What Is Generative AI

hello all my name is krishnaik and welcome to my YouTube channel so guys a new playlist some days back I had actually uploaded like how chat GPD was trained and chargpt as you know it is an llm model again it is based on generative AI techniques itself so in this video also I am going to discuss what exactly is generative AI completely from Basics uh the reason why I'm creating this particular playlist because I feel that in this upcoming two years right there will be job related to specifically generative AI because there are now many many startups getting opened who really want to work in generative AI they're creating their chat Bots they're creating their uh you know image generation tools video generation tools and many more things right and this is all possible because of generative AI they are there I hope everybody has also heard about prompt engineering right and a lot of job openings are also related to prompt engineering this playlist will focus on covering all these kind of topics uh in this video we're just going to focus on generative AI but later on as we go ahead we are going to discuss about models Lang chain I'll be showing you practical implementation how you can create your own custom land chain models or llm models just by using open AI API itself and again some amount of prompt Engineering also I'll discuss about all these terms in this video also but as I start every playlist of mine I really want to consider it from basic topics I really want to discuss where does generative AI falls into picture whether it is a part of deep learning where does it fall in deep learning what is the differences between the Deep learning models that we learned in CNN RNN when compared to generative area all these things will be covered in this topic if you are new to this Channel please make sure that you subscribe the channel or share with all your friends because there are a lot of interesting contents that are going to come up right so let me go ahead and let me share my screen because this is some amazing materials that I've created you for this particular tutorial itself right related to generative AI so first of all I hope everybody knows about what is AI artificial intelligence what is machine learning we usually say machine learning is a subset of AI as you can see in this diagram and deep learning if I probably consider it's a subset of machine learning in machine learning and deep learning we basically learned two important types that is supervised machine learning and unsupervised machine learning I hope everybody knows the different references about supervised machine learning and unsupervised machine learning supervised machine learning is nothing but it is specifically if you have labeled data right at that specific point of time we use supervised machine learning you can solve problem statements like uh regression classification forecasting many more things as such in deep learning also we have supervisors and supervised techniques in this specifically we learn about CNN we learn about RNN we learn about different different NLP Concepts you learn about object detection and many more right and before I think in my channel I've covered almost everything along with projects end-to-end implementation and many more things everything you'll be able to find out that is the reason why I'm now upscaling more because definitely many people had actually requested about generative AI but now the question Falls that where does generative AI fall into picture so that is the reason I have created this diagram generative AI is the subset of deep learning very much clear you can see this green color circle it is subset of deep learning right and again what is the meaning of generally where I'll just let you know in some time right you will probably be heard about large language models llm models right one of the example is chat GPT right so chat GPT is an llm model you also have Google bard right this is all are llm models and this kind of models are trained with huge amount of data right and definitely all the NLP simpler tasks like touch translation text conversions or acting as a chat bot itself text summarization all those tasks can be done by this llm models itself where does llm model fall into picture again it is a subset of deep learning and it has some properties merge with generative AI I will be discussing more about large language models but if you see if you probably check out my previous video how chat GPT is trained right I have created a dedicated 30 minutes video with respect to that charge if it is a kind of llm model and is based on generative AI we'll discuss about more like how you can create your own custom llm models by using Lang chain many more things as we go ahead but right now let's focus on the basic guys always try to learn things from Basics when you will try to learn things from Basics trust me your knowledge will be very very good you will be able to crack interviews right now let me go ahead and let me talk about deep learning in deep learning till now probably uh whenever I say about CNN a and RNN right these all kind of models are called as discriminative models in deep learning you specifically have two things one is discriminative right we basically say discriminative techniques the other one that we specifically say is called our generative techniques right now that is the reason why I'm saying you because generative AI will be a part of generative technique in discriminative technique you will be doing tasks such as classification prediction right the data set is trained on label data set right indiscriminative so image classification cat and dog classification object detection and then classifying right all those part are the part of discriminative technique whereas if I talk about generative tasks such as you generate a new data trained on similar data set right so let's say that I am training and training my generative model on a huge Wikipedia database so what it will be it will be able to tell you different kind of answers sentence completion right it can be it can actually help you to generate new data itself let's say you have a generative model which has been trained with some music it will be able to generate its own music or just you give the part of the music it will be able to train with with respect to the it will be able to give you the output of the new music itself charge GPT what it is doing you're giving some uh input over there it is being able to explain you some output or give you some output with respect to considering some context and uh again the output is fabulously right right and that is how you are doing a conversation with chargpt right so generative models right so usually in deep learning when I say discriminative this is based on supervised unsupervised and semi-supervised right and similarly in generative you don't need to provide supervised data set itself you can provide different kind of data set because here you also require reinforcement learning right so in generative AI what you'd specifically do is that you give unstructured content for the training purpose it will be huge data set huge data set from internet from websites from tutorial websites right it will not be a specific label data set as such right it will just try to you know in in in in this kind of supervisor and supervised model tries to see the uh relationship between the input and the output here it tries to see the relation between the distribution of the data specifically in the generative AI right and this is becoming really popular specifically if I talk about two amazing things that are probably going to come up right and many many companies are working in this one is generative language models generative language models charge GPT is a generative language models right and one more specifically is called as generative image models I hope you have heard about dally too what is Dali 2 doing it's taking a text it is converting into an image right so it is being able to generate a new image from a specific image it can even create videos super important guys it is important that you know where does generative AI fit it is a subset of deep learning and it specifically works on huge amount of data you don't need to specifically provide a label data it is not possible when you have a huge amount of data you'll just try to see the distribution of the data to try to find out the features that are related to the distribution of the data within that and you'll be able to create this kind of models right now let's see one example with respect to discriminary technique let's say I have some label data I have some music over here and this is labeled like this whether it is law or Rock classical right so this is my label data and this kind of data I passed through my discriminative model and it is being able to classify whether it is Rock whether it is classical or whether it is romantic right so this is basically a kind of prediction classification right this is called as a discriminative technique over here obviously I can use RNN I can use lstm RNN I can use different kind of techniques over here right to do the prediction but if I talk about generative technique in generative I will give the music right I'll be probably training my generative model with different different kind of music and then based on this distribution it will be able to generate the new music or it will be able to complete some music like let's say I'm giving some initial tone to my genetic models that will be able to complete the remaining tone right stories right if you if you ask chargity write me a story where two AI are probably talking about like how AI is going to change the future in the world if I just give this text my generative models will be able to create a story regarding that right so that is how in short whenever we talk about generative AI it is creating new data it is generating new data in short right so this was what is generative techniques again let's see one content over here how the training process works I will be having an unstructured content which will be going as an input with respect to the generative AI model it learns patterns and distribution in unstructured content obviously it is not so simple uh there are a lot of other techniques in generative AI the main purpose to make the accuracy better human supervision is also required or the kind of reinforcement learning the feedback through reinforcement learning is also required if you really want to know the detail more about it like you can definitely see my check out my video like how chat GPU is trained from there you will be able to clearly understand this kind of materials only I have created over there right so the generated output will be the new content okay it can be a text it can be music it can be image it can be videos it can be multiple things so in order to just distinguish between generative AI whether an app is a whenever application is a generative AI or not if your output is in the form of number class probability categories we have we can categorize that it is not a gen AI if the output is in this form but if the output is in the form of text audio images video frames then it is a gen AI okay so basically it is generating text it can generate audio it can generate in images it can generate videos many things as such right now guys why gen AI is becoming very famous why this open AI API Google body API can really really play an amazing work because later on I'll be using even Google vertex there is something called as generative AI Studio I'll be showing you every practical example over there like how quickly it is being able to generate and even you can use that particular apis itself in order to create your own custom chat bots in order to create your own custom uh this kind of models which I have actually mentioned over here generative image models even that option is also there right so how machine learning is involved till now initially in machine learning in deep learning or let's say in a n we usually had features we have independent features we have output feature right my model used to see how is the relationship between this independent and the output feature and based on that the prediction was done then came deep learning models in deep learning models we were able to give images we are able to give video frames and then we are able to perform tasks like image classification object detection right even there when we are giving these images and images specifically or video frames we are not saying that okay this feature belongs to this this like it has cat eyes let's say cat versus dog there we are just specifically giving images in this we give like okay cat has two years to two eyes you know one mouth something like that the ear shape can be pointy the nose shape can be this right these are the features that we usually given machine learning but in deep learning models specifically related to CNN rcnn all the object detection technique there we are just giving images or video frames and then we are able to do the task like image classification object detection in case of generative AI it is very simple consider you are a human being okay I give you a book that is related to cat let's say okay I've given you a book related to cat let's say 500 pages five different books you have learned that all in one month now generative AI will act in such a way that if I ask you any questions that are related let's consider that you have now become a chat Deputy for cats okay now if I ask you a question you are able to give the response and that response I'll just ask okay tell me about cat and you will be giving multiple different answers two to three lines of answers then I can ask you more information about a cat what does cat usually like in terms of food then you will be able to give the own answer that is how this kind of genitive AI language models are basically trained specifically I'm talking about generative language models there will be different kind of data that will be available the models will be trained on that specific data and then as a generative AI whenever a question is specifically answered you will be able to generate your own output okay through this kind of models and similarly you also have generative image model where you gives kind of images you'll be able to generate images you will be able to generate videos you'll be able to generate another image right and similarly in generative language model whenever you're giving text right you can see in charge repeat if I give text translation can be happening summarization can happen q a can happen image generation Can Happen video generation text to speech can also happen in charge GPT right now this three functionalities is not there probably in charge GPT 5 in GPT 5 it will be coming they are targeting obviously it has been announced that even image Generation video generation text to speech will happen right now you may be thinking crash fine this all are very very good but how do we use this and that is where you can probably use this open AI API and you can create your own chat bot through prompt engineering prompt engineering basically means through what format you will be able to get your response that kind of format you can basically train your own custom models with right and this is the job right now it is very much popular even in U.S market you can see that people are getting good amount of money people who know how to work with open AI API along with prompt engineering techniques different different techniques right that how you can probably get the response that is specifically called as problem engineering the way you get a response from an llm model is all dependent on this prompt engineering work and again I'll be creating a series of tutorial with respect to prompt engineering where we will be using open AI API itself right so I hope you'll you like this particular video uh I hope you got an idea with respect to generative AI because there are a lot many things that are going to come up with respect to this in the next video we're going to discuss about LM models as I always suggested I always try to take it from basics please do wait there will be many videos as we'll go ahead in this specific playlist so yes this was it for my side I will see you all in the next video have a great day ahead and please do make sure that you subscribe Channel I'll see you on in the next video have a great day bye bye take care

Channel: Krish Naik

Views: 75,365

Keywords: yt:cc=on, Generative AI, Artificial intelligence, Machine learning, Deep learning, Neural networks, GANs (Generative Adversarial Networks), Creative AI, AI-generated content, Computer vision, Natural language processing, Image generation, Text generation, Music generation, AI in art, Design and creativity, Future of AI, Innovative technology, AI applications, Algorithmic creativity, krish naik

Id: 2m7Pgl-84F8

Length: 15min 50sec (950 seconds)

Published: Sat Jun 10 2023