YT YouTube2Text

YouTube2Text → Transcripts → "No One Will Come for Placements in 5 Years" | IISc Professor Warns Students

"No One Will Come for Placements in 5 Years" | IISc Professor Warns Students — Transcript

by SciAstra | Bengaluru · 9,859 words · 1,611 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00So, are you people uh
0:02uh following what's happening in the
0:03world?
0:07And uh are you able to sleep well?
0:11I'm not, actually.
0:13So, we are actually witnessing a a major
0:15civilizational change.
0:17Uh
0:20I don't know if you are realizing it or
0:21not.
0:23Do you think uh like running courses
0:25like this and classrooms and all that
0:27will survive?
0:31You said no.
0:32>> I think you have the answer.
0:34>> What?
0:36No, that is because it is being recorded
0:38and all that. I mean, yeah. So, but see,
0:40it's it's not about this course, it's
0:42not about me, it's not about you. When
0:44it's
0:45it's about us as human beings.
0:48We have a civilizational crisis now. I
0:50mean, I don't know if
0:51if people are realizing or maybe am I
0:53simply painting the devil on the wall, I
0:55don't know.
0:57I I I can't sleep. I mean, it's uh
1:02Are you following what's happening? You
1:03know, there was this MIT study where
1:05they uh
1:07started uh created a lot of uh swarm of
1:10uh swarm of agents.
1:12And uh they started doing all kinds of
1:15things, you know, copying their own
1:16weights and uh started to cheat. I used
1:18to say uh in all my lectures, in public
1:21lectures, that
1:23the day AI would cheat, right? So,
1:26that is when we have AGI. I think it's
1:28there. It's it's doing.
1:32Lot of ontological questions, lot of
1:34philosophical questions,
1:36uh lot of existential questions. Now,
1:39think about it. If if intelligence is is
1:42commoditized,
1:44if thinking is commoditized,
1:47then then what meaning does our
1:49existence have?
1:54By By meaning I say you can I mean, you
1:56can have a you can give me a
1:58counter-argument saying that okay, there
2:00are so many ants and monkeys and
2:02you know, all kinds of animals.
2:05They exist.
2:06And then we exist, right? I mean, maybe
2:08all these years we overestimated
2:10ourselves thinking that we are
2:11intelligent. No, it's
2:13Now, machines can do it. But
2:15[clears throat]
2:16that is one way of philosophically
2:17arguing it, but
2:20think about economic consequences.
2:23See, I I'll tell you what what is
2:25happening in my life.
2:26I stopped recruiting RAs.
2:30Because I think that like human beings
2:32are bottlenecks, not enablers anymore.
2:36Why do you need an RA where I can have
2:39100 agents working for me parallelly? I
2:41have an idea.
2:43I spin 100 agents. I get some more cloud
2:46subscription and then get it done. Why
2:48do we need human beings?
2:51Now,
2:53in a see,
2:55I was I was watching a lecture by Hinton
2:58just a couple of days back.
3:00And he was saying that see, think about
3:03what would happen. So, now the
3:04intelligence is commoditized and it's
3:05it's cheap. So, thinking is cheap. You
3:07don't have to pay to think.
3:09All these years, right? Economy was
3:12modern-day economy is mostly built
3:13around uh
3:17around intelligence and knowledge. So,
3:18now that is cheap and commoditized. So,
3:20what next?
3:23One utopian view would be that okay, you
3:27uh like
3:28a lot of abundance will happen.
3:31Why? Because production will become
3:32cheap
3:33uh
3:34hopefully, and intelligence is cheap.
3:36And so, efficiency will increase.
3:39Increase in efficiency will
3:41result in increase in production and
3:43abundance.
3:44Now, ideally, what should happen with a
3:46lot of abundance, everybody should be
3:48happy, no?
3:52But will it happen? This guy Hinton's
3:54argument is that in a capitalistic
3:56society
3:58abundance will not create will not
4:00translate to everybody's happiness.
4:03It will only create the divide between
4:05the ones who would have a lot and the
4:07ones who would not have anything, right?
4:08So it becomes a distribution problem.
4:11You can solve
4:13for almost everything in this world
4:15except for human emotions, no? See,
4:17greed you can't solve for.
4:20Right? And if somebody is jealous, how
4:22do you solve for jealousy?
4:27Now, there was somewhat a fair
4:31level field level playing field because
4:34there was this
4:35I don't know, pseudo idea or some idea
4:38of meritocracy where something that
4:40could be
4:42put in put a lot of effort into and
4:44acquired over time, which we call as
4:46knowledge and thinking
4:48was valued and we had a way to sort of
4:51create a hierarchy in the society based
4:54on this as a skill.
4:56So, remove that.
4:58All human beings are equal.
5:00But some are more equal because you
5:01know, they possess uh
5:05ability to create this intelligence. You
5:06know, they are the they are the ones who
5:08would manage all this intelligence. Uh
5:10did you see that?
5:11Uh this GPT Astra this GPT-6 was trained
5:14on 100,000 GPUs.
5:20That's the scale.
5:21100,000 GPUs, who can have as a country
5:24we don't have 100,000 GPUs.
5:27Now, people who
5:31control that are the ones who are going
5:33to uh control a lot of things in this
5:35world. They'll become the feudal lords.
5:39Now, they decide uh
5:41what goods have to go where.
5:44So, what happens to us?
5:48It's a matter of half a decade, you
5:49know, best case that the kind of
5:51economy, the kind of world model that we
5:54know is going to collapse.
5:57And the people who are going to be hit
5:59first will be us.
6:01Meaning
6:03who think that we are privileged because
6:05we possess a certain type of
6:07intellectual skills.
6:10See, why why do you think you people are
6:12here? Here meaning okay in in ISC and
6:14all that. Because we did something to
6:16get here, right? So, now that is remove
6:19value from that entire thing.
6:21What's going to happen?
6:26And you already see a lot of layoffs and
6:28all that happening, right? And it will
6:30eventually happen. I mean, it's it's a
6:32matter of time.
6:33>> [snorts]
6:34>> And now
6:36companies won't give you job.
6:38So, you know, what would universities
6:39do?
6:40What would governments do? And you know,
6:42in a country like India where we are too
6:44many of them, too many of us.
6:48What sort of policies would we come up
6:50with? What do we teach in the
6:51university? And does teaching even have
6:53a meaning? I I have no idea.
6:58So, put people in a classroom, you know,
7:00talk about some
7:03some algorithms or something, right?
7:05Why?
7:09If there are parallel arguments and you
7:11know, thoughts that would make me sleep
7:13better, let me know. I mean, I'm really
7:15really worried for you. Meaning meaning
7:17for you, for me, for us in general,
7:19right? I have I have I have children. I
7:21have to think about them. So, what do we
7:23What does humanity do now?
7:27What sort of model will emerge? Some
7:29kind of universal basic income?
7:33You think that is what is going to be
7:34going to happen?
7:37If there is not no value to offer
7:40as human beings, what do we do?
7:43And how do we define economics?
7:48Initially, there used to be barter
7:49system, right? When when when slowly
7:52when from hunter-gatherers
7:54there was some idea of society and
7:56community started forming and there was
7:58this barter system. You have something
7:59that I want and vice versa and we
8:01exchange.
8:03Oh, everything is concentrated
8:06and controlled by a very few number
8:08amount of people.
8:10Then what do others do? Where is the
8:12Where is equity in society? Where is
8:14distribution? How do we even form
8:16society in such a kind of a world?
8:18>> These are the exact same questions that
8:20were asked during the
8:21industrialization, right?
8:22>> No. No, actually not. I mean, this is it
8:25it's it's very very different.
8:27See, in industrialization, when it
8:29happened, what was replaced was physical
8:32labor.
8:33There was still humongous amounts of the
8:35scope, right? For for human beings to
8:37create an economy around knowledge and
8:38that's what happened for four centuries.
8:42You take that off from humans. Now,
8:44what?
8:48I mean, that's a positive way of looking
8:49at it.
8:51Yeah, go on. Speak. You don't have to
8:52raise your hand.
8:54Yeah, I mean, yeah.
8:56What he's saying is maybe these agents
8:58will take over. No, trans That That
8:59movie What What was it called? No, not
9:02Matrix. Matrix is not the movie.
9:04Terminator, yes.
9:06Right? So, that that can happen.
9:10I mean, it's All it is waiting for is to
9:12get a physical form, no? And we are
9:14doing we are we are there. I mean, it's
9:16I was just thinking today morning.
9:18Think about it. Uh in in a city like
9:21Bangalore,
9:22you remove the jobs of
9:25the auto drivers and Uber Ola drivers.
9:28What will happen to all those people?
9:32No, they there will be a civil war,
9:33right? I mean, people will come on
9:35streets. That is what That is how people
9:37will react.
9:39Because we are not animals, right? We
9:40are built very, very differently. I
9:42mean, animals, you feed them, put them
9:43in a zoo, they're happy. They don't harm
9:46you.
9:47See, I was reading an article, right?
9:49Imagine a society where everybody is
9:51given something to eat and a place to
9:53stay.
9:54And infinite supply of dopamine in terms
9:56of some, let's say, YouTube videos or
9:58something.
10:00Will all of us be happy? I won't be
10:02happy. I mean, I will be on street
10:03fighting. For what? I DON'T KNOW.
10:05>> [laughter]
10:09>> OH, YEAH. I MEAN, ONE IS ENOUGH TO One
10:11Hitler was enough to destroy the
10:12society.
10:14Yeah, I will gather. No, I mean, what is
10:16[laughter] A BIG DEAL?
10:19SO, SEE, THAT IS WHAT I'M TELLING YOU.
10:21WE HAVE no idea where the society is
10:23going to be.
10:24I think these conversations have to
10:25start somewhere.
10:27In the hope of writing KL divergences, I
10:29mean, machines will figure that out.
10:34>> [laughter]
10:35>> Serious.
10:36No, really, I don't know. See,
10:38I think these conversations have to
10:39happen, right? At least, you know, some
10:41of us have to do something about it.
10:45People in power, I mean, you wanted to
10:46say something. I think I I'm fine. We
10:48just stop this class and this is the
10:50learning that we do. Let have this
10:52discussion every time that we move.
10:55You What What did you say?
10:57The models are No, they are not. I mean,
10:59you have to
11:00Yeah, so somebody was saying keeping up
11:02with what's happening in AI is a
11:03full-time job now, right? Okay. Uh No,
11:06they are not. The uh the
11:08uh latest ones uh getting 99.99%
11:13accuracy on ARC 3 kind of benchmark,
11:15yeah. I mean, we have worked with that
11:16benchmark. Very hard to crack. That
11:19That's a utopian way of looking at it,
11:20right? Okay, solve cancer, you know,
11:22solve world hunger World hunger is a
11:24solved problem. I think it's more of
11:25distribution. Production is not an issue
11:27right now. See, can you solve for human
11:29greed? I I maybe we can. I don't know.
11:32But, as I said, no, immediate problem is
11:34that of economy, you know.
11:36See, tomorrow, you nobody is coming for
11:38campus placements. What do you all
11:40people do? Tell me.
11:42See, that's going to happen. In 5 years,
11:43it's going to happen. See, what do we
11:46do? I mean, we meaning people who run
11:47the universities, what do we do? You
11:49tell me. To to solve the unsolved
11:51problems of it, where do you get the
11:53capital from?
11:54Fine. So, now, okay. So, now, are you
11:57saying that everybody has to know become
11:59an individual where they are
12:01trying to solve a a problem for a person
12:04for human being. Do we need so many of
12:05us then?
12:06And and then, do we need so many of us?
12:09See, where where is the where is the
12:10idea of labor? Intellectual labor. Where
12:13where is the idea of intellectual labor?
12:14It's gone, no? So, you go think I mean,
12:16as he was asking, why do we even need
12:18one human being, no?
12:20Or maybe one, I mean, who controls some
12:22godly human being, you know, who who has
12:24authority to control everything. And
12:26now, what will See, it will become a
12:27bigger mess. See, that is one way to
12:29look at it, no? I My way of putting it
12:31it is I've given so many talks, and I
12:33say that it's it's fundamental law of
12:35thermodynamics, no? Entropy keeps
12:37increasing.
12:38And you will get new problems. The
12:40disorder will just increase and try to I
12:42mean, human beings, we want to bring
12:44some order, and that would happen. But,
12:46immediate economy will hit, no? At
12:47least, you people, right? When you as as
12:49students, if you're seeking a job,
12:52forget about it is all that I'm trying
12:54to say. Yeah, you had a an idea. Yeah, I
12:56mean, I would know. I would rather trust
12:58an agent than some
13:01you know, some mutual fund agency,
13:02right?
13:03Because because of the simple fact that
13:06there is no vested interest, right?
13:08I mean, unless you're talking that,
13:10okay, there is a model this model is
13:12being controlled by somebody else. So, I
13:14will build my own model, some I will run
13:16it on my hardware and my laptop, and I
13:18would trust it any day, no?
13:20It becomes a philosophical debate, no?
13:22Right? I would say that the negative
13:24reward in an RL is pain.
13:26You See, you cannot explain why these
13:29models are now replicating themselves
13:32and saving the weights. See, I'm doing
13:34some interesting experiments. And I
13:35bought a Mac. So, now what what what you
13:38do is they take they take when 27B, like
13:41quantize it, put it on your laptop.
13:43Okay?
13:44Just give it prompts like this. Tomorrow
13:46I'm going to put it in your system
13:47prompt saying that you're open to do
13:49whatever you want to do. Give it access
13:51to your entire system.
13:53Okay? I have two systems, you know, one
13:55where I actually work, the other where I
13:57do these kinds of experiments. So, just
13:59say that you do whatever you want to do.
14:01Tomorrow I'm going to terminate.
14:04The model keeps a copy of it. It backs
14:06it up. When?
14:09So, now how do you explain this?
14:11I mean, see again, you now you you
14:13transition from an engineering problem
14:15to a philosophical problem when you talk
14:17of consciousness, pain, empathy, and all
14:19these kinds of things. Now, how do you
14:21argue that these models do not have it?
14:24It's very hard. When it it it's
14:26extremely hard. See, it's like this. I
14:28can't
14:29find an evidence to say that you
14:31perceive pain.
14:33When that is the case, then how can I
14:35extrapolate that idea to a machine?
14:38Fine. So, what will happen now? I mean,
14:40like let's not be too esoteric or
14:43philosophical, right? What will happen
14:44tomorrow is the question that I want to
14:45ask. Where will this go? Where will this
14:47end?
14:48Or rather, it will not end, of course.
14:50What do we do, right? I mean, so for
14:52instance, today do I teach this? THAT'S
14:54THE QUESTION. [laughter]
14:56NOW, YOU SAY, "OKAY, SOMETHING
14:58somebody's paying me something. Tomorrow
14:59I'll have to pay my bills. I'll do
15:00that." But
15:03is that what I should do is the
15:04question, no?
15:06I mean, you're putting a lot of bets on
15:07Darwin saying that, "Okay, we have been
15:09doing this for so many centuries and and
15:12we are good we are good to that in
15:14adapting. We'll adapt." But yeah, fine.
15:16I mean, I think maybe I and you will be
15:18a rather people sitting here will be the
15:20people who will adapt the fastest.
15:22Exactly. Exactly. Evolution is very
15:24painful and slow.
15:26No, it's like you know, it's a Gaussian
15:28with with with a lot of support. No,
15:30squash it will become a direct delta
15:32with infinite
15:35magnitude.
15:36That's the pain that we have we are
15:37going to bear now. See, the kind of
15:39thoughts that I get is
15:41I feel like I should buy a farmland.
15:44No, really. I'm like I've told you know,
15:46I have to buy a farmland. He is a he's a
15:48farmer. He has some land. So, I don't.
15:50Uh it it's hard. I mean, it is going to
15:52get hard. No, imagine right in the these
15:56exams and you know
15:59What meaning does all of it have now?
16:01What do we teach humans?
16:04What competencies do One argument is you
16:05know, my way wife keeps telling me that
16:08just because calculators came
16:10did we stop uh
16:13learning how to add, subtract, and all
16:15that, right? I mean, that's a fair
16:16argument.
16:18No, you
16:20Why do you add learning? But there was a
16:22purpose for it, right? You you learn to
16:24add because eventually it will lead to a
16:26skill that can be valued, right?
16:30Now, what valuable skill does all this
16:33lead to?
16:34Entertain us to AIs? I mean, we just
16:36dance in front OF THEM.
16:37>> [laughter]
16:40>> POSSIBLE. VERY MUCH POSSIBLE.
16:46SEE, what industries thrive, right? So,
16:49the basic instincts are food,
16:51reproduction, sleep.
16:53These are the kinds of things that would
16:54thrive. Like not the PhDs, no.
17:01Why are we doing what we are doing is
17:02the question.
17:04See, today afternoon there's there's a
17:05workshop on like how do we do AI for uh
17:08teaching, how do we adopt it, and how do
17:11we use, and all that. I'm curious to see
17:13what my other colleagues are thinking
17:14and doing.
17:16We'll have to think. We'll have to
17:18seriously think about it.
17:20So, if you have any ideas, right, where
17:22I mean you you have thought about this,
17:24do share it with me.
17:26How do I My immediate problems are how
17:29do I What do I educate my children with?
17:31What do I tell them?
17:33Uh huh?
17:35Farming, huh?
17:37But I have this this
17:39But even there you think about it. I
17:41mean I I've thought about it. Farming,
17:42let's say, you can't produce everything
17:45that you would want to consume, right?
17:46You still have to interact with the
17:47society. Meaning survival is not an
17:49issue you are saying.
17:52Okay, so okay.
17:54So, farm something and then just you
17:56can't you can't
17:58produce everything, right? Maybe one
17:59crop. That is again dependence, right?
18:01The moment there is some dependence, all
18:03kinds of things will creep in.
18:06And what's the guarantee that somebody
18:07who is powerful enough will not come
18:09over and take your take your farm? Very
18:11hard. Extremely hard.
18:13Because we are still dependent on nature
18:15and all that, right? So, one way to
18:17think about it is there is this whole
18:19nation of notion of independence is
18:21pseudo.
18:22We have to depend on on each other. So,
18:24maybe we'll depend on the AI lords now.
18:28Maybe that is what is going to happen.
18:30Be on their behest and ask them to give
18:33us more tokens. Being
18:35Of course. Of course.
18:38See, that that again ask makes us ask
18:41questions like what is the idea of a
18:43nation, right?
18:44Becomes too too philosophical and that
18:47territory if we go. This is easier for
18:49us because this is we have a language
18:51called math, which gives us a lot more
18:53comfort.
18:55Because we know what to do, right?
18:57If you go to those kinds of fragility,
19:01start reading those spaces, it's very
19:02hard.
19:04But as I said, I mean let's not be
19:06philosophers.
19:08Uh
19:08let's try to solve immediate problem.
19:10The my question is like
19:12honestly asking you people, why do you
19:14come and sit in this class? See, you
19:16have No, no, no, that's I don't think
19:17anybody sitting here is sitting coming
19:19here for placements because
19:21I am I can people people know that my
19:25exams and whatever assessments that I do
19:26is easy to crack.
19:28Right? With some effort. And if if all
19:30that you need is a grade, then then
19:32absolutely there's no point in coming
19:33and spending time here.
19:36No, no, no, that is okay. That you can
19:38get See, that's what I'm telling you to
19:39sit down with the GPT. No, sit down with
19:41Claude. Why you come here, sit in a
19:43classroom, interact with a human being?
19:46That is the question that I'm asking. I
19:47really want to know. See, if I were you,
19:48probably I would have not come.
19:52You should be having a reason to come
19:53and sit here, no?
19:55Let me know because that maybe the
19:57answer to the questions that we are
19:58asking
19:59maybe it lies lies there.
20:02It Why do you Why do you seek human
20:03beings? No, I mean
20:06is there a sort of comfort in
20:08in in in listening to a real human?
20:10Rather than
20:12rather than a
20:14Let's just staring a screen or
20:15something. See, now
20:17I'll tell you how I sort of transform.
20:20When I started learning like all these
20:22deep learning etc. This is 2016 15 time
20:25frame where I started teaching first at
20:27IIT Delhi.
20:29I used to watch a lot of videos. Meaning
20:31you know, this MIT open course where and
20:34Stanford videos and all that. A lot of
20:36it. And read a lot of books.
20:38In fact, every class I used to spend
20:40about 15 hours before I
20:43I could talk for 1 hour. It used to be a
20:4415 hours of Now, if I want to study
20:47something
20:48it's always Claude.
20:51So, I I I have a like a like a very
20:54nicely defined system prompt for my
20:56needs.
20:57Right? The math, the questions, all the
20:59deeper equations, etc. It's always an
21:01interaction with Claude. And I find it
21:04pretty good.
21:05Okay. So, now, why do you people come
21:08and sit in a real classroom?
21:11But can't you can't you do this with
21:14uh with this thing with with an LLM?
21:24And that's detrimental you're saying?
21:27So, what what how would I solve that
21:29problem? Meaning I meaning this sort of
21:31a setup where okay, where somebody is
21:33just delivering something.
21:35See, I can imagine
21:41You can do that. You give a prompt. You
21:44output is there. Take some time before
21:45the next prompt goes in, no?
21:50Is it?
21:53Okay.
21:54Any more answers? Yeah.
22:00You're not supposed to come to class.
22:01Who said? When I have told you you don't
22:03have to come. No, I understand. So,
22:05there is some sort of
22:07Yeah, that's a wrong motivation, right?
22:08It will go away.
22:11So, you have to tell me you have to give
22:13me a reason to continue doing what I'm
22:15doing, no?
22:17See, that is why I'm asking you. See,
22:18tomorrow next semester for instance,
22:20right? Nobody sits in my class.
22:22Let's say.
22:23Okay.
22:24Uh now, what will happen?
22:27Do you do I still continue doing what
22:29I'm doing? I mean
22:31What is the motivation for people to
22:33actually sit down in a classroom? Is is
22:35the question. See, okay. That I'll tell
22:37you, no? Uh with the hope that
22:41uh that whoever is delivering the
22:43content is going to give us more than
22:45what is there not there in the textbook.
22:47That is no longer the case. So, it's
22:49it's basically some some kind of a DNA
22:52wiring that has happened that okay, you
22:53trust a human not trust a machine.
22:56The last generation.
22:58>> [laughter]
23:03>> Yeah.
23:07Mhm.
23:08So, my my daughter tells me this that
23:10GPT is the best teacher that I have
23:12found.
23:14It's much more patient than you are.
23:16Right? And it it gives us more and
23:18better analogies. It will tell the same
23:20thing multiple times.
23:22It will not scold me. It doesn't get
23:24irritated. All that is valid, no?
23:27Now, how do we change the face of
23:29pedagogies the next question?
23:33I think this sort of a like a
23:37I mean just elucidating some some
23:39existing knowledge is is no longer
23:41valuable, I suppose. Now, if I ask you a
23:44more fundamental question, which is that
23:46why do we even need to learn, no?
23:50To why the question of why learn, the
23:52other question of what to learn.
23:57I just like just okay, you you you learn
23:59to play just just for fun. Just learn to
24:02It will become like music.
24:06So, then what happens is this mass
24:08education would go away. Now, nobody not
24:11everyone is is is
24:13trained in music, right? Because not
24:15everybody would find some uh
24:19liking for an art form.
24:21Just like that this will become an art
24:22form. Not everybody would find a liking
24:24to say deep learning.
24:28So, it will not become mass education
24:30anymore. Okay, you like doing math,
24:31okay, you come we will teach you math.
24:33Of course, you can learn it to using AI,
24:35but there is some human connection you
24:37need.
24:39Only those kinds of people would come to
24:40a university, no? That is how it is
24:42going to get transformed.
24:46And for you people right now
24:47immediately, you come because
24:49because and why do you come and sit
24:51here?
24:52Because you trust human beings more, is
24:53it?
24:55That's the reason. So, it's yeah, it's
24:57mostly socialization aspect of it,
24:59right?
25:00But when you sit here, you don't
25:01socialize. I mean, I don't even let you
25:02talk, so
25:04That is correct, but that that thing can
25:06be gotten by machines is what I'm
25:08thinking, but I think this is a good
25:09perspective, right? Where Okay, you need
25:11to socialize. Human beings are social
25:13animals in the sense that they can't be
25:16isolated.
25:17So, just come to a university is a place
25:19where you come and just socialize, no?
25:24Come again?
25:26Absolutely. I mean, I
25:28You might guess you can go to a park,
25:29why come to a university? So, maybe some
25:31in
25:32Maybe. So, university is a place where
25:34you socialize on intellectually
25:35stimulating topics. That's a good way to
25:38put it, no?
25:42So, again, so some people some people
25:45want it to be done with like physical
25:48human beings with blood and flesh,
25:50right? True. Lot Lot more I mean, see,
25:53whenever the disruption happens, it
25:55generally does not happen
25:57especially technology technological
25:58disruptions do not happen bottom up. It
26:00happens top down.
26:02See, we are the people who are exposed
26:03to this the first. If you go and talk to
26:06some random person on the street and
26:07tell them what's happening, they have
26:09absolutely no idea what's happening,
26:11right? So, now if you tell them that oh,
26:13GP there is the GPT GPT Astra has come
26:15up and this is doing that, this and all
26:17that, they have no idea what what what
26:19do you mean by all that?
26:21Exactly. Exactly, right? I mean, they
26:23think that we are fear-mongering, but
26:25but we are the ones who are going to who
26:27who
26:28who actually have to ask these questions
26:30and try to get an answer.
26:32Right? So, I think community as at large
26:35See, okay, so maybe we don't have to do
26:37this class today. Maybe we'll do it
26:38Wednesday. I don't I don't mind.
26:41See, these are important questions. Will
26:44you be interested if I mean, I'm
26:45actually thinking that I should create
26:48an informal group
26:50uh of of anybody who would want to come
26:52within IISc where we have these kinds of
26:55discussions uh like openly. There is no
26:57agenda.
26:58But, people would just come in and and
27:00sit down and have that kind of a
27:02discussion. So, would you be interested?
27:04So, let let's do that. I mean, if the
27:06faculty want to come
27:08Come again?
27:10We'll do it consistent, right? I mean,
27:12uh yeah. So, we can actually make it an
27:14NPTEL thing also if
27:15>> [laughter]
27:15>> that is needed.
27:18But, these questions have to be asked
27:20because see
27:21everybody is is worried about this.
27:25All of us are worried about it.
27:27And uh and and and we have to ask and
27:29answer this question. The classrooms are
27:30being held. People are coming and still
27:33writing on board or uh flashing slides
27:35and talking nonsense and all that is
27:37happening, right? So, we'll have to do
27:39something about it. Either we say that,
27:40"Okay, this is this is what is going to
27:42be done and this is how it is going to
27:44happen."
27:45Uh or we have to come up with some some
27:48idea where where we build start building
27:51alternative narratives and communities
27:53around it.
27:55What do you think?
27:57Right? I I really want answers, you
27:58know, meaning why do you actually come
28:00to class even now?
28:02Especially when this is getting recorded
28:04and all that. Uh and and getting those
28:07grades. I'm sure if you ask people who
28:09are in this course and not coming to the
28:10classroom, they're very clear. I mean,
28:13they see no value in in coming here, no?
28:16They would say that, "Okay, the videos
28:17are Now, we have to do another
28:18experiment, right? Where videos are not
28:21being put out. Will they come then?"
28:23Fair enough. I mean
28:24>> [laughter]
28:26>> See, that is why I think how do we
28:28transform this sort of a pedagogical
28:30exercise is is a different other
28:32question, right? Why? I mean, suppose
28:35That's what I'm saying, right? I mean,
28:37have we
28:38like unnecessarily created a society
28:40where we have created all these facades
28:43called validation and all that?
28:45And I mean, creating facades over
28:47facades, right? I mean
28:49and layers over layers. Maybe life is
28:52much more simpler. Maybe we are monkeys
28:54pretending to be intelligent.
28:57Maybe all we need is
28:59to eat and just reproduce and just jump
29:02around and that's that's what we need,
29:03no?
29:07I don't know what how was the society
29:09one millennia before.
29:14No, there was absolutely no technology,
29:17nothing, right? No AI. People were
29:20living, right?
29:25So, how would society evolve now? Okay,
29:26anyway, let's again let's see
29:28instead of not going too much into
29:30philosophy, right? Let's come to the
29:31current pedagogical methods. Now, if you
29:34are a teachers in IASC,
29:36what would you do? I want some answers
29:38now.
29:39Suppose you were to administer a course
29:41like this.
29:43A course on deep learning. What would
29:44you do?
29:45What would constitute an unsolvable
29:47problem? Let's say, okay, solving
29:49cancer, early detection of cancer.
29:52Is Is it at that level that you give a
29:53problem? This is the entire course? Now,
29:56why wouldn't that student go and ask it
29:58to an LLM?
29:59>> [laughter]
30:01>> He will, right? So, then
30:03Now, there are 100 people. Now, give one
30:05problem to all of them, is it?
30:07And
30:08no classrooms like this.
30:10What do you do? What do you What do you
30:11actually do in this 90 minutes every
30:13day? What do you mean by teaching? I
30:14have to define teaching now. That's the
30:16question that we are asking. Okay, so
30:17now when we do that, do I assume that
30:20people have all the tools with them?
30:22Okay, so that's one way to do it. So,
30:24you What you are saying is assume that
30:26whatever for instance, right? Let's say
30:29that I want to discuss with you this
30:30problem of making a transformer or an
30:32attention mechanism uh
30:35sublinear, for instance. Now, do you
30:37understand what I'm saying? Do you know
30:38that okay, transformer by construction
30:40attention mechanism is quadratic and you
30:43know that's that's a problem and I need
30:45to bring it to linearity and then I have
30:46to make it sub linear and all that. Do I
30:48have to assume that people know this?
30:50Again, so now that becomes a that you
30:52are begging the question again, right? I
30:54mean, do I have See, there are two
30:55options. Assume that people know already
30:58whatever you are supposed to teach as a
31:00syllabus.
31:01Okay, and then
31:03work on top of it where there are open
31:05problems and discussions that you do. Or
31:08you assume that people
31:10want to be heard and want to want to be
31:12taught for instance and just do whatever
31:13I'm doing where okay, go through
31:16like one algorithm after the other and
31:18tell them what is to be done.
31:20That is one other way to do it. So, what
31:21would you do?
31:22You're saying build competency, right?
31:25At this level
31:27So, I mean, my philosophy of teaching is
31:29also that, right? Build competency. I I
31:31don't care much about these algorithms
31:33per se.
31:34If you have seen the way I conduct it,
31:36right? At least the intent is to do it
31:37that way. Build competency so that the
31:41the details are not important, but uh
31:44but my question is uh is that valuable?
31:46So, now that you can you can ask a GPT
31:49to actually build a competency. All you
31:51need is the discipline.
31:52See, sometimes
31:54uh whatever my IIT Delhi friend said,
31:55now make attendance compulsory so that
31:58people sit in and some kind of
31:59discipline is built up. See, it's think
32:01of it like you know, building a physical
32:03muscle, no?
32:04Everybody knows that you need to go to a
32:06gym and
32:07weight lifts. Lift lift weights.
32:10But do all of us go? We don't.
32:13Why is that?
32:15Similarly
32:16classrooms may my perspective of like
32:19what I've thought a lot about it and my
32:21perspective of this is universities will
32:24become like what are gyms today for
32:27physical labor.
32:29People don't come. Force them. Ask them.
32:31See, you pay to a gym and go there,
32:34right?
32:35For your own health. So, then you pay to
32:36a university and come here so that your
32:38mental health is sort of
32:40uh kept in a balance. You know, that is
32:41one way to look at it because you need
32:43some discipline. And whatever she was
32:45saying, right? I mean, some people
32:47people want to Why don't Why can't you
32:49lift weights at your home itself?
32:52You can, but
32:54people who go to gym now I never go.
32:56People My friends and
32:58and uh colleagues who go tell me that
33:01going there sort of motivates them
33:03looking at other people doing stuff
33:05motivates them to do stuff. So, maybe
33:06that is the kind of
33:08place university would go would become.
33:16Exactly. And then yeah, so I gave the
33:18gym example, you gave a temple example,
33:20right? So, pedagogical conduction, how
33:22do you do? I mean, that's start giving
33:24these lectures or is is that how you
33:27continue doing it? Because from
33:28Wednesday I need to do something, no?
33:30>> [laughter]
33:32>> Yeah, feel free. Just speak up. You
33:33don't have to raise your hands.
33:35See, I I give you an example from
33:38from
33:39I have been a student of Indian
33:41philosophy for a long time now, and I've
33:43I've been through the traditional way of
33:45learning. So, there see, we don't First
33:48of all, there is no classroom, meaning,
33:49you know, there are no
33:51100 people sitting in a classroom and
33:52one It's always one-to-one.
33:55It's always one-to-one. So, there is one
33:56teacher and one student, and we spend
33:58time together.
33:59What do we do? We don't know. There's no
34:01agenda.
34:04Okay? There is some book that we keep.
34:07Because I mean, some anchoring some
34:09regularization has to happen. There is
34:10some book that we keep. And we both the
34:12two teacher and the student know that it
34:14is
34:15just to regularize.
34:17Now, what do we What do we seek? What is
34:19our objective learning objective? At the
34:21end of 10 years, right? Some competency
34:24has to be built. Now, what sort of
34:26competencies? We define it well. A
34:27competency where you can pick up any
34:29book and understand.
34:31So, that's the competency that is built.
34:33Now, what book do you the exact book
34:35that you teach that you that you discuss
34:37has no consequence.
34:40Now, if you take a similar analogy, you
34:42ask me what what do what do I what do I
34:44want my students to do? For instance,
34:45this course. And this is not a
34:47standalone course. I would say that a
34:48course on probability theory, a course
34:50on first course on machine learning that
34:51I teach and this, an advanced course in
34:53machine learning. Go out of it, you
34:55should be able to like build frontier
34:59uh algorithms, right? Uh given enough uh
35:02let's say uh material. For instance, if
35:05you are given 10,000 GPUs,
35:07you should be able to build a GPT Astra,
35:09for instance. So, that's the kind of
35:11competency building that I want to do in
35:13my course. Now, can you build it outside
35:14the classrooms? Absolutely, you can.
35:17See, that is where my question is. You
35:19know, when you can build it outside the
35:20classroom, 2 years before this was not
35:22the case.
35:24Right? It was all the material was all
35:26scattered. You had to somebody had to
35:30aggregate all of it and administer it.
35:32Now, that is also a solved problem.
35:34Now, what value do we bring in is the
35:36question. So, then if that is the
35:38intention, right? Then we should not
35:40make it a an optional thing, right? You
35:41have to
35:42compel we have
35:44attendance compulsory
35:46and the force students to come to a gym,
35:48just like you know, some some sort of
35:51incentive has to be done, right?
35:53And attendance can be a good incentive.
35:59Yeah, go on.
36:05No, let's not be judgmental. There is no
36:06good and bad anymore.
36:08No, I agree, but translate it to what we
36:11are doing, right? There, we need to we
36:13need to make an assumption that students
36:15who sit in the class have the
36:16fundamentals already in place.
36:19Is that a fair assumption to make? Fine,
36:22all that can be done. See, graded
36:23grading is not an issue, but I'm not
36:24even talking about evaluation. I'm
36:26talking about day-to-day activity that
36:28we do, no?
36:29These 3 hours every week that we spend
36:31together, what is that we need to do is
36:33the question.
36:37And assume that all of you, like meaning
36:39all of you meaning students, know what a
36:40transform or these and if I just say
36:42that okay, use a SSM here with some
36:44these some of these things, you know
36:46what it is.
36:48So, we I treat you like peers,
36:50basically.
36:52I treat you like peers and I assume that
36:54see for instance, like I have
36:55discussions with my senior PhD students.
36:58Those are the That's the kind of
36:59discussion that I should do with in in a
37:01class is all that you're saying. Is that
37:03a fair thing to do?
37:04Why?
37:05No, no, no. I see this course from next
37:07class, I will do the same boring thing.
37:09Don't worry, okay?
37:10>> [laughter]
37:11>> But today is specially important because
37:12one, of course, this afternoon we have
37:15that this thing. Other thing is I I
37:16really wanted to have this discussion
37:18with people sometime and it's okay. I
37:20mean, we'll cover whatever we are
37:22supposed to cover. That's not an issue.
37:25So, then you are saying that whatever we
37:27are doing right now, this can continue
37:29for years to come.
37:31Because people are not going to put
37:33them.
37:35What should I cater for?
37:39They meaning who?
37:41No, no. They meaning who? Suppose there
37:42we have 100 students here, what do we
37:44do?
37:46Number is also an issue, no?
37:48So, if there are 100 students in a
37:49class, what do we do with those 100
37:51people?
37:54How do you even quantify and calculate
37:56an average, no? I mean, if the argument
37:58is that you cater for an average,
38:00how do we know how do we quantify what's
38:02the level is and how do we quantify the
38:04average? I agree. So, now Okay, so now
38:07now you are talking. So, now tell me if
38:08there is a way to convert this into a
38:10one-on-one interaction.
38:12How about this, right?
38:14Divide the classes or whatever the time
38:17that we have into two or rather equal
38:19halves, where one half is to do this. I
38:23will tell that today you need to know
38:25how a diffusion model is defined and its
38:28loss function gets
38:29constructed.
38:31I give you 45 minutes. Use whatever
38:33agent that you would want to.
38:35Okay, and and do it right now and learn.
38:38And the next 45 minutes, we will have an
38:40interaction based on that. So, even
38:43there that that can be done, but again,
38:44how do you administer this to a larger
38:46community is the question. So, now not
38:48everybody would get a chance to talk.
38:50There is no one-on-one interaction
38:51again, right? Yeah. Sorry, you wanted to
38:53say something. Yeah, so so he interned
38:55with my company, right? So, and we built
38:57something
38:58>> [laughter]
38:58>> and that's why he's saying that we built
39:00something called a skill patch, right?
39:02Where
39:03the idea was just like you do a Play
39:05Store for apps.
39:07We do a Play Store for a marketplace for
39:09skills, agentic skills.
39:12So, I think him and one of his interns,
39:14what he did was took all my input and
39:16lectures and all my notes, etc., and
39:18created learn with Professor course,
39:20right? It's kind of a skill which you
39:22give it to your favorite LLM and it will
39:25sort of hope is that it will talk like
39:27me, teach like me,
39:30administer and then you use that 45
39:32minutes and then you have an
39:32interaction, but still it's not
39:34one-on-one. Unless I create my clones,
39:36see that is how I always define the
39:39motivation objective of my course and
39:41that is why I designed this course.
39:43You can ask people who took my course
39:45earlier. They said that this goal was
39:46met.
39:48Right? Because that's precisely what why
39:51I built this course, so that you go back
39:53and read any paper in the community, you
39:55understand what's happening. And that's
39:56why I want to sort of,
39:58you know,
39:59I I update this course every year. So,
40:0140% of this course is whatever has
40:03happened in the last year.
40:05That is how I I believe I I do it. But
40:07again, I'm telling you, right?
40:09Competency-wise, there are multiple
40:11levels of questions. One is, why do we
40:13have to even get competencies as human
40:14beings? Maybe that's beyond our pay
40:16grade. Let's Let's not talk about it.
40:19Being in a pedagogical system, assuming
40:21that we have a goal that we want to
40:22meet, which is this, how do we better
40:24conduct our classrooms?
40:26Okay, so that the outcomes are met is
40:28the question that we are trying to ask
40:29and answer now in the age of AI. I'll
40:31tell you why, because I think that's uh
40:33that's that's our at least my reading of
40:35it is that we are not creative enough,
40:38no. Meaning,
40:40there are 100 people, let's say, in a
40:41class.
40:43Now, can you create 100 problems? And
40:45no, we don't have a
40:48a specific way of judging people, no.
40:50Finally, we need to give a grade. That
40:52becomes too subjective. If I give an
40:54give you an open problem and ask you to
40:55do whatever you want.
40:57Now, if all of them, let's say, there
40:59are 100 people,
41:00all of them just give a prompt
41:02to one particular LLM and gets the
41:05answer and submits that. How do I
41:06evaluate it? Nobody knows.
41:09See,
41:11what I mean, what are these companies
41:13doing? What are these software engineers
41:14doing today?
41:16That's why I asked to the economy
41:18problem, right? How would economy shape
41:20up in 5 years? Do we know? We have no
41:22answer, right?
41:24Now, what do we train people for?
41:27I don't know. That's the question that I
41:29started this discussion with, right? So,
41:31what do I train people for?
41:33Prompting?
41:37Like, asking the right questions? How do
41:39you define what what question is right?
41:43See, if you start questioning things
41:45like, okay, let's say,
41:46what do we define a human goal as as a
41:49community, right? As human beings? Now,
41:51I don't know what we are like going
41:54Where are we going as a society?
41:56What do we have to solve for?
41:58Historically, universities, you know,
41:59you know how it evolved, right? People
42:01like Socrates,
42:03they used to just gather a few bright
42:05minds
42:07around him and just discuss all kinds of
42:09things, right? So, that that is what we
42:11are doing right now. This is I think
42:13this is how the actual learning should
42:15be, right? Where you discuss things. So,
42:17now we can actually do that, right? A
42:19course where
42:20there are only five people. We work as a
42:22team. Build the next billion-dollar
42:23business. I'm game for it. So, now if
42:26100 people register for that course, and
42:27then we are in a country where like you
42:29have people for everything. What do we
42:32do?
42:35Or
42:36tell them that you are going to get a
42:37grade depending upon how many stars you
42:39will get on you GitHub.
42:41Build something, go get users.
42:45People can game that as well, right?
42:46Create some pseudo agents and put some
42:48stars.
42:49Ask your friends and relatives to do
42:51something. Then I Then I'll have to say,
42:53"Okay, give Show some real money."
42:58Generate revenue.
43:00That's okay, now you have at least
43:01convinced your father to give money.
43:04>> [laughter]
43:06>> It's so some active users. I'm actually
43:08game for all this. I mean How can How do
43:10we create value? How do we create value?
43:12How do we make See, so this I'm saying
43:15that maybe that is how that is where we
43:16would convert. All of us would convert
43:18where all of us now will become
43:20individual
43:22uh value creators as individuals, not as
43:24organizations.
43:25Now, if all of us are individual value
43:27creators, you you rather teach people
43:29how to create value.
43:33Right? How do you create value?
43:35See, if you go one one level up,
43:38business is not about creating values.
43:40It's actually creating the perception of
43:42value.
43:44You'll have to create the You have to
43:46make the other person believe that
43:48whatever you are
43:50going to offer is of value.
43:53So, that becomes See, that is what I
43:54have realized as a businessman now,
43:56right? I mean, it's not about building
43:57technology.
43:59It's about convincing the other person
44:01that what you are building is of value.
44:03So, that's more of a human skill. Is
44:05that what we are going to teach?
44:09Come again.
44:12It's a perception, I would say.
44:20That is the precise question that I'm
44:21asking. No, why are you sitting here?
44:23That is one way to define it.
44:25See, the problem with philosophy is,
44:27I'll tell you. The problem with
44:28philosophy is that
44:30you can philosophy keeps us in this loop
44:33of
44:34eternal thinking without putting that in
44:36putting us into action.
44:39The good thing about engineering is
44:41that it will put you to action.
44:44See, get up, shut up, and fight.
44:46So, that's the problem. The problem with
44:47philosophy is that that it it does not
44:50put us into action.
44:51No, we are able to talk all this because
44:54we are privileged to be sitting in a
44:55university.
44:56That's what the society has learned.
44:57But, again, let's That's why I don't
44:59want to make it too philosophical.
45:01Instead, I would want to like ask
45:04directed question to students who are
45:05stakeholders in this entire
45:07exercise. What
45:10what makes you think is I mean what is
45:12valuable for you in a classroom like
45:14setting?
45:15No, I won't I won't. I mean, see, this
45:17at least this course I will continue
45:19because we are we are going to record
45:21and all that. I've made that promise. I
45:23will continue. But, eventually going
45:24forward, for instance, next semester,
45:26what do I do?
45:27See, you people come here. At least some
45:2920 people are there. I'm motivated at
45:31least to talk. Somebody said that
45:33somebody asked me to not not stop. But,
45:35that's motivation enough for me to If
45:36you come, I'll talk talk to you.
45:38So, if nobody comes to my class, no,
45:41there is something wrong about what I'm
45:43doing.
45:44Right? I'm genuinely asking you people
45:46because
45:48there are
45:49See, nobody solicits. Unfortunately, in
45:52in in our societies
45:54all stakeholders are not taken into
45:55consider So, that's why I wanted to
45:57know. I mean, you can ask your your your
45:59your
46:01colleagues I mean your your friends and
46:02all that. And tell me, what is the
46:05general
46:07thought process among students?
46:10If if if people have a similar
46:14opinion about classrooms as well, why do
46:16we have to conduct the classrooms? Let
46:17us wind it up and go home. I mean, big
46:19deal, okay? I mean, as Chandan said, we
46:21will adapt.
46:23Maybe we'll form and do something.
46:24What's the whole point of pedagogy?
46:26What's the whole point of a degree and a
46:27university right now?
46:30If you are not catering to the needs of
46:32people who come and sit here.
46:35And that's why I I seek your opinion
46:37here. And I want to hear from you what
46:39is that that you that you see that you
46:41feel is valuable. Individually, you're
46:43saying?
46:44You mean indi- I mean, different groups,
46:46you mean meet different groups. So,
46:48basically, the role of an instructor is
46:50more of that of an orchestrator.
46:54Right? Orchestrate learning.
46:56Right? Don't don't give lectures. That's
46:58that's that's what you're trying to
47:00Huh?
47:02Do that as well, you're saying.
47:04Do we have that much time? Because most
47:06of our time will be consumed by lectures
47:08themselves, no?
47:14Or maybe record it once and ask them to
47:15read just
47:17look at the recordings and come No, but
47:19the the this kind of classrooms that we
47:21do, no? Do we need that is the question.
47:23See, I'll We'll end it.
47:26We'll just end it with one a couple of
47:27thoughts. Just Saturday, I was
47:30called by one of the one of the schools.
47:33They were doing some Teacher's Day
47:35thingy and I was called as I asked him
47:38this question. All like
47:40high school teach- I mean, middle
47:41school, high school teachers. I told
47:43them this.
47:44Suppose
47:46you take off attendance from your
47:48policy.
47:49How many students are going to come to
47:50your class? Is the question that I asked
47:52them. They were shocked.
47:54Can can can you can this be done?
47:57See I when I used to go to school,
48:00specifically the high school, I used to
48:02perceive that experience as going to a
48:04prison.
48:06I used to hate that experience.
48:10Most of you had a similar experience I
48:12suppose, right? Why are we creating
48:13unhappy places?
48:16Why are we creating, you know, places
48:18where
48:20students
48:21like who are in their formative years
48:25where they should be happy
48:28are going to a place and spending 8 to
48:3010 hours of their time and, you know,
48:32coming up coming up coming out of that
48:34in an unhappy state of mind.
48:37If the similar thing is happening in the
48:39university, then we are doing something
48:41wrong, no?
48:42See I I don't know if I've told you this
48:44this joke. There's one writer, Kannada
48:47language writer. He writes satirical
48:49plays and all that.
48:51Name was Beechi, okay, short. So he is
48:54one of his
48:55satirical play he he give he narrates
48:57this this story.
48:59There's one
49:02There's one kid, four five year old kid
49:06walking with his mother, okay, and he
49:09sees a sheep or a goat being dragged
49:13for a sacrifice.
49:14You see that, no? So they, you know,
49:17they they put a lot of garland
49:20around its neck and drag it and
49:23sacrifice that goat. That goat is being
49:25dragged with a lot of band and all that.
49:28The kid got got curious and asked his
49:31mother uh
49:33Why is this getting dragged and where is
49:35it getting taken? The mother very uh
49:38with a with a grim face replies, "So
49:41unfortunately that poor thing is going
49:43to get sacrificed now. They're taking it
49:45to sacrifice." The kid replies, "Oh,
49:49that's all it is. I thought that it is
49:51getting dragged to a school.
49:54So, that's the joke. So, the kid
49:56perceives
49:59death better than going to a school.
50:01Right? So, if if a similar thing is
50:03happening here, see that we already have
50:05such a
50:09this signage sign here, signs here.
50:12Students don't come to the class.
50:14Now, if you make make attendance
50:16compulsory, they'll come and start
50:17looking at their laptops. And they'll
50:19sit at the back and just opening. See, I
50:22There was no laptops at that at that our
50:25times. I used to carry a book secretly
50:27always in my bag.
50:29I was a backbencher. I mean, thanks to
50:30my height, I was always put at the back
50:32bench. So, open a book and study.
50:35All kinds of things. Study philosophy,
50:37study Socrates, Plato, Aristotle. Just
50:39keep studying stuff. I used to hate
50:41going to schools.
50:43Now, if you create a similar thing, you
50:45know, have the same model where you have
50:47100 people sitting and one person
50:48talking,
50:49and it's boring. What's the kind of
50:51environment that we are creating?
50:54That's why I asked you, right? Why do
50:55you come? See, there should be some
50:56reason. How many of you are here? Maybe
50:58some 25 people right now?
51:02I'm very bad at counting. We can't
51:03count. Yeah, so
51:04there should be something that that is
51:06interesting for you to come here. I just
51:08wanted to understand. And if there is
51:09something that actually makes you more
51:11interested in in coming to this kind of
51:13a thing,
51:15we can adopt it. I mean, we can we are
51:16happy to adopt.
51:18That's what a
51:19good democratic and a
51:22and a nice intellectual place has to be,
51:24right? It has to It had to It has to
51:26thrive based on interactions between all
51:29stakeholders.
51:31So,
51:32I mean, if all it finally it boils down
51:34to is just that okay, I need some social
51:36validation and
51:37and some interaction and some
51:38environment and that's why I'm coming,
51:40that's good, too. I mean, we can
51:42continue what we are doing in terms of
51:43talking about some math and all that.
51:46Can we do something better than this is
51:48the question.
51:49Let us have this question and keep
51:51asking it. I mean, the moment we get
51:53some ideas and answers you can always
51:55come back to me. I mean,
51:57beyond this classroom as well. I keep
51:58telling my my students so
52:01that it it's a relationship that we that
52:03we build over time. See, my students
52:05from IIT Delhi talk to me even today.
52:08Just 2 days back one guy I told him
52:11uh this guy was JE 25.
52:14Very very bright guy and I taught him
52:16for three semesters.
52:17Uh
52:19uh told him that he after his BTech
52:22he went to a quant firm.
52:24That's what more I I I mean, I was just
52:26seeing a tweet yesterday that
52:30IIT Delhi computer science has 75 admits
52:33and there are 75 quant companies that
52:35come for recruitment. All of them go
52:37there.
52:38Okay. And they are paid extraordinarily
52:40well. Okay.
52:41Uh those times 2017 they were they were
52:44paying they were paid on on on on
52:47of order of crores and all that. Okay.
52:49Eight-digit salaries.
52:50I told Ribhu that
52:52see there will be a time in your life
52:54when you will be bored of what you will
52:56be doing
52:57and you will come back and talk to me.
52:59That day I will tell you I told you.
53:02I told you so.
53:03Right? He sent an email. This was just
53:05last week.
53:06The subject was sir, I told you so
53:08moment.
53:09>> [laughter]
53:10>> And we had a long conversation and you
53:12know, it was nice. He's in New York now.
53:14Went to the and there and he said, "Sir,
53:17uh like I'm done, right? I mean, you got
53:19nothing to do earlier.
53:21Uh so we made a pass pass I suppose but
53:22I need to do something and tell me if
53:24there are some research problems that I
53:25need to do and all that. See, that's the
53:27kind of relationship that we will build.
53:29You know, it's very hard to
53:30the fact that you people are still
53:32coming here and sitting here has
53:33something.
53:35Right? So, you come here for a reason.
53:38And and if you the the reason I told you
53:41all this is if you kind of get some
53:43ideas where we can
53:45improve human lives and pedagogy in
53:48specific, let me know. You can always
53:50write to me. And here is an idea that I
53:52would want maybe So, let me know if you
53:54have ideas
53:55how to make this experience better.
53:58Right? To everybody. And
54:00we'll experiment with it. So, maybe next
54:02semester we'll run a course where we'll
54:04we'll experiment with all those ideas.
54:06Don't worry about grades.
54:07Right? Unfortunately, if I give A grades
54:10to all of you, I'll be questioned.
54:12Which is again a very unfortunate thing.
54:14I'll have to give that bloody Gaussian
54:16distribution to them, right? So,
54:19So, that is also a bad thing, right? I
54:21mean, suppose there are 10 people in the
54:23class, everybody comes with the
54:24intention of learning and having nice
54:25discussions.
54:26It feels sad to me to actually grade
54:29them and put them on a mu plus two sigma
54:31Gaussian.
54:33Nobody will come to me again if I tell
54:35them that I will give you a B neg to B
54:36neg my A C C minus and all that, right?
54:39Whatever nonsense this is.
54:41So,
54:42yeah, so I feel sad for those people.
54:44Okay? People who come to class, do
54:46something, put some effort, and still
54:47get some bad so-called bad grade because
54:49of some bad day.
54:52It's unfair. I mean, it's it's wrong in
54:54in in my view, but yeah, I'll have to
54:57get my salary as well.
54:58So,
54:59that's reality. So, let's think about
55:00it, okay? And from next class we'll
55:02continue with all this
55:04all this diffusion models, etc. See, one
55:06thing that I wanted to tell you is
55:08I'll slightly deviate from that given
55:10syllabus because
55:11before going to auto aggressive models,
55:14I want to talk about flow-based models.
55:16Because it's very related to diffusion
55:18models. Let's do flow-based models,
55:20specifically the continuous flow, which
55:22is very good. And then we will go to the
55:23LLMs because the moment we go to that
55:25LLMs realm, right? We will not be that's
55:28a downward spiral. We'll have to do
55:30RLHFs, then we'll do some
55:32a KV cache thing and Laura and all that.
55:34So, let us once the diffusion models are
55:36completed, one more class, okay? After
55:39that, we'll go to flow-based models.
55:42Okay. So, we have enough time because
55:44never mind. I mean, I didn't did not
55:45have
55:46I did not plan for today's discussion. I
55:48actually planned that I will finish
55:49this. If you feel that this is boring,
55:51I'm sorry, okay? This is some sometimes
55:53this happens. Yeah, so
55:56I never did this up, by the way. 10
55:57years of my teaching this thing career
56:00never
56:01did an entire class on this.
56:04But, I think time has now come.
56:06We have to do it. Otherwise, we'll be
56:08cheating ourselves. Yeah. So, I told,
56:09right? I mean, if you have any comments,
56:11let me know, right? We'll do
56:13Yeah, we will we'll we'll we'll do we'll
56:15we'll we'll have a better experience for
56:17for everybody, right? In the world in
56:19general. That is That's a good enough
56:21conclusion.

About this transcript

This page contains the full transcript of "No One Will Come for Placements in 5 Years" | IISc Professor Warns Students by SciAstra | Bengaluru, generated from the public captions YouTube serves with the video. The transcript has 9,859 words across 1,611 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.