YT YouTube2Text

YouTube2Text → Transcripts → Orientation Session - Bharathidasan University - Google Cloud Generative AI

Orientation Session - Bharathidasan University - Google Cloud Generative AI — Transcript

by Chikka Prathibha · 9,034 words · 1,520 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

18:46Hello students.
18:50Good. Good morning.
18:59Can you all please lower your hands?
19:03I request everybody to lower your hands.
19:06[snorts]
20:19Hello students, am I audible?
20:23Can you please lower your hands? Please
20:26do not raise any hands.
24:20Hello students,
24:22good morning.
24:24Me Pratusha, I'll be the trainer for
24:27next uh six sessions
24:31throughout six weeks. Okay. So we will
24:34be learning about Google cloud
24:37generative AI. So I request everybody to
24:42uh do not raise your hand so that we'll
24:46get some idea like the people who have
24:49queries can raise your hands. Okay.
26:35Okay.
26:39>> So you are from Barati Dasan University,
26:42right? So I hope the chat is available.
26:46[clears throat]
26:47You can
26:50ask your queries in the chat box and the
26:53people who have joined from YouTube
26:55channel you can also comment your query.
26:59Okay.
27:05Okay. Uh so here in Google meet there is
27:08a limit only till 500. after 500 you can
27:13join the session from YouTube live.
27:19Okay, I hope everybody got your YouTube
27:21live uh links where you can join the
27:24YouTube YouTube live. Okay, so make sure
27:28you are attending the sessions regularly
27:30and getting the attendance. Okay.
27:36So in in this sessions you'll be
27:39learning about Google cloud generative
27:41AI where you'll be learning the
27:43generative AI modules and along with
27:46that you will also work on uh Google
27:50skills lab and also uh you will work on
27:59you will work on uh
28:03anti-gravity apps where you will be
28:05creating three apps apps using
28:06anti-gravity. Okay, you'll be learning
28:09all those things in this 12 sessions.
28:12Okay, make sure you are regular to the
28:14sessions. In case if you have missed any
28:18session, you can go through the YouTube
28:21link and learn. Okay, so hope this is
28:24clear for everybody.
28:27So before starting the session, I just
28:30want to start something important. Okay.
28:33So
28:35the first and foremost thing I wanted
28:38you to tell is about skill valid
28:40platform. This is our platform. Uh I
28:44will just share my screen.
28:49I hope everybody can see the screen. Can
28:52I get some confirmation in the chat box?
28:56Are you able to see the skill valid
28:58platform screen?
29:08Okay, thank you Yashwini. Thank you so
29:11much. Okay, so
29:15here where you will be joining the
29:19YouTube, you will be joining the
29:21session, you'll be entering to the
29:23projects, everything will be handled in
29:25the same platform. Okay. So this is the
29:28dashboard of the skill valid platform.
29:33The same way you can see in your login.
29:36Okay. The login credentials will be
29:38shared through your registered mail ids.
29:41Okay. So this is the track right. Google
29:46generative AI. Google cloud generative
29:50AI right. So
29:57yes.
29:59So and
30:01when you click on the Google cloud
30:04generative AI you can see all the
30:08instructions all the mandatory task to
30:10be completed. Okay. The first mandatory
30:13task is to enroll
30:16the Salesforce account creation. So once
30:20you log in you will be having the course
30:23access along with that a reference
30:25video. Okay. And the other one is
30:28service now account creation where you
30:30will find uh find all the steps
30:34everything in the reference video. Okay.
30:38So everything is been
30:42shown in the reference video.
30:44Accordingly you have to create a
30:47account. You have to create a service
30:49now account. Okay. Along with that you
30:52should also register future skills
30:55prime. Okay. So I will be sharing you
30:59all the links all the reference links
31:02everything in the group by end of the
31:05day. Okay. Make sure you are doing all
31:08the things all the mandatory task to get
31:10your certificate. Okay. Make sure you
31:12are doing doing it. Okay. along with
31:16future skills prime registration you
31:19should complete the anti-gravity
31:21workshop you should complete this okay
31:25so
31:33one again.
32:07So all the mandatory links everything
32:10will be shared in by your faculty or I
32:14will be sharing you in the session too.
32:17Okay. Or else if you have any access uh
32:20for skill valid right now, you can also
32:23go through the reference videos and
32:26access all the mandatory task and
32:29complete all the mandatory task
32:32that is mandatory guys to complete all
32:34the task. Okay, there are five four
32:37things to be completing complete one is
32:39trial head service now account creation
32:42fuser skills prime and Google
32:44anti-gravity workshop. So these are the
32:46five four things where you have to
32:48complete. Okay. And about the learning
32:50journey
32:53we have shared you the schedule like
32:55training calendar accordingly we will be
32:57having sessions that is Monday and
32:59Thursday. Okay.
33:02And
33:06in courses you will be all the mandatory
33:10task as I already mentioned. uh NASCOM
33:14future skills prime registrations you
33:17just click on enroll
33:21you'll be enrolling
33:24okay after enrolling you can go through
33:28the learning path and using a reference
33:31video and enroll into the STEM course
33:34okay
33:36and along with that the be the beginner
33:40lab where you'll be learning in the
33:42sessions beginner, gale and advanc.
33:45Okay, so beginner, gale and advanc along
33:49with that you also have service now
33:51account creation. So what all things we
33:54have to do as a mandatory task?
33:59Can you please tell me what all things
34:01to be covered in the mandatory task?
34:13Yes, Chikoshi, you can ask whatever
34:16question you have. You can post it in
34:18the chat box. Okay. Uh so here is my
34:21team. Diva and Turuna will be handling
34:24all your queries. Uh Tarun and Diva.
34:28Okay. Both of them will be handling the
34:30queries both in YouTube as well as
34:33Google Meet. Okay. So these are the
34:36mandatory task guys where you have to
34:39complete it for sure. Okay.
34:42Now coming to the group projects. So you
34:46are not allowed to enroll to the
34:49projects as an individual. So your
34:52faculty will be giving you one project.
34:54Your faculty will be forming a team. So
34:58everything about the projects will be
34:59handled by your faculty. Okay? Is that
35:03clear?
35:13Is that clear about f uh projects? You
35:17are not allowed to enroll to the
35:19projects. Everything will be handled by
35:21your faculty. All the team formation and
35:24project selection everything will be
35:27done from your faculty. Okay.
35:31Yes.
35:44So once you complete all the uh courses,
35:48all the assessments, all the courses in
35:51the sense these five until unless you
35:55complete these five your bar will not
35:57get increased to 100%. Once this is 100%
36:00you are eligible for the certificate.
36:03Okay. Along with that we will keep a
36:06grand assessment by by end of the
36:09sessions like on the 12th day of the
36:12session one particular day we will be
36:14keeping one grand assessment throughout
36:16the course complete syllabus. Okay. So
36:19and the final project submission if you
36:22have done only these three you are
36:24eligible for
36:26certificate until unless you do these
36:29three things you are not eligible to get
36:31the certificate. Is that clear?
36:44Okay.
37:43Okay. Is that clear about the skill
37:45valid
37:47platform?
37:48So these are the instructions.
37:52So once you open the service now account
37:56creation you can see the platform. Okay.
38:01Using the reference video using the
38:05reference video you can enroll you can
38:08create an account in service now
38:11university. Okay. And in future skills
38:14prime you have to enroll to a specific
38:17course called digital application
38:19fundamentals. It's a STEM course where
38:21you have to enroll and work on it. Okay.
38:25So, and comes with Google anti-gravity
38:28workshop where you be creating the three
38:32apps. Okay. So, these are the
38:36four things you have to mandatorily
38:39complete. Okay.
38:43So in the project workspace once you are
38:46enrolled to the project I will just show
38:48you one thing. So
38:51let's say you have enrolled to this
38:53particular project.
38:56So all these things will be done by your
38:58faculty. So this is the overview of your
39:01project uh about the description
39:03scenarios everything what are the skills
39:05required for your project everything are
39:08mentioned here for each specific
39:10project. Okay. And this is the workspace
39:13where you will be working on each and
39:15every module in the workspace. Okay. So
39:19everything will be mentioned clearly
39:22where you can go through each step and
39:24work on it.
39:26Okay. And you have to submit your demo
39:28video link along with that your GitHub
39:31link. All your project related files
39:35should be pushed into a GitHub link.
39:38make it public public repository and add
39:42your public repository GitHub link here
39:45in the platform. Okay. So once you added
39:49this you have to push this into
39:51progress.
39:53Okay. So after each task completion you
39:57have to push into progress. Once this is
40:00in progress we will review it. The
40:03mentor I will be pushing it to be
40:06reviewed. Okay, sorry. You have to push.
40:09So once the task is done, you have to
40:13push everything to be reviewed. What? So
40:16you can't simply push it until unless
40:18you submit the demo and GitHub link.
40:22Okay? Once we get the review thing, we
40:24will be reviewing and we'll be assigning
40:26you with marks.
40:28Is that clear?
40:50Hope it is clear for everybody. Uh I
40:53hope I've given you all the clear
40:55details about the course, about the
40:57instructions, group projects and the
41:00eligibility criteria to get your
41:01certificate. Okay. So, and one more main
41:05thing you have to remember is you are
41:08not eligible to enroll one specific
41:10project. So, everything will be handled
41:12by your co college faculty. Okay?
41:15Whatever things you wanted to know about
41:18the projects, you can just go contact
41:20your faculty. Okay?
41:24Is that clear for everybody?
41:33Give me a second.
48:43Everyone uh my name is T Singh. So I'm
48:46your mentor and uh there is some
48:48technical issue. So I would request
48:51everyone to uh leave the meeting and
48:53join again in 10 minutes.
51:52Uh students can you please leave the
51:54session and come back in 10 minutes.
51:57Join the session in 10 minutes again.
52:00Okay. Can you please or leave the
52:02meeting due to there is a technical
52:05issue small technical issue is
52:06happening. So for that for that reason
52:11we request everybody to rejoin the
52:14session in next 10 minutes. Okay.
55:08Hello students. Uh can you please all
55:11leave the meeting and rejoin at 11:30?
1:13:02Okay students, sorry for the
1:13:04disturbance. Uh due to some technical
1:13:06issue, we asked you to leave the
1:13:09meeting. Thank you for that and I hope
1:13:11uh everybody is back again. So if any of
1:13:16your friends are ready yet to join the
1:13:18session, please ask them to join the
1:13:19session. Uh the YouTube lines though the
1:13:23limit here in Google meet has been
1:13:25exceeded already. So ask your friends to
1:13:28join in the YouTube link. Okay. So ask
1:13:31your friends to be back to the session.
1:13:34Okay. I'll be waiting for two more
1:13:37minutes. Ask your friends to join. Okay.
1:15:36Okay. Thanks for the wait students. Uh I
1:15:40hope uh it's clear about the skill valid
1:15:44platform about the projects about the
1:15:47certification eligibility criteria.
1:15:49Everything is clear I hope. So make sure
1:15:52you are doing you completing all the
1:15:53mandatory task uh before itself.
1:15:58So about the projects we have five
1:16:00projects here. Comic graph, pocket,
1:16:03smart, fit buddy,
1:16:06uh endog journey and legal ease. So
1:16:08these are the five projects where you're
1:16:10going to work on
1:16:12this sessions. Okay. So
1:16:16make sure you are
1:16:20working on the mandatory task. All this
1:16:22project related stuff will be handled by
1:16:24your faculty. Okay. So we will I just
1:16:27wanted to start with the session. Now I
1:16:30hope about the skill valid it is very
1:16:32clear. Right.
1:16:35So in this 15 12 sessions we will be
1:16:38learning about uh Google generative AI
1:16:42track. Okay. where you will working on
1:16:44anti-gravity three app three apps along
1:16:47with that you will be working on Google
1:16:50skill boost okay so you will uh
1:16:55get a credit access
1:16:58we will be sharing you the Google uh
1:17:01form link shortly for your faculty
1:17:04faculty will be sharing with you okay so
1:17:06make sure you are doing the mandatory
1:17:08task along with that you have to attend
1:17:11the sessions regularly to maintain your
1:17:14certificate. Okay,
1:17:19is that clear?
1:17:35Give me a minute.
1:17:51So whoever are late to the session uh
1:17:55ask your friends to join in the YouTube
1:17:58link for uh the session. So here in
1:18:02Google meet there is only a limit of
1:18:05500. Okay. So ask your friends to join
1:18:08is a YouTube live. Okay. And if you have
1:18:12any queries you can post it in a chat
1:18:14box here in Google meet and in a comment
1:18:16box in YouTube link. Okay. So make sure
1:18:20you are following the things properly.
1:18:23Okay.
1:18:26So
1:18:35I hope my screen is visible for
1:18:38everybody. The
1:18:40banner image of Google Cloud Jedu AI.
1:18:49Are you able to see the screen?
1:19:01Yes.
1:19:03>> Okay.
1:19:06>> So in today like today we are starting
1:19:10our journey into one of the most growing
1:19:13areas of technology that is generative
1:19:16AI, generative artificial intelligence.
1:19:20Right outside everything is completely
1:19:23AI
1:19:25right so we will be using uh chat GPT
1:19:28Gemini and Google maps everything is
1:19:31handled by this genative AI models
1:19:34itself so most of us have already uh
1:19:37interacted
1:19:39with AI when you use Google maps and we
1:19:44watch Netflix we watch
1:19:48Amazon Prime where everything is handled
1:19:50by Google generative AI. Okay. So
1:19:55it is AI is working in the background
1:19:58for Netflix, Google maps or face ids is
1:20:02the mobiles or the people who are using
1:20:05iPhone.
1:20:10Everybody are like the people who are
1:20:12using iPhone uh they can use Siri right.
1:20:17So everything
1:20:20can be h everything is handled by AI
1:20:24which is working back as a background.
1:20:26Okay. So but generative AI is something
1:20:30different. So it doesn't only analyze
1:20:33the existing information. It can uh
1:20:37create
1:20:39completely a new content such as text,
1:20:43images, audios, videos, everything can
1:20:46be created. Okay, as a new image. So as
1:20:50a new content. So by end of this session
1:20:53you will understand what is generative
1:20:56AI is about and how it generates the
1:20:59content and the important modules behind
1:21:02the generative AI and how it is used
1:21:06responsibly everything you will
1:21:09understand by end of the session. Okay.
1:21:13So coming to
1:21:18generative AI like before going into the
1:21:20definitions or thinking about the
1:21:22technologies what we use in our daily
1:21:25life everyday task like using chat GPT
1:21:28for summarization images we use some
1:21:30nano banana or Gemini right so you
1:21:35let's say your phone recognize your face
1:21:38how that happens because AI is working
1:21:42as a background from okay
1:21:46so YouTube channel recommends videos
1:21:49Google maps predicts the traffic and
1:21:52chat GPD answers the questions and
1:21:54creates the content all these things are
1:21:56working using an AI okay as a background
1:22:00background in background for every app
1:22:04AI is running okay all of these
1:22:07technologies are connected to AI but
1:22:10they do not all work in the same Google
1:22:12map will not work as chantity chat
1:22:15cannot work as YouTube. Right? So today
1:22:19we will understand how AI evolved from
1:22:23performing specific task to generate a
1:22:26new content.
1:22:29Okay.
1:22:32So
1:22:34in today's session we will be learning
1:22:36about what is AI and difference between
1:22:39AI, ML, DL and G AI okay and how these
1:22:45generative AI models will work and what
1:22:48are the types of models what are the
1:22:50applications what are the limitations
1:22:53and how responsible AI will be there and
1:22:56along with that what is the future of AI
1:22:59so all these things we will be learning
1:23:03it today's session. So for this will be
1:23:05the agenda for today's session. Okay
1:23:09one second.
1:23:13So today's session is completely
1:23:15structured like a journey like first we
1:23:18will understand the AI at high level and
1:23:21then we will understand the relationship
1:23:23between a IML DL and generative AI.
1:23:27After that we will explore the genative
1:23:29AI things. Okay.
1:23:38So first we will learn about AI. So what
1:23:42is AI? How many of you know what is AI?
1:23:49You can answer in a chat box. What is
1:23:52AI?
1:23:55What is the definition of AI?
1:23:59artificial
1:24:09intelligence. Uh, definition of AI.
1:24:14I need a definition of AI.
1:24:25>> Okay, I will let you know. So artificial
1:24:28intelligence is not one single
1:24:31technology guys. It's a
1:24:35board field that enables machines to
1:24:38perform task that normally require human
1:24:42intelligence. Okay. It's not a single
1:24:45technology. It's a
1:24:48machine where it performs task similar
1:24:51to humans. Okay. Humans can hear, can
1:24:55see, can understand, learn, make
1:24:57decisions and create. But here AI
1:25:01systems are designed to perform some of
1:25:03these capabilities. Okay? Uh
1:25:07like to make some decisions, to create
1:25:10some content, to learn something from
1:25:13the data, everything can be done, right?
1:25:16So similar to humans.
1:25:23So think of AI as giving computer the
1:25:27ability to learn patterns from the data
1:25:30and perform intelligent task. Okay? So
1:25:33AI doesn't necessarily think exact like
1:25:38human. AI will not think exactly like
1:25:40human. Okay? So it performs intelligent
1:25:44task by using data by algorithms,
1:25:48mathematical models, computer computing
1:25:51power. So based on this AI will learn
1:25:55and predict the things. Okay. AI will
1:25:58learn the patterns and perform the task.
1:26:01Okay?
1:26:05For example, uh let's take uh like uh
1:26:10can we can a calculator be called as AI?
1:26:21>> So a normal calculator
1:26:24follows a fixed rules, right? It doesn't
1:26:26learn from the data. Therefore the basic
1:26:29calculator is generally not considered
1:26:33as an AI because it's not learning
1:26:35anything. Whatever task you ask like 2 +
1:26:372 is four that is all it will be it will
1:26:40not learn from the history or it will
1:26:42not learn from the data. You're not
1:26:44giving any data right?
1:26:49So now so this is called artificial
1:26:54intelligence. Okay.
1:26:57Did you get the point? What is
1:26:58artificial intelligence?
1:27:08>> Can you understand?
1:27:19>> Okay.
1:27:23Yes.
1:27:27>> So as I told you right what is AI? AI
1:27:30will think, see, hear and decide. Okay.
1:27:35So the main core capabilities of AI is
1:27:38like to see, hear, understand, predict,
1:27:41recommend and generate. So artificial
1:27:44intelligence enable machines to perform
1:27:46task that are generally associated with
1:27:50human intelligence. Okay. So I will go
1:27:53through each capability okay each
1:27:56capability separately. So C see C in the
1:27:59sense uh face recommen recognition
1:28:02computer vision enables AI system to
1:28:05understand images and videos. Okay.
1:28:08Whenever your face uh ID is unlocked to
1:28:13unlock your mobile you will use a face
1:28:15ID. Right.
1:28:23you will use a face ID to unlock your uh
1:28:27phone.
1:28:29So here what happening
1:28:32face recognition is happening where uh
1:28:35AI is working as a background. Okay. In
1:28:39in retail like companies you they use
1:28:43computer vision to analyze the computer
1:28:45customer movement or to monitor the
1:28:48products or to detect some missing
1:28:51instages
1:28:54everything comes under face recognition.
1:28:57Okay. Now here here in the sense we can
1:29:00take an example for voice assistant. So
1:29:03AI can process human speech and convert
1:29:06spoken languages into text or an action.
1:29:11Okay. So
1:29:13let's take uh Google Alexa. Okay. Hey
1:29:18Alexa, set an alarm for 6:00 a.m. So the
1:29:22system will understand your voice and
1:29:24perform the task. So immediately what
1:29:26Alexa will do? It will set an alarm for
1:29:296:00 a.m. Right? If you want to play
1:29:32some song,
1:29:34>> you just tell, "Hey Alexa, play so and
1:29:36so song." It will start playing, right?
1:29:39So system will understand your voice and
1:29:41perform the task. So this is what the
1:29:44voice assistants do like speech to text
1:29:48or it can be also convert calls, call
1:29:52recordings into a text format. Okay.
1:29:56like Amazon transcribe or uh speech
1:30:00recognition tools.
1:30:02Okay. Now comes the third one is
1:30:05understand.
1:30:08AI can process languages and understand
1:30:16identify meaning intent and content.
1:30:19Okay. If a user search the best laptop
1:30:23for AI development.
1:30:25So what what is the
1:30:29understanding here? You're searching
1:30:30some engine right an intelligent system
1:30:33with which under understand the user is
1:30:36looking for like what kind of work. So
1:30:39he's he's a developer right. So that is
1:30:42the reason he is looking for a best
1:30:43laptop for development purpose. So so it
1:30:47will recommend the best laptops which is
1:30:50helpful for the user to develop some
1:30:52applications.
1:30:54Okay. So in that way the understanding
1:30:58happens.
1:31:00Okay.
1:31:09Now comes predict. Predict in the sense
1:31:12predicting the traffic, predicting the
1:31:14weather. So AI will identify the
1:31:17patterns from historical data and
1:31:19predicts the possible future outcomes.
1:31:22Okay. So based on the historical data
1:31:25which is trained and it will predict
1:31:28based on that. Okay, it will predict
1:31:31based on the historical data.
1:31:36So Google maps will estimate how long it
1:31:41to reach your destination based on the
1:31:44current traffic, historical traffic,
1:31:46road cartitions and travel patterns. So
1:31:49based on these things the analysis
1:31:52happens and then it will tell by what
1:31:54time you will reach your specific
1:31:56destination. Okay. So it will analyze
1:31:59the weather, it will predict the traffic
1:32:02and all.
1:32:06Now the next one is
1:32:09recommend.
1:32:10So recommend is nothing but recommending
1:32:13something like YouTube, Netflix,
1:32:16shopping. So in Netflix if you watch a
1:32:21horror movie continuously for four to
1:32:24five times the next time whenever you
1:32:27open definitely you will be recommended
1:32:30with the horror movies only right. So
1:32:33previously watching history and genres
1:32:36what you have watched. So similar user
1:32:39preferences everything will be
1:32:43recommended based on the history. Okay.
1:32:45and generate. Generate is like
1:32:48generating the text or uh generating the
1:32:52images. So generative AI creates new
1:32:55content based on the patterns learned
1:32:58from the large data set. Okay. Let's
1:33:02take a prompt called
1:33:08let's take a prompt called uh create a
1:33:11professional email requesting project
1:33:13approval. requesting for project
1:33:15approval. So the AI can generate a
1:33:20complete email using chargi
1:33:24or cl mid journey or there are so many
1:33:27other tools where you can use it. Okay,
1:33:31is that
1:33:34clear for everybody?
1:33:38Is it clear for everybody?
1:33:53Okay.
1:33:59>> Okay. Okay, guys.
1:34:20Now we will see the difference like
1:34:24comparison between AI ML and JA it's
1:34:28kind of a family tree. Okay. So
1:34:36think of
1:34:38so on the screen you can see the image
1:34:40right it's kind of a family tree
1:34:48it's kind of a family tree okay so think
1:34:50of AI AI is a university
1:35:00uh AI is a university and machine
1:35:03learning is one of the department inside
1:35:07the
1:35:08university. Okay. Now comes the deep
1:35:11learning. Deep learning is one of the
1:35:13spec specialization inside a machine
1:35:16learning
1:35:18inside the department. Okay. So every
1:35:21department will have different
1:35:23specialization CSE and CSE A IML CSE
1:35:28data science or sim right. So
1:35:32AI is kind of a university under
1:35:35university there are multiple
1:35:37departments one of the department is
1:38:46students give me a 2 minutes time. Uh
1:38:50I'll be back. Okay.
1:40:19Okay students uh so whenever if your
1:40:23friends are not joined the session ask
1:40:25them to join through YouTube video link.
1:40:28Okay. Ask your friends to join YouTube
1:40:30video link.
1:40:34Yeah. Okay. We'll continue the session
1:40:37now.
1:40:39So, so we are discussing about the
1:40:43family tree of AI,
1:40:46MLG and generative AI. Let's think AI is
1:40:50kind of a university and a university
1:40:52there you will be having multiple
1:40:54departments, right? So each so one of
1:40:58the department is machine learning under
1:41:01department we will have multiple
1:41:03specializations. So one of the
1:41:05specialization is deep learning. Okay.
1:41:08And generative AI is an advanced
1:41:11application area that uses deep learning
1:41:14models to generate a new content. Okay.
1:41:17So
1:41:19AI is a huge one. Under AI we will have
1:41:22machine learning. Under machine learning
1:41:24we will have deep learning. Using deep
1:41:26learning models generative AI will
1:41:28create a new content. Okay. Hope uh this
1:41:33uh family tree makes you clear
1:41:36understanding. Okay. So machine learning
1:41:39will perform intelligent task where
1:41:42machine like artificial intelligence
1:41:44perform intelligent task. Okay. Machine
1:41:47learning will learns the patterns from
1:41:49the data. When it comes to deep
1:41:52learning, this deep learning uses uh
1:41:57multi-layer neural network to learn
1:42:00complex patterns. Okay. And generative
1:42:04AI will
1:42:06create a new content using learned
1:42:09patterns. Okay. This is how AI, ML, TL
1:42:14and generative AI will actually work.
1:42:17Okay. So generative AI is not a separate
1:42:20not separate from AI. It's part of a
1:42:23larger ecosystem AI. So AI is kind of an
1:42:27ocean. Okay. One of the drop is
1:42:30generative AI. Okay.
1:42:43Now we'll see about the difference
1:42:46between traditional AI and generative
1:42:48AI. So many of you might uh get a doubt
1:42:53what is the difference between gen AI
1:42:55and what is the difference between
1:42:56traditional AI? So why genai? Why not
1:42:59real AI? Right? So traditional AI
1:43:03usually work with existing information
1:43:07to classify, predict, recommend and
1:43:10identify the patterns. Okay. So where
1:43:14you where the user will be given the
1:43:18input data and then model will learn the
1:43:20input data and it will predict it. So
1:43:24using an existing data, using an
1:43:26existing information, it will predict
1:43:29the output.
1:43:31clear.
1:43:32Now comes generative AI. Here in
1:43:35generative AI everything is newly
1:43:37generated. Everything creates a new
1:43:41content. Okay. So in traditional AI
1:43:45input by input it will analyze the model
1:43:49will analyze the input and it will do
1:43:51the prediction. But when it comes to uh
1:43:54genative AI the prompt whatever the
1:43:57prompt is given by the user it will
1:43:59learn the pattern it will generate a new
1:44:02content. Okay. For example in
1:44:05traditional AI
1:44:08is this transaction fraud. So this is
1:44:11the prompt you have given. So output may
1:44:14be yes or no. But when it comes to
1:44:17generative AI, create a customerfriendly
1:44:21explanation about why transaction is
1:44:24blocked. One second.
1:45:07Okay. So,
1:45:28okay. Sorry for the dist. Yeah.
1:45:33So in traditional AI you
1:45:37we will perform task like uh spam
1:45:42deduction uh sales forecasting or this
1:45:45is diagnosis but when it comes to
1:45:48generative AI we will create text
1:45:51generation image generation music video
1:45:54code all this generated newly okay
1:45:59so generative AI doesn't simply retrieve
1:46:02give an existing answer. It generate a
1:46:05new response based on the learning
1:46:07patterns.
1:46:09Okay.
1:48:55Okay. So I hope it is clear guys. Uh if
1:49:00you have any queries you can just ask in
1:49:03Q&A section or else you can keep clear.
1:49:06Is that clear for everybody?
1:49:17If you have any queries, you can just
1:49:19post here in the Q&A.
1:49:26>> Okay.
1:49:39Yes. Now we will see what generative AI
1:49:43will actually create. Okay. So
1:49:46generative AI will create is called
1:49:49multimodel because it can work with
1:49:52multiple types of information.
1:49:56Multiple types of information. It can
1:49:59create
1:50:01text, images, audios, videos, code,
1:50:05slides, everything. Okay, let's see one
1:50:07by one. Okay, so the first first
1:50:10category what genative AI will create is
1:50:14text. Okay,
1:50:17let's take an example for text is it can
1:50:20create articles, it can create the
1:50:23reports, it can create the stories, it
1:50:25can also create emails and lesser
1:50:27graphs. Okay, when it comes to tool uh
1:50:30tools to create this text uh kind of
1:50:33content, you can use chat GPD, you can
1:50:36use Gemini, you can also use cloud,
1:50:39Microsoft copilot, all these things you
1:50:41can use to create uh text kind content.
1:50:46Okay. So when it comes to images,
1:50:50images is like it can be posters, it can
1:50:53be designs, illustrations and product
1:50:56related concepts. So for to create an
1:50:59image you can use Gemini image
1:51:01generation and uh Firefly and Mid
1:51:05Journey. There are many other AI tools
1:51:09where you can use uh
1:51:13to create an image. Leonardo all these
1:51:16things are the images where you can
1:51:18create using an AI. Okay. Now comes
1:51:21audio.
1:51:23AI generated
1:51:25audio uh voices, AI generated music, AI
1:51:30generated voice overs, everything can be
1:51:33generated using Google AI studio or 11
1:51:37labs or suno or Amazon poly. All these
1:51:41things are the tools where you can
1:51:42generate an audio. Okay. Now comes
1:51:46video.
1:51:47Video is like educational videos where
1:51:52you can create educational videos, you
1:51:54can create marketing videos, AI avatar
1:51:57videos or product demonstration videos.
1:52:00So multiple things can be done using an
1:52:03AI. So the tools kind
1:52:09all these things are the things
1:52:11similarly code. Okay, you can also
1:52:15generate a code using GitHub copilot
1:52:18chat GPD cursor cloud all these things
1:52:22and slides. Slides is like presentations
1:52:25or visual content or training materials
1:52:28using karma canva and uh gemini or
1:52:33copilot. So there are multiple things
1:52:36can be generated using generative AI.
1:52:39That is the reason it is called as
1:52:41multi-model.
1:52:44Okay. So, generative AI is powerful
1:52:47today. But why did it become popular
1:52:50only recently? Can anybody answer?
1:52:56Because we have started learning an AI
1:53:00learn using an AI tool. We are into AI
1:53:03settings, right? So that is the reason
1:53:06all these apps all the AI tools are
1:53:08coming outside. Okay. If you could
1:53:13use genative AI properly, you can create
1:53:16multiple things in a very unique way.
1:53:20Okay.
1:53:24>> Hope uh this is clear for everybody. Is
1:53:28that clear?
1:53:41Okay.
1:53:49>> Yeah. Okay.
1:53:51>> Thank you guys for your responses.
1:53:58Now we'll see the evolution of AI how it
1:54:03has been transformed how this is
1:54:06happened like started from uh rules and
1:54:11the logic
1:54:13AI is not
1:54:15AI didn't started
1:54:17today guys it's been there in it's been
1:54:21there outside
1:54:23in the society in
1:54:26uh industry from 1950.
1:54:29Okay. From so many years it was there
1:54:34it was keep on building and coming on
1:54:36the screen you can see the evaluation of
1:54:39evolution of AI. So in 1950 to60
1:54:44AI has been worked using logic and the
1:54:46rules a small foundation kind of thing.
1:54:50When comes to 1970 to 80 they have
1:54:55created expert systems slowly it's keep
1:54:58on growing right and 90s to 9 2000 the
1:55:03machine learning has been started slowly
1:55:06and then deep learning in 2010 and now
1:55:08in 2020
1:55:10and beyond it's generative AI so now
1:55:15which stage we are in which state we are
1:55:19like we are introduc deep learning or we
1:55:21are into generative AI.
1:55:26We are in which stage?
1:55:37Yes, it's geni right. We are in jai
1:55:43stage where people are started using the
1:55:46tools. people are starting creating the
1:55:48unique content outside and multiple
1:55:51things right. So now generative AI is
1:55:58ruling outside. Okay.
1:56:04So why generative AI became popular? Why
1:56:08it happened?
1:56:10So the major four reasons is because the
1:56:13data the computation
1:56:17algorithms and user interface. So
1:56:20generative AI did not suddenly appear
1:56:23from somewhere else. Okay. It the
1:56:27underlying ideas existed from many many
1:56:31years. Before slide you have been seeing
1:56:34the evolution of AI, right? So keeping
1:56:38on growing the things now we are in the
1:56:41stage of using an AI right it became
1:56:44widely accessible because multiple
1:56:47technologies improved together.
1:56:50Okay multiple technologies improved
1:56:52together.
1:56:56So data data is like modern AI models
1:57:00are trained using large amount of data
1:57:04large the data like text, images, audios
1:57:09and other data. Okay. So
1:57:14the language model will learn the
1:57:17patterns from the large collection of
1:57:19data and it will generate the content.
1:57:21when it comes to com compute uh training
1:57:25large AI model will require
1:57:29powerful and uh huge computing power. So
1:57:33this generative AI have that huge like
1:57:36fast GPUs fast working computing uh
1:57:40power to run a huge models. Okay. So
1:57:45these GPUs can perform many mathematical
1:57:49calculations parallelly. Okay. So this
1:57:52make model train much faster. That is
1:57:55one of the reason to become popular.
1:57:59Okay. And
1:58:01algorithms new algorithms made it
1:58:04possible for models to understand the
1:58:08relationships in the language and any
1:58:10other uh complex data. So this uh
1:58:15transforms
1:58:16uh understanding the relationship
1:58:19between different words in the sentence.
1:58:21So based on the transformers the
1:58:24understanding between the words and
1:58:26understanding between the data the
1:58:29content will be generated properly. So
1:58:31this is also one of the reason and other
1:58:34reason is interface.
1:58:37interface is like uh chatbased interface
1:58:42made
1:58:44AI easy for everyone. Okay, AI easy for
1:58:49everyone. Earlier users needed
1:58:53programming language but now you can
1:58:56just simply uh ask the chat GPD or
1:59:00gemini Gemini or any other AI tool ask
1:59:04to write 500 lines of code in a minutes
1:59:07it will write a code. So that will be
1:59:09based on your prompt. Okay. Or else you
1:59:12can also ask uh you can also ask uh a
1:59:17prompt like explain me generative AI
1:59:21like I am a beginner to I'm beginner
1:59:24where I wanted to learn genative AI can
1:59:27you guide me so if you ask like this it
1:59:30will definitely help you right so
1:59:34AI became smarter and easier to use so
1:59:39it became accessible to everybody. Okay.
1:59:42Easy to access, smarter. Uh you'll be
1:59:45getting very good outputs, everything is
1:59:48properly made. So that is the reason it
1:59:50is became popular. So to become popular,
1:59:53the data, compute, uh algorithms,
1:59:57interface are the b major reasons of
2:00:01becoming popular
2:00:04nowadays.
2:00:06Okay.
2:00:09Is that clear for everybody?
2:00:27>> Okay.
2:00:29Do you have any queries?
2:00:39Any queries?
2:00:51>> Okay. Okay, guys.
2:01:02>> [clears throat]
2:01:16>> Okay, guys. Am I audible?
2:01:19>> Am I audible?
2:01:29>> Okay. So,
2:01:33Somebody asked me a query to explain
2:01:38Priya. You have asked me to explain uh u
2:01:42the algorithm. Give me a minute.
2:02:54Okay. Sorry guys. So about the
2:02:57algorithm. So let's think. Uh Shakia, I
2:03:02think you are there in the meet. Are you
2:03:03in the meat?
2:03:11Shaki Priya can I get a response from
2:03:13you?
2:03:15>> Yes. Okay. So that I just wanted to
2:03:18solve your query. Algorithms in the
2:03:20sense new let's take one new algorithm
2:03:24is outside. Okay. We have to make sure
2:03:27it is possible
2:03:30for the models to understand the
2:03:33relationship between the language and
2:03:36the complex data or to understand the
2:03:39different relationship between different
2:03:41words in the sentences to create
2:03:44something
2:03:45the understanding should be there right
2:03:48between the different words. So
2:03:51algorithms like transformers let's take
2:03:54a transformer what transformer will do
2:03:56transformer [clears throat] can
2:03:57understand the relationship between
2:03:59different words in the sentences once
2:04:02the understanding is done what happens
2:04:06you will be getting a proper answer
2:04:09right so this transformer will help us
2:04:11to understand the difference between
2:04:14both the things the words in the
2:04:16sentences or the task between the next
2:04:19coming up task
2:04:21Okay. So this is how the transformer or
2:04:24the algorithms will actually work. Okay.
2:04:28Is that clearly
2:04:30clear?
2:04:39>> So basically these algorithms we use it
2:04:41for the understanding purpose. Okay. Now
2:04:45comes the next topic is
2:04:54how does AI work?
2:04:58How does it work?
2:05:03So on the screen you can see
2:05:07the training data and comes the learning
2:05:11pattern and your prompt then process the
2:05:14generate and process and generates a new
2:05:17content and after generating it will
2:05:20provide the content as a new generated
2:05:24content. Okay. So
2:05:27the generative AI will start working
2:05:30with the training data. Once the
2:05:32training data is done, the training is
2:05:35done. Then it will learns the patterns
2:05:38from the trained data. Okay. So after
2:05:41learning the patterns whenever you ask
2:05:44any kind of prompt
2:05:46whenever you ask any kind of prompt so
2:05:50it will start learning the p it will
2:05:53start analyzing the patterns and it will
2:05:56analyze the trained data and it will
2:05:59process after processing it will
2:06:02generate some new content the output.
2:06:05Okay.
2:06:06Then it will give you give the user as
2:06:09an output. The new generated content
2:06:12will be given as an output. Okay. So
2:06:15this is how the
2:06:18generative AI will actually perform.
2:06:23Okay.
2:06:26Is that
2:06:28>> clear?
2:06:38Is that clear for everybody?
2:06:51>> Okay.
2:06:52Okay, guys.
2:06:55Now we'll see generative AI model types.
2:07:01So to work generative AI the models
2:07:05should be work right work properly. So
2:07:07what kind of models will this generative
2:07:10AI have?
2:07:12So models like autoenccoders,
2:07:15variational encoder, autoenccoders,
2:07:18GANs, auto reggressive models and LLM
2:07:22large language models. So different
2:07:25generative AI models are designed for
2:07:28different types of task. So there is no
2:07:31single model that is best for every
2:07:34applications.
2:07:37Okay, I'll be going through each and
2:07:40every model. On the screen you can see
2:07:42five models. Okay, I'll be going through
2:07:44each and every model. Now first one is
2:07:47autoenccoders. So autoenccoders will
2:07:50learn efficient representations and it
2:07:54will reconstruct the data. Okay, it will
2:07:58reconstruct the data.
2:08:02When comes to vans,
2:08:06what does VAN do? It generate a new
2:08:09content, new versions of variations of
2:08:11data, new versions, new variations or
2:08:14new in a different way. It will keep on
2:08:17generating the new data, new content.
2:08:20Okay. GANs. GANs is like we use GANs
2:08:24computation between two neural networks.
2:08:28Okay. So we compute GANs to create.
2:08:33Okay. It's a competition between two
2:08:35network and produce the better output
2:08:39outputs.
2:08:40Okay. And the fourth one is auto
2:08:44reggressive models. Here auto
2:08:46reggressive models will generate content
2:08:50step by step. It predicts the content
2:08:52next step. Okay. The process here using
2:08:56auto reggressive models will be step by
2:08:58step. When it comes to lll it will
2:09:01understand and generate human like
2:09:04language, human language and human
2:09:06understanding and then it will generate
2:09:09the content.
2:09:12Okay.
2:09:15So please do not memorize only these
2:09:19names. Understand the purpose of each
2:09:23model. Every model have its own purpose.
2:09:27So based on the purpose, the content
2:09:30will be generated. Okay. So
2:09:35is that clear? So you will be learning
2:09:38each model like auto encoders and vans
2:09:42in coming slides. Okay. Is that clear
2:09:45about the types of generative AI models?
2:10:03>> Okay.
2:10:06One minute.
2:10:26Now we'll learn about one two of the
2:10:29models
2:10:31two of the generative models that is
2:10:34autoenccoders and vans. So autoenccoders
2:10:38first we will learn about autoenccoder
2:10:40and then we'll go with okay
2:10:44so autoenccoder what does it do? It
2:10:47takes the input data, it compress it
2:10:50into a proper representation and then it
2:10:54reconstructor the original data. Okay.
2:10:58The input compress and reconstruct.
2:11:02Okay. The input will be given input will
2:11:05be sent to encoder. What does encoder
2:11:09do? It will compress the representation
2:11:12and will send to the decoder. So what
2:11:14does decoder do? it will reconstruct and
2:11:17send to the output. So this is the basic
2:11:20process of order encoder. So input
2:11:23compress in input to encoder encoder
2:11:26will compress the representation and
2:11:28then the compressed representation will
2:11:30be shared with decoder. Decoder will
2:11:34reconstruct the output.
2:11:52So this is how autoenccoder will
2:11:54actually work.
2:11:56Okay. Imagine
2:11:59taking a large text and creating short
2:12:03notes. What happen? The short notes will
2:12:06contain the most important information.
2:12:08Later you can use that modes to
2:12:11reconstruct the main topics. Right? So
2:12:15these autoenccoders are used for image
2:12:19compression, no power, noise removal,
2:12:23fraud detection or anomaly detection.
2:12:26Okay. In [clears throat]
2:12:28manufacturing and an autoenccoder, you
2:12:32can learn what normal machine sensor and
2:12:35data looks like. Okay.
2:12:38When comes to VAN, VAN is nothing but
2:12:42variational autoenccoder.
2:12:44So this is not only reconstruct the
2:12:48existing information, it learns the
2:12:52distribution of data and can generate a
2:12:55new versions,
2:12:57variations of data.
2:13:00Okay. So auto encoder will reconstruct
2:13:04the existing thing that will create a
2:13:08new content in a similar way. So this is
2:13:13the difference between van and
2:13:14autoenccoder. Hope it's very clear for
2:13:17everybody.
2:13:20Is it clear?
2:13:40Okay. Okay, guys.
2:13:58One second.
2:14:04[snorts] Okay. Now
2:14:08we'll see GANs and GANs. What is GAN?
2:14:14Generative adversarial network.
2:14:17Okay. So, GAN stands for generative
2:14:20adversarial network. So, which contains
2:14:24generator and discriminator. Okay. So,
2:14:28what does generator do? Generator
2:14:30creates uh synthetic and fake data.
2:14:35Discriminator will check whether
2:14:40check whether the output is real or
2:14:42fake. Okay. Whatever data is created by
2:14:46the generator, it will be checking
2:14:49checked by
2:14:52the discriminator. Okay. So generator
2:14:57will create the data. Discriminator will
2:14:59check the output whether it's real or
2:15:02fake.
2:15:04Okay.
2:15:06So the process of this GAN will be like
2:15:10gen after creating the generator will
2:15:12create the output. So after the
2:15:15evaluation is done by the discriminator
2:15:17whether it's a fake or a real one. So
2:15:20after that evaluation
2:15:22the generator will receive a feedback.
2:15:25So based on the feedback the generator
2:15:27will improve and process will process
2:15:30the output. So this process will be
2:15:32continued. Okay. So it will continue
2:15:35it's a continuous process. So disc
2:15:38generator will create the outputs.
2:15:40Discriminator will evaluate whether it's
2:15:42real or fake based on the out uh
2:15:45feedback. The generator will improve and
2:15:48regenerate it regenerate the process. So
2:15:51this process continues many times. Okay.
2:15:57So this gan will improve through the
2:16:01computation. The next model generate the
2:16:04information sequentially. Okay. It's
2:16:07kind of a sequential process.
2:16:15>> Okay. Now, auto reggressive models.
2:16:20So, what does this auto reggressive
2:16:22model does? It will generate the output
2:16:27step by step. So, one step at a time.
2:16:32Auto regressive model will generate
2:16:34output one step at a time. Okay. After
2:16:37one step is done, the next step and the
2:16:40next step. It's kind of a step-by-step
2:16:43process. So each prediction depends on
2:16:46the information which is generated
2:16:48previously. Okay. So
2:16:52on the screen you can see the image.
2:16:56Okay. Picture. So give me a minute. See
2:17:00here it's a input so it is linked with
2:17:04the next one and this output will be
2:17:06linked to the next one. This output will
2:17:08link to the next one. So it's kind of a
2:17:11step-by-step process. It's a sequential
2:17:13process. Okay. So
2:17:18each prediction each new prediction will
2:17:20be depend on the information which is
2:17:22generated previously. Okay. So
2:17:27let's take an example like uh the input
2:17:31input will be like the sun rises in the
2:17:35okay the possible prediction should be
2:17:37east. So the first word is connected
2:17:40with the second word called sun the sun
2:17:43and the third word is connected again
2:17:45rises then fourth word then fifth word.
2:17:49So based on the process based on the
2:17:51connection between each word the
2:17:54prediction happens. Okay. The sun rises
2:17:56in the east. If there is no connection
2:17:59between this sentence what happens? Will
2:18:02you get a proper output?
2:18:17Okay. So there should be having some
2:18:21connection so that the next word will be
2:18:23predicted. If there is no connection
2:18:25between the sun rises in the will you
2:18:28get a proper output
2:18:31yes or no?
2:18:42Will you get the output if there is no
2:18:44connection?
2:19:04We will not get the output right. So we
2:19:07will not get any output if there is no
2:19:10proper connection. Okay.
2:19:13[clears throat]
2:19:18So that is about auto reggressive
2:19:21models. When comes to the LLM, the fifth
2:19:25model.
2:19:39So about the attendance I can see few
2:19:42messages related to attendance. Your
2:19:45attendance is
2:19:47shared by your faculty. Your faculty
2:19:50will uh actually your faculty is there
2:19:52in the meeting. So they will be handling
2:19:55your attendance. Okay.
2:20:17Okay. Your attendance will be handled by
2:20:19your faculty. No sheet or no Google form
2:20:22will be shared here in the chat. Your
2:20:25faculty will look after into that. Okay.
2:20:28So now we'll be learning about what is
2:20:31an LLM. LLM stands for large language
2:20:36model. So it's a large AI model.
2:20:40It's a large AI model trained on
2:20:44innumerous data, huge data of language
2:20:48data. Okay. So huge data is been trained
2:20:51for that large AI model called large
2:20:55language model. Okay. It is main
2:20:58capability is to understand language
2:21:02patterns and generate human like
2:21:04responses.
2:21:07Okay,
2:21:10human like responses. Example, Gemini,
2:21:14chip, claw and llama all these things.
2:21:25So the capability of LL is to write is
2:21:30to summarize translate answer questions
2:21:34generate some code explain the concepts
2:21:37all these kind of task can be done by an
2:21:40LLM. So every track GPD every Gemini
2:21:44tool or any other AI tool comes under
2:21:47LLM. Okay. So, LLM is a large language
2:21:52model where it understand the language
2:21:55patterns and generate the responses.
2:21:57Okay. All the AI tools are comes under
2:22:00LLM.
2:22:04Okay. LLM is not uh automatically a live
2:22:08search engine. It generate the responses
2:22:11based on the learned patterns. Okay. It
2:22:15may not have
2:22:17current information until unless it is
2:22:20connected to the search engines. Okay.
2:22:23If you want something updated data, you
2:22:25have to train the LLM first and then
2:22:28work on it. Okay. Sorry. Uh so to get a
2:22:32current data it it will not connect uh
2:22:36to the current information until unless
2:22:39it is connected to the search engines or
2:22:41retrieval systems or any other external
2:22:43tools.
2:22:45Okay.
2:22:50So I just have a small question for
2:22:52everybody.
2:22:54uh if an LLM gives an answer confidently
2:23:00LLM in the sense any AI tool gives
2:23:02answer confidently does that guarantee
2:23:05that it was a correct answer
2:23:09all the answers what you are getting in
2:23:11chat or what you are getting in Gemini
2:23:13or any other AI tool is correct answer
2:23:19yes
2:23:21oh how many
2:23:31It's a no guys. No. All the answers may
2:23:34not be correct. Okay. That is the reason
2:23:37people tell us do not trust AI blindly.
2:23:40Okay. So we should not trust AI blindly.
2:23:46Okay. So AI may make mistakes too.
2:23:53So that is the reason we should not
2:23:55trust AI blindly. Okay. So
2:24:01now let's understand how an LLM convert
2:24:06a prompt into a complete response. So to
2:24:09understand that we should learn how this
2:24:13LLM will generate
2:24:16the text. Okay. So
2:24:26There are
2:24:28so this LLM will generate the steps uh
2:24:32generate the text step by step guys.
2:24:35There are three steps on the screen you
2:24:37can see the three steps like uh write
2:24:39like ask tokenize understand and predict
2:24:43and build right. So the first step here
2:24:47is to ask in the sense prompt and
2:24:50tokenization. So individual the input is
2:24:54divided into small units called tokens.
2:24:58Okay. So every prompt is divided into
2:25:03small units called token. Okay. For
2:25:06example,
2:25:07I will be giving a prompt called AI.
2:25:10Generative AI is powerful. Why? This is
2:25:14the prompt. So the model may process the
2:25:17text as smaller pieces rather than
2:25:21treating it as one single statement. So
2:25:24it will divide like generative AI is
2:25:28powerful. So here there are four tokens
2:25:31or four units. Okay. So this is how the
2:25:34prompt is divided into tokens. Okay.
2:25:37After division, after the tokenization
2:25:39is done,
2:25:41the understanding happens. The model
2:25:44analyze the relationship
2:25:47between tokens and identify the
2:25:50patterns. Okay.
2:26:01Okay. After analysis is done, what will
2:26:04be the third step? predict and build the
2:26:08model will predict the next suitable
2:26:11token or next suitable unit or a word.
2:26:15So based on the analysis okay so
2:26:19each prompt is divided into token then
2:26:23understand then predict and build the
2:26:26output. So this is the basic pro process
2:26:30how this LLM will generate the text. Is
2:26:33that clear? may know the three steps.
2:26:37Can you please tell me the what are the
2:26:39three steps we do whenever we give a
2:26:42prompt?
2:26:53>> What are the three steps we have learned
2:26:56to process the prompt to output?
2:27:12Okay, I repeat the process again.
2:27:15Let's take a prompt called
2:27:22generative AI is powerful. Okay, or else
2:27:26I'll take in general prompt. India is a
2:27:30country
2:27:32in case
2:27:35or um
2:27:38okay
2:27:43prompt is like India is a dash okay it's
2:27:48a fill in the blank okay India is a that
2:27:51is only the prompt the next word should
2:27:53be country right how does this happen so
2:27:56first the sentence called India is a
2:28:00that will E divided into small units as
2:28:04India is one unit is is one unit A is
2:28:08one unit. So here we have three
2:28:10different tokens three units right. So
2:28:14after division happens after the
2:28:17tokenization is happen the understanding
2:28:20happens now. So understanding between
2:28:22three tokens India is A. After getting
2:28:26that understanding it start predicting
2:28:28the next word. So India is the country.
2:28:31Once the analysis is done, it will
2:28:34predict the output called country. Okay.
2:28:37So our answer will be depend on the
2:28:39previous words. Okay. Is that clear?
2:28:56>> Okay. Is that clear about the LLM?
2:29:01How this LLM will generate the things.
2:29:15>> Okay. Okay.
2:29:19So now we'll see
2:29:27we'll learn about hallogenation.
2:29:31Allergenation occurs when AI system will
2:29:34generate information very confidently
2:29:37and believe that it is correct but it is
2:29:41wrong.
2:29:43AI will generate wrong answers very
2:29:46confidently.
2:29:48that is called hallogenation. Okay. Or
2:29:52completely wrong answer or
2:29:56misinterruption
2:29:58or miscommunication, misunderstanding,
2:30:00all these things comes under
2:30:03allergenation. It it may the comment
2:30:05types may be like fact fake facts, fake
2:30:09information, incorrect numbers, wrong
2:30:12calculation, incorrect code or in
2:30:15incorrect research paper, incorrect
2:30:18answers, everything comes under
2:30:20hallucination. It will give a output
2:30:22very confidently but it is wrong. Okay.
2:30:27So that is the reason we tell do not
2:30:30trust AI blindly until unless it is
2:30:32verified.
2:30:34Okay.
2:30:36So,
2:30:40so we can verify using tools like Google
2:30:43Scholar. All these things based on the
2:30:46tools you can based on the output you
2:30:49can verify it whether it is correct or
2:30:52not. Okay. So, hallogenation is one of
2:30:56the reason why we need responsible AI.
2:30:59So we need responsible AI just to avoid
2:31:03halogenation to reduce hallogenation.
2:31:06Every AI tool must have this uh
2:31:11every AI tool must have this uh hallogen
2:31:15responsible AI thing. Okay.
2:31:20So hope it's clear for everybody.
2:31:24Allination is nothing but giving a prom
2:31:28giving output very confidently. Hope
2:31:32it's clear for everybody.
2:31:36Is it clear?
2:31:49>> Okay.
2:31:53And now
2:32:02now we'll see few applications of
2:32:06generative AI. So applications like chat
2:32:09bots, virtual assistance,
2:32:11summarizations,
2:32:13research, code generation, translation,
2:32:17automation, data analysis, content
2:32:19creation. See on the screen you can see
2:32:22various uh
2:32:27seven applications on the screen. Okay.
2:32:30So these are the seven applications of
2:32:32generative AI where you can use
2:32:34generative AI very easily. Okay. You
2:32:38chatbot is also generative AI for
2:32:40content creation. You can use generative
2:32:41AI for data analysis also you can use
2:32:44for process automation. You can use for
2:32:47translation the languages you can for
2:32:49code generation for summarization for
2:32:51all kind of purposes you can use
2:32:54generative AI okay so these are the
2:32:58basic in general applications of
2:33:01generative AI okay now we'll see the
2:33:04limitations
2:33:06and
2:33:07sorry we'll talk about responsible AI
2:33:11Okay.
2:33:20So as I already told you while we
2:33:22discussing about the hallogenation
2:33:24slide, I already used a word called
2:33:27responsible AI. Responsible AI means
2:33:31designing, deploying and using AI
2:33:35fairly.
2:33:37Okay. So fair enough, safe answers,
2:33:41being private with the information like
2:33:44do not leak the data or being
2:33:47transparent everything comes under
2:33:50responsible AI. Okay. So AI can provide
2:33:54the suggestions but humans must make
2:33:58important decisions and verify the
2:34:01results what you got from AI. Until
2:34:03unless the verification is done, please
2:34:06do not believe AI blindly. Okay.
2:34:12So,
2:34:16principles of responsible AI is like
2:34:19being fair enough. Fair in the sense AI
2:34:23should not produce unfair outputs. Okay?
2:34:27So, AI should not give biased biased
2:34:31outputs. So if your AI is trained on
2:34:34biased data definitely you will get a
2:34:37biased output. So that is the reason
2:34:39whenever you train your AI model be sure
2:34:43that you're training the fair
2:34:45data. Okay. So that is how one thing and
2:34:50the private and safe. So all the
2:34:53sensitive data must be protected. No
2:34:57personal data should be linked right
2:34:59keeping the data very secure. So that is
2:35:02the reason we should not upload any
2:35:05private data or your images or your
2:35:08email ids, contact numbers, all these
2:35:11things you should not upload or you
2:35:14should not give to AI tools. Okay. And
2:35:19being transparent
2:35:21users should know when AI is being used.
2:35:25So if AI is transparent
2:35:29that is enough. Okay. So if AI is not
2:35:34transparent user will not have any trust
2:35:37on that particular data particular AI.
2:35:41Okay. Next comes human oversight. Human
2:35:44oversight in the sense AI should support
2:35:48human decision should not replace
2:35:50human's judgment. Okay. End of the day
2:35:54human will take the decision. It only
2:35:57assists us. AI will only give us some
2:36:01assistance but it will not replace the
2:36:03humans. Okay. And evaluations.
2:36:08Evaluations is like AI will generate
2:36:10multiple outputs. So everything should
2:36:13be evaluated properly. Should check the
2:36:15accuracy. Should check whether the
2:36:18generated output is relevant or not. and
2:36:20should check it is biased or not, it is
2:36:23safety or not. Everything should be
2:36:25checked prior itself before using. Okay.
2:36:31So this is how the responsible AI will
2:36:35actually work. Okay. So today we have
2:36:40learned the foundations of generative
2:36:42AI. Okay. We have learned about what is
2:36:45AI, what is the difference between
2:36:49workflow of AI, MLDDL and genai
2:36:52everything and
2:36:55comparison between traditional AI and
2:36:57generative AI and how this generative AI
2:37:00will work. What are the important models
2:37:02in generative AI? How responsible AI
2:37:05should be there and applications
2:37:08of genative AI. So these are the few
2:37:13things these are the major foundations
2:37:16of generative AI we have learned today.
2:37:19So is that clear for everybody? Do you
2:37:22have any queries?
2:37:27If you have any queries now the session
2:37:30is open for next one minute you can ask
2:37:33your query. Okay. If it is clear, please
2:37:36mention clear in the chat box and in
2:37:40YouTube comment box.
2:37:59Okay, I think nobody have any kind of
2:38:02queries. I hope it's clear guys. Uh
2:38:05sorry for the disturbance happen in
2:38:07between the session in the beginning of
2:38:09the session uh due to some technical
2:38:11issue but still uh work on the mandatory
2:38:16task. Uh please download the skill valid
2:38:20application and uh you can access your
2:38:23daily sessions like you can join the
2:38:26sessions from the skill valid itself.
2:38:29Okay.
2:38:36Is that clear for everybody?
2:38:47Oh yes.
2:38:59>> Yeah. app is like a skill valid
2:39:01platform.
2:39:04The skill valid platform I have told you
2:39:06in the beginning right the skill valid
2:39:09platform where you have to where you can
2:39:12access your course you can access your
2:39:14projects everything okay
2:39:18yeah that is all for today's session
2:39:20guys uh hope it's clear for everybody
2:39:27so thank you so much for joining the
2:39:29session see you in the next session
2:39:31bye-bye

About this transcript

This page contains the full transcript of Orientation Session - Bharathidasan University - Google Cloud Generative AI by Chikka Prathibha, generated from the public captions YouTube serves with the video. The transcript has 9,034 words across 1,520 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.