AI-3018: Fundamentals of Generative AI – Complete Beginner's Guide! - video Dailymotion

SkillTech Club

Welcome to this in-depth AI 3018 session, where we explore the Fundamentals of Generative AI! Generative AI is transforming industries by enabling machines to create text, images, code, music, and more using advanced deep learning models like GPT, DALL·E, and Stable Diffusion. Whether you’re a beginner or an AI enthusiast, this tutorial will break down the core concepts of Generative AI, its applications, and how it works.  🔍 What You’ll Learn in This Video: 1️⃣ Introduction to Generative AI & How It Works 2️⃣ Key Technologies Behind Generative AI (LLMs, GANs, VAEs, Diffusion Models) 3️⃣ Understanding Training Data & Model Optimization 4️⃣ Applications of Generative AI in Real Life (Content creation, AI coding assistants, Image & video generation, Chatbots) 5️⃣ Ethical Considerations & Challenges in Generative AI 6️⃣ Getting Started with Generative AI on Cloud Platforms (Azure, OpenAI, Google AI, etc.)  🛠️ Who Is This For? Beginners & AI Enthusiasts exploring Generative AI Developers & Data Scientists interested in AI model training Professionals & Businesses adopting AI-powered solutions Students preparing for AI certifications & careers  📌 Key Highlights: ✅ Clear & simplified Generative AI explanations ✅ Real-world examples of AI-generated content ✅ Hands-on introduction to AI model architectures ✅ Discussion on ethical AI & responsible usage  💡 Get started with Generative AI today & unlock the future of AI-driven creativity!  Explore Our other Courses and Additional Resouces on: https://www.youtube.com/@skilltechclub

Transcript

00:00hey friends good morning good afternoon or good evening thank you so much for joining us today

00:14today we are going to focus on fundamentals of generative ai the agenda of this particular

00:19video is going to go like this first we are going to talk about what is generative ai and then after

00:24that we'll talk about different language models which are available we'll talk about large language

00:29models small language models also we are going to talk about copilots and ai agents and how we can

00:35build or customize them we'll also talk about adopting generative ai in your business so what

00:40kind of benefits you will get if you adopt generative ai in your business we'll talk about microsoft

00:46copilots which are available in almost every product of microsoft nowadays and then we'll talk

00:50about considerations for prompt how we can exactly use generative ai for generating desired content

00:58using different techniques of prompt engineering and then after that last but not the least we'll

01:02talk about extending and developing generative ai apps so let's get started first of all what is

01:10generative ai if that is your question let me start with what is ai artificial intelligence

01:16definition says that ai is going to imitate human behavior by using machine learning to interact with

01:22environment and execute tasks without explicit directions on what to output basically that's

01:28going to imitate a human kind of behavior with your software applications now this kind of ai enabled

01:35software application will imitate human behavior so that you can feel like they are able to understand

01:40what kind of an input you are providing on the other hand when we talk about generative ai

01:46generative ai is basically a subset of ai in which an ai model is going to create original content in

01:53the response of your natural language prompt so basically user will provide some kind of a natural

01:59language input by either saying something or typing something and then based on that the generated

02:05result is going to be an original content it's not something which is coming from some existing data and

02:11search it's not something which is copied from somewhere it's a newly generated original content

02:17and that's what generative ai is all about now generative ai application is going to take this

02:22kind of natural language input and it's going to return an appropriate response in variety of formats

02:29as you can see in this slide the formats are actually three different formats natural language

02:34generation image generation and code generation basically natural language generation is going to give

02:40you a response in a human specific language and in this case you might submit a prompt like give me

02:47three ideas for a healthy breakfast or maybe you can ask for write a cover letter for my resume and when

02:54you give this kind of an input prompt is going to give you a response in the natural language text

02:59on the other hand when you're using things like image generation you're going to give a prompt create an

03:04image of an elephant eating a burger now when you do this thing it's going to generate a new image

03:10with the same kind of requirement or maybe you can try this one which is create a logo for a florist

03:16business and it's going to give you a logo associated with that third but important one code generation

03:22is also the third thing which is associated with generative ai in which you can actually generate

03:28code if you're a developer in this case your ai applications are designed to help software developers

03:34to write a code for example you could submit a request like show me how can i code a game of a

03:40tic-tac-toe with python or maybe you can just put a python code to add two numbers kind of a prompt and

03:46it's going to generate a logic associated with that remember it can generate logical code with multiple

03:52languages like python javascript c chart and few more but if you want to understand generative ai properly

03:59the next thing which you need to understand is language models now while the mathematical

04:04principles behind language models can be very complex a basic understanding of the architecture

04:10which is used to implement them can help you to gain a conceptual understanding of how they work

04:16now in today's cutting-edge technology large language models are based on transformer architecture

04:22this is one of that famous architecture based on which you will find all the latest language models

04:27like gpt3 gpt4 or some other models from other companies like meta and microsoft this kind of

04:34models are transformer models and these transformer models are trained with large volume of text which

04:40are enabling them to represent the semantic relationship between words and use those relationship to predict

04:46probable sequence of the text to make sense about that so if you're interested in learning transformer

04:52model in depth you can just comment down in this particular video and we will create a separate

04:57video for you on that but as of now this slide is actually trying to show you that in the left

05:02side first section you're going to provide your training text now remember this is a training text

05:07which is a text-based raw content you're going to provide that thing and transformer model is going

05:12to first take it in the encoder box this encoder block is actually going to encode the text is going

05:19to convert the text into tokens and then in the vector and once this tokens and vectors are generated with that

05:25is going to understand what kind of an input prompt you have provided the generation of the new text

05:31content is going to happen in the second block which is known as decoder block in the decoder block

05:37they are going to understand the relationship between each word with the other words in that particular

05:41statement now obviously this is a complex process but this encoder and decoder blocks are actually going

05:48to have multiple attention blocks inside which this encoder block and decoder block are going to have

05:54multiple attention blocks in between which will actually help this process in the step-by-step way

05:59at the end of the decoder block you're going to get your output which is generated based on your

06:04input which you have provided now this diagram is just trying to give you a very very high level

06:09bird's eye view of how exactly transformer model works but in order to understand this thing in depth

06:16you can search for transformer model architecture and that actual architecture is actually going to take

06:21few hours to understand that in depth but yes it's going to be worth of your time it's very important

06:26if you are a data scientist then you understand a transformer architecture if you're not a data

06:32scientist you're just a business user or a developer maybe this won't be that much useful for you so you

06:37can just focus on generative ai applications more than anything else now let's talk about some language

06:43models which are available in today's time the first thing which we want to discuss right now is

06:48foundation models remember organizations and developers can train their own large language

06:54models from scratch but in most cases it's very practical thing to use an existing foundation model

07:02and then optionally if you want you can fine-tune those foundation models with your own training data

07:08when you do this thing you're not only going to save a lot of time you're also going to save a lot of

07:13cost which is involved in creating large language models on microsoft azure and the azure open ai

07:19service which is available inside that it's actually including curated set of models from open ai which

07:25are hosted on microsoft azure cloud this is actually going to offer a benefit of cutting edge language

07:31models like generative pre-trained transformer models like gpt models what we call and you also have some

07:38kind of image generation models like dal e which is helping you to generate images with that now all

07:44these models are actually available in the model catalog of your azure open ai service and your azure

07:50ai foundry portal is also having this now if you don't know about azure ai foundry or azure open ai

07:56then stay tuned we are going to learn this thing in this particular course after some modules then stay

08:02tuned we are going to learn this kind of models and this kind of portals very soon in this particular course

08:08if you ask me what kind of models are available as of now in the model catalog of the services

08:13well some of the names are mentioned on this slide you can found microsoft models open ai models in that

08:19hugging face models mistral even meta models are also available in this the very recently some of the

08:26new models like you have deep seek models are also available in the model catalog of azure ai foundry portal

08:32so this is really cool thing and it's going to take you a lot of time once you decide which model you

08:37want to go on with you can fine tune you can customize and you can use it now let's talk about

08:43a little in depth about large language models and small language models now as the name suggests there

08:50are many language models which are available that you can use to power generative ai applications and in

08:56general we always have two different differentiations large language models and small language models as the

09:02name suggests large language models are trained with the vast quantity of text that represents wide

09:08range of subject matter related data now in this data you are actually going to have billions and

09:13even trillions of parameters inside that on which those large language models are trained basically if

09:19your model is trained with the higher number of parameters and more amount of data then it means that

09:25your large language model is more capable compared to your small language model but processing that

09:31particular data and then generating the responsive text from that is also going to be more expensive

09:36because it has to go through that whole bunch of data whenever you are going to use fine tuning or

09:42other customizations with the models so basically large language models will be more capable more accurate

09:49but it's also going to be more expensive on the other hand when you're focusing on small language models

09:54these are the trained with more smaller or some subject focus data sets it's not going to have a wide

10:00variety of data sets and parameters on this but it's going to be focusing on some particular focused subject

10:06on that this focused vocabulary is going to make them very effective in a specific conversational topic

10:12but it's going to make them less effective at more general language discussions so basically this is going

10:18to be something like that small language models will be expert of a specific focused topic but not

10:24about some general discussion kind of a thing the smaller size of slms are going to provide more

10:30options for deployment including local deployments to devices and even on on-premise computer and that

10:36is going to make them much faster and easier to find you compared to large language models well if

10:42you ask me right now which one is better well do not come to this conclusion that this is better for me

10:47because in your case depends upon your project depends upon your client and depends upon your company

10:53you maybe have to choose small language models or large language models and based on that you have

10:58to decide how you're going to use them how you're going to fine-tune them or where exactly you're going

11:03to deploy them now let's talk about copilot and ai agent now generative ai apps are often integrated

11:11into applications as a chat interface and this is something which most people know because of the famous

11:17application called chat gpt they always provide contextual support for common tasks in those

11:23applications you ask a question and it's going to give you a response microsoft copilot is also a

11:28generative ai based app that is integrated into a wide range of microsoft products if you are a business

11:34users then business users can use generative ai to boost their productivity and creativity with ai

11:41generated content and automation of tasks developers can also extend microsoft copilot by integrating them

11:49into their business process and data and even they can create a copilot like custom agents in to

11:55incorporate a generative ai capabilities into their own applications and services which they are developing

12:01this slide is actually showing you microsoft copilot which is enabling you to summon a generative ai

12:07chat app where you are working on a windows or a microsoft 365 application like microsoft outlook or

12:14microsoft word and then using generative ai you will be able to chat with this copilot and you can ask

12:21questions and you can generate related content with that now if you have a question that where exactly

12:26gen ai can help me in businesses well the answer is there are three levels of generative ai adoption in

12:32organizations the first one is you can use off-the-shelf generative ai apps like microsoft 365 copilot to

12:40empower your users and increase their productivity this is one of the way by which most organizations

12:46are going to use generative ai the second one is you can actually extend microsoft copilot to support

12:52custom business processes or tasks basically in this case you are going to use your own data to control

12:59how your copilot is going to respond and you can also use your own customized user prompts in the

13:05organization basically organizations can add their own data and they can control the behavior of the

13:11microsoft copilot with this kind of extension and the third and the final usage of this is you can build

13:17your own copilot like agents to integrate generative ai into a business apps or to create a unique

13:24experience for your customers so think about you are a developer who's already developing

13:29application for your clients and customers you can integrate generative ai into those custom

13:34applications using your own copilot like agent and then you can provide generative ai facilities into

13:40your own applications as well now let's talk about different microsoft copilots which are available

13:45in various microsoft products now first let's talk about microsoft copilot which is available at the url

13:51copilot.microsoft.com this is going to provide microsoft copilot's home on the web browser page

13:58you can go there you can ask questions you can generate content such as text and images

14:04next you can go for microsoft copilot which is integrated into your bing search engine while

14:10searching you can use bing.com and then you can just use a copilot which is available in the chat

14:16interface of your bing search engine when you do this thing is going to help you in getting very

14:21specific into your search results and the task which you want to achieve with that when you browse with

14:27microsoft edge browser there also copilot is available on the right top corner you can just

14:32click on that and you can get copilot pan in your microsoft edge browser this is going to help you

14:38to research a specific topic it can help you to generate a new content for example you can maybe

14:44publish a blog post using that while all of these copilots options are signing in working with the work

14:51or a school account is enabling you to use copilot in the context of your organization's data and services

14:57with all of this copilot's options you have an option to sign in with your work and school account

15:02which basically allows you to incorporate your work and school related information data with your

15:09copilots and then it can work very well with your own configurational organizational data also

15:15next we have microsoft 365 copilot this copilot can be an ai assistant for your information workers

15:22microsoft 365 copilot integrates copilot into a productivity applications that information workers

15:28use every day for example you can use copilot in microsoft word to generate a new document based on

15:35the natural language prompt you can also refine summarize and improve the document with the few prompts

15:41you can use copilot same way with microsoft powerpoint also where you can generate presentations based on

15:47the content of a document or an email you can add graphics you can reform slides or maybe you can

15:54improve your presentation with some of the edit animations with that you can also use microsoft

15:59outlook copilot where it can help you to summarize your email threads it can help you to configure your

16:06schedule and you can even find relevant emails and documents to prepare for a specific meetings

16:13basically all these tools are actually going to help you to save you a lot of time and giving you

16:18very specific insight which you are looking for a particular meeting or a document or a client

16:25next we have copilot in microsoft dynamics 365 now maybe if you have never used microsoft dynamics then you

16:31don't know how exactly this is going to work but in this case also microsoft dynamics 365 is actually

16:37a suite of business tools that is helping your users into a specific role to perform a business

16:43processes copilot in microsoft dynamics 365 is going to provide contextual assistance into these tools and

16:51is going to help users to be more efficient and effective for example let's say you are a sales

16:58professional and you can use copilot to quickly find relevant customer and industry information by

17:05integrating with the company's crm when you do this thing your crm database and some other data which

17:11you can associate with copilot can help you to get the desired data from that even in the other case

17:17let's say customer service agents can use copilot in dynamic 365 for a customer service to analyze support

17:24ticket or maybe to research similar issues or maybe to find a resolution of a specific issue which is

17:30happening on a day-to-day basis they can communicate with the end users with just a few clicks and prompts

17:36which are available in microsoft copilot if you are someone who is already comfortable with microsoft

17:42dynamics 365 then you can check out how you can use generative ai with this by following this particular link

17:48next but important one which i use on a daily basis which is copilot for microsoft azure cloud

17:55i hope you all know that azure cloud is available on portal.azure.com this azure portal is basically

18:01allowing you to create everything in azure cloud when you're using copilot with azure portal it can

18:07actually assist you in the infrastructure administration related work basically that works with all your cloud

18:13services and it can help you to do infrastructure based deployment you are going to manage

18:18your it infrastructure with the help of copilot to check and learn more about azure cloud specific

18:24copilot you can follow this particular link and remember copilot is available inside azure cloud

18:29as a separate product itself next we can use microsoft copilot for security also which will provide

18:35an assistant for security professionals as they are going to assist mitigate and respond to security

18:41threats which are happening in that the screenshot is actually showing you how we can use copilot in

18:47microsoft 365 defender when you do this thing the defender for the cloud and defender for your

18:53organization is easily able to integrate with your copilot and you can dig deep into the specific

18:58security vulnerabilities and issues with the help of copilot prompts not only that if you are someone

19:04who's going to analyze data and you want to use ai in your data analytics you can do that also

19:10your data analysts need to work with the code and visualization tools in order to analyze data and

19:16report inside in this case your microsoft fabric is going to have a copilot associated with that

19:21which enables you to analyze your data automatically and is going to generate the code that needs to

19:28analyze manipulate and visualize the data in the spark notebook so as you can see in this first screenshot

19:34we are actually having a notebook which is available inside microsoft fabric that notebook is generated

19:40with the help of python code and this code is generated with the help of copilot so basically as the data

19:45analyst you can quickly generate this kind of a notebook code all you have to do is run the notebook

19:51analyze the data see the data frames or the charts or dashboards which are generated and associated with that

19:57next if you are dealing with power bi and you want to manage power bi dashboards reports copilot can

20:05actually analyze your data and then they can suggest and create appropriate data visualizations from it

20:11and that is what which is exactly visible in the second screenshot last but not the least we have

20:15github copilot which is one of my favorite undoubtedly because this is going to help you to generate code

20:22yes github is the world's most popular place for developers to manage their code and the repositories

20:29to develop applications and to collaborate with their team members undoubtedly github has the largest community

20:35of developers associated with each other using github copilot these developers are going to maximize

20:41their productivity by analyzing and explaining code by adding code documentation with ease by generating new

20:49code based on the natural language prompt by refactoring and optimizing code by generating test cases

20:56for their existing functions and by integrating into their existing development tools like microsoft

21:02visual studio code and many more if you are interested in developing applications with the help of github

21:09copilot i strongly recommend you to check out this link this link is actually giving you step-by-step process

21:15of github copilot with a lot many tutorial labs with that now after all these variations of multiple microsoft

21:22copilots let's talk about considerations for prompt because ultimately when you are going to provide a proper

21:28prompt then only you are going to generate desired result so while some generative ai applications are

21:35going to provide buttons and some kind of a visual tools to interact with the language model most of

21:40the time you have to depend on this prompt which you are going to enter into the chatbot by typing or

21:47maybe by speaking and then if your prompt is proper then only you're going to get desired result

21:53now we know that our large language models are very smart enough to understand our inputs

21:58but there are some common prompting techniques which you can apply to get the best out of your

22:04generative ai app and some of the techniques are actually mentioned here you can see that in this

22:09slide there is one prompt which is showing you that summarize the key considerations for adopting copilot

22:15now with this there is a number one which is mentioned in this prompt which is telling you

22:20that we have just followed the first guideline which is start with the specific goal for what you want

22:26the generative ai app to do then we are going to the second one which is provide a source to

22:31ground the response in the specific scope of information where we are saying that describe

22:37in this document so basically you are mentioning that in this document we want to describe that thing

22:42then we are going with the third one add a context to maximize response appropriateness and relevance in

22:48which we are saying for a corporate executive so we are basically strictly saying that we want this

22:52thing for corporate executives only that's the context which we are specifying then we have set

22:57a clear expectation for the response in which we are seeing that format the summary as no more than six

23:04bullet points with a professional tone so basically you are specifying this fourth point that i want

23:10six bulleted points only and then finally we have an iterate based on previous prompts and responses to

23:17refine the result now this is the fifth one which we have given in the right side section where we are

23:22going to specify in the copilot so in the copilot you can actually have system message configuration

23:28conversational history which is maintained with that and your current prompt is going to be keep adding

23:33new new prompts inside that now when you're doing this thing at that time only you'll be able to understand this

23:39for you guys if you have never used copilot or if you have never used prompt engineering techniques

23:46i strongly recommend you to check out my videos which are mentioned in the description of this

23:50particular one we have a dedicated video on prompt engineering techniques and we have a separate video

23:56which is showing you how you can use microsoft copilot how you can use microsoft copilot in the edge

24:01browser i surely recommend you to check out those two videos without fail now let's talk about the last

24:08and the final topic of this particular module which is extending and developing generative ai apps now

24:14obviously llms are capable to generate a generative content based on my prompt and if i want i can do

24:21fine tuning and configurations with my own data also inside that but still there are many cases where you

24:27decide for your organization you're going to develop your own copilot kind of an agent when you decide that

24:33we want to develop a generative ai apps you have two different options you can use copilot studio or you can use

24:41azure ai foundry now this slide is actually showing you in which cases you are going to use copilot studio

24:49if you want a low code development tool for creating agents using power automate and you want to extend

24:56microsoft 365 copilot then you're going to go for copilot studio if you want fully managed hosted software

25:03as a service kind of a sas model then you will go with this you can have a dialogue and conversational

25:10orchestration so basically the flow of that particular dialogue conversation you can control

25:15you're going to have built-in analytics with security and governance control which is a part of

25:21the copilot studio integrated with that and then you can deploy to common chat channels like web apps social

25:28channels and teams and this can be associated with that now if your work is limited to this kind of

25:34features then we strongly suggest you go with the copilot studio specifically if you're not a very

25:40hardcore developer using your programming languages then you are going to use this one on the other hand

25:47if you say no i want to develop generative ai applications i'm a person who is a pro code and i can

25:53develop any kind of logical code with my programming language knowledge then azure ai foundry is your thing

26:00this is a development environment which is a pro code development platform with full catalog of models

26:06and fine-tuning capabilities remember whatever model catalog which we have mentioned in the beginning of

26:11this particular video all are available this is not a software as a service it's a platform as a service

26:18with full control over cloud infrastructure so obviously compared to copilot studio this is going to be

26:24more expensive but it's going to give you more control over whatever deployments you want to

26:29do with that your prompt and model orchestration will be there you also have an evaluation engine

26:35to taste the performance reliability scalability and responsible ai safety with content filters

26:42all these additional features which you will get in azure ai foundry portal will not be available in

26:47your microsoft copilot studio you can also deploy as an endpoint in azure for use in custom apps and services

26:55basically that's going to make your deployed model as an api and anyone who knows how to consume api

27:03can actually make a request to your deployed model for your kind information there are so many videos

27:08on azure ai foundry labs which are available on our youtube channel so i strongly recommend you to check

27:14out that one of the video which i have mentioning right now in this side is very important for you to

27:20start with azure ai foundry portal with practical labs step by step that's it for today's video i

27:26want to thank you for being with me till the end of this video and i have a request out of all this

27:32microsoft copilots can you please comment which one is your favorite in the comment box of this particular

27:37video i will make sure that we will create a separate video on that particular microsoft copilot and

27:43will tag you in that video so that everyone knows that you requested this i'll see you soon tomorrow

27:48this is maruti signing off thank you happy learning

AI-3018: Fundamentals of Generative AI – Complete Beginner's Guide!

Category

Transcript

Recommended