning works on the neural network which mimics the human brain structure and excels in NLP, which is natural language processing, and does the image recognition part. So how it works is it takes the neural network, it creates multiple layers, multiple depth layers to understand and to save the data, and then it trains the model, tests the model, integrates with the smart appliances. And in today's world where we have the generative AI like OpenAI-based ChatGPT models or Bard or Gemini, or we have other Bing. So all these things are generative model which gives us the content creation which makes us feel like our personalized environment is being created. On the same thing, you can generate new tasks for particular, uh, task to be done. So all this generative AI is the— beyond the predefined rules which was done in the rule-based system. Rule-based system was all based on the rules which has been programmed by the engineer. But in case of the generative AI, we code nothing. We just make our model learn from the data, understand the pattern, and understand what has been defined in the data, what are the labels used, what are the things which we can do with the data. So that's how our artificial intelligence industry is working. And there is one more model coming into the picture, which is the diffusion model. But still, the model is not yet deployed or completed, we can say. So this is all about the generative AI. So in case if you guys have any questions, you can drop it down in the chat section so that I can take your questions from there. For more learning about artificial intelligence or generative AI or Python programming language, how you can learn, or how you can be an artificial intelligence engineer, or how you can create models, you can subscribe to this channel and like to this webinar so that we can create more content on the same.