# Day 47 — YouTube 自动生成字幕（AUTO_GENERATED）

覆盖训练段 000–10 分钟。已去滚动字幕重叠。ASR 生成可能含识别错误。

==================================================

[Music] [Music] hello and welcome to another episode of our own devices uh this time we going to talk about a very interesting concept uh which is on how to use generative AI as a digital public good and and in a country like India that sort of opens up a lot of opportunities to talk about it is a very familiar face in the Indian tech scene Mr jpr bindra who is now the founder of tech Whisperer uh so J prit welcome to the welcome to the show

thank you very much nandu great to be here and I look forward to our conversation and and for the benefit of the listeners that introduction was very short and and you know that's also because you know you can you can introduce J in many different ways and it can be a very long introduction so so I I guess we'll go directly into the point so so just prit you know generative AI is being you know is being thought of in many different scenarios but but your

concept of using it primarily you know in a country like India as as something that you know that is transformational right like you know so so if you could just explain that concept a little bit no thanks nandu and this is this is like my big passion and my big um area of of of evangelism in a sense right now as to how in India and can create a model for generative AI to lead to lead the world uh really comes from three places one is

you know as generative a and have taken hold you know us and China have clearly emerged as the two big powers in it and they have their own models you know to to uh to propagate it uh secondly you know India has obviously had a very successful story in digital public infrastructure being treated as a digital public good so things like UPI Adar built at scale for more than a billion people offered to you free has completely transformed uh uh the way we do pretty much

everything it's the digital transformation of a country and and then the third bit is I was my last two years I was doing a masters in uh Ai and ethics at Cambridge University and again we were grappling with how countries like India can have their own kinds of ethical rules guidelines based is our culture and our society on on AI and a combination of these three gave um me and a few of us this uh idea that we should look at generative AI in India or

geni is what I call Ji okay so Jan is obviously for janta people and Jan AI really is very simple yet audacious that offer generative AI to 1.4 billion Indians as a digital public good the same way that we offered payments identity Commerce etc etc and transformed them and what it entails really are two things and I'll keep that at a high level one build your own India bat llm large language model which many people are speaking about with your own context data culture language Etc

but importantly take this llm and in some way create it as a layer in the India stack which has given us all these digital public INF structure and goods and offer this to Indians to Kickstart the whole creative economy entrepreneurship and multiple other things that's the J concept it is not that easy also right because you have to literally do it ground up because whatever you have with the CH gpts and of the world is also heavily biased towards their context and we need to create

something which is positively biased to our context our problems understands the you know it's also very complex India itself is so complex right so so how do you think that goes about is it is it going to be a government driven kind of a thing or because you have so many Indian companies already working in the space do they all come together sort of contribute to something like well Big transformational Ideas are never simple uh when we when India created the digital public infrastructure you know

and which we are so familiar with today it was a massive huge complex exercise over many years with lots of money being you know lots of in investment going in ETC uh the good news is that we know how to do it we've done this successfully and so we know how to do it now what does it take really four things okay number one is that you need uh data and that is probably the most important part and and you know think about it this way

uh if you take all dur dasan recordings till now take all all India recordings till now in all the languages take all BBC Hindi Etc recordings till now take every every newspaper publication published in India etc etc most of these are available in a digitized manner and you know you will actually have the data sets available okay along with other ones which are there like bashini which is a language data set available already so in many ways while you know it's it's still difficult to get

all the data together a lot of it is available that's number one second is the is the compute power the you know the you know the the investment and compute that will go into it with all the gpus which we keep on hearing about well already as you said few private sector companies have signed up with n viia u Mr Chandra shakar is talking about getting 25,000 gpus into India for you know a test bed kind of thing and so you know and and frankly what

is going to happen to GPU costs is that a year from now probably those costs are going to be half of what they are today two years from now they're probably going to be 10% of what they are today so it isn't as big a cost that we think third is Talent which we know we have so I'm not going to dwell on that too much and the fourth most important nandu is the partnership and the governance structure this has to be much like BPI has

to be a government private sector and Academia partnership which got created and then we had the right governance bodies like npci uid IC Etc which kind of then made sure that this worked in a great public private academic partnership manner we don't have to reinvent the wheel these governance structures already exist okay and we just need to use that uh knowledge and create perhaps another governance structure which will do this and add on to the uh India stack and offer Jan as a digital public good

A lot of the structure as you said is already there we just need to start using that for something new for something new right that that's right well it might be simplifying it too much in that sense but what I'm trying to say is that you know we are not we're not a country where we have not done this before in fact we the only country only country in the world which has done this before okay we know how to create public infrastructure at scale and

offer it with the right governance as a digital public good and so for us to do something like this is much much easier than for any other country to do it and then I also spoke about the data and the talent and the other missing pieces of it in that sense and uh therefore you know creating a body which does this obviously under the sponsorship of the government much like uh you know Mr nil did and his team did the entire uh area I think is

is much easier to do for us so so this collective intelligence sort to you know of India that as you said like you know you know millions of hours of TV audio all of that but but it also comes with the added layer of complexity of language right and it's not in one language you know as India is so is that a good thing because you know in a way this entire gen AI space could get supercharged because of this new complexity that's coming in it's

solving a big problem and maybe solving it for the world oh absolutely so so the complexity is a problem and an opportunity yeah now again let me give you and the D dasan all India radio thing I said were just couple of the massive other small example but let me give you another small example uh land in every part of India is measured in a different way it's not standard there are bigas and there are Kels and there are you know Mara yeah and so you

know if you have like a chat GPT which is trained on a you know largely Western database you know it's useless for for most Indians or Indian you know let's say the Indian farmer or whatever and so there are these complexities okay which uh are there and then there's obviously language uh you know 25 offic almost official languages hundreds of thousand and thousands of dialects and it's going to take time to kind of you know bring them in ETC but once you have that think about

the benefits and let me just throw again three quick benefits at you okay yeah uh you you might have been exposed to an app called jugal Bundi you know which was uh well which has been created by Microsoft thoughtworks and a couple of other partners on generative AI which has actually Satya nadela himself kind of put it out on LinkedIn ET because he was filled with it it basically is an a generative AI based app uh which can tell any poor underprivileged Indian about the schemes

available for her you know if she's a 80-year-old Widow living in biar somewhere you know she doesn't know there are 300 500 government schemes so in her natural language she can ask an app I'm so and and what is available for me and you know it kind of then you then you interact back in that Lang language etc etc now that has been big but think of

# Day 47 — 火山引擎 录音文件识别（volc.seedasr.auc）

豆包大模型 ASR，异步接口。可能含识别错误，仅供听力对照。

==================================================

Hello and welcome to another episode of our own devices.

This time we are going to talk about a very interesting concept, which is on how to use generative AI as a digital public good.

And in a country like India, that sort of opens up a lot of opportunities to talk about.

It is a very familiar face in the Indian tech scene, Mr Jaspreet Bindra, who is now the founder of Tech Whisperer.

So, Jaspreet, welcome to the show.

Thank you very much, Nandhu.

Great to be here.

And I look forward to our conversation and, and for the benefit of the listeners, that introduction was very short.

And, and you know, that's also because, you know, you can, you can introduce Jaspreet in many different ways and it can be a very long introduction.

So I, I guess we'll go directly into the point.

So, so just please, you know, generative AI is being, you know, is being thought of in many different scenarios, but, but your concept of using it primarily, you know, in a country like India as, as something that, you know, that is transformational.

Right like, you know, so.

So if you could just explain that concept a little bit.

No, thanks, Nandhu.

And this is, this is like my big passion and my big area of evangelism in a sense right now as to how India and can create a model for generative AI to lead, to lead the world.

It really comes from three places.

One is, you know, as generative AI and AI have taken hold.

You know, US and China have clearly emerged as the two big powers in it and they have their own models, you know, to, to, uh, to propagate it.

Uh, secondly, you know, uh, uh, India has obviously had a very successful story in digital public infrastructure being treated as a digital public good.

So things like UPI Adhar, built at scale for more than a billion people, offered to you free, has completely transformed, uh, uh, the way we do pretty much everything.

It's the digital transformation of a country and and then the third bit is I was in my last two years, I was doing a Masters in, uh, AI and ethics at Cambridge University.

And again, we were grappling with how countries like India can have their own kinds of ethical rules, guidelines.

This is our culture and our society on.

On AI and a combination of these three gave me and a few of us this idea that we should look at generative AI in India, or Gen AI is what I call Jan AI.

Okay, so Jan is obviously for Janta people.

And Jan AI really is very simple yet audacious that offer generative AI to 1.4 billion Indians as a digital public good, the same way that we offered payments, identity, commerce, etcetera, etcetera and transformed them.

And what it entails really are two things, and I'll keep that at a high level.

1, build your own India Bharat LLM large language model, which many people are speaking about, with your own context, data, culture, language, etc.

But importantly, take this LLM and in some way create it as a layer in the India stack, which has given us all these digital public infrastructure and goods and offer this to Indians to kick start the whole creative economy.

Entrepreneurship and multiple other things.

That's the Janai concept.

Number.

It is not that easy also, right, because you have to literally do it ground up because whatever you have with the chat GPTs and of the world is also heavily biased towards their context.

And we need to create something which is positively biased to our context, our problems, understands the, you know, it's also very complex.

India itself is so complex, right?

So, so how do you think that goes about?

Is it, is it going to be a government driven kind of a thing or because you have so many Indian companies already working in the space, do they all come together, sort of contribute to something like, well, big transformational ideas are never simple.

Uh, when we when India created the digital public infrastructure, you know, and which we are so familiar with today, it was a massive, huge, complex exercise over many years with lots of money being, you know, lots of investment going in, etc. The good news is that we know how.

To do it.

We've done this successfully, and so we know how to do it.

Now what does it take?

Really, four things.

Okay?

Number one is that you need data and that is probably the most important part.

And, and you know, think about it this way.

Uh, if you take all Buddhist recordings till now, take all, all India recordings till now, in all the languages, take all, uh, BBC, Hindi, etcetera, recordings till now, take every newspaper, my publication published in India, etcetera, etcetera.

Most of these are available in a digitized manner and you know, you will actually have the data sets available, okay, along with other ones which are there like Hashini, which is a language data set.

Yeah, available already.

So in many ways while, you know, it's it's still difficult to get all the data together, a lot of it is available.

That's No..

1.

Second is the, is the compute power, the, you know, the, the investment and compute that will go into it with all the GPUs, which we keep on hearing about.

Well, already, as you've said, a few private sector companies have signed up with Nvidia.

Mr Chandra Shekar is talking about getting 25,000 GPUs into India for, you know, a test bed kind of thing.

And so, you know, and, and frankly, what is going to happen to GPU costs is that a year from now, probably those costs are going to be half of what they are today.

Two years from now, they're probably gonna be 10% of what they are today.

So it isn't as big a cost that we think.

Third is talent, which we know we have.

So I'm not gonna dwell on that too much.

And the fourth most important Nando, is the partnership and the governance structure.

This has to be much like DPI, has to be a government, private sector and academia partnership which got created and then we had the right governance bodies like NPCI, UIDAI, ONDC, etc.

Which kind of then made sure that this worked in a great public, private, academic partnership manner.

We don't have to reinvent the wheel.

These governance structures already exist, okay?

And we just need to use that, uh, knowledge and create perhaps another governance structure which will do this and add on to the, uh, India stack and offer Janai as a digital public good.

A lot of the structure, as you said, is already there.

We just need to start using that for something new, right?

That.

That's right.

Well, it might be simplifying it too much in that sense, but what I'm trying to say is that, you know, we are not, we are not an a country where we have not done this before.

In fact, we are the only country in the world which has done this before, okay?

We know how to create public infrastructure at scale and offer it with the right governance as a digital public road.

And so for us to do something like this is much, much easier than for any other country to do it.

And then I also spoke about the data and the talent and the other missing pieces of it in that sense.

And therefore, you know, creating a body which does this obviously under the sponsorship of the government, much like, you know, Mr Nilakani did and his team did the entire DBS area, I think is, is much so.

So this collective intelligence sort of, uh, you know, of India that, as you said, like, you know, uh, you know, millions of hours of TV, audio, all of that.

But, but it also comes with the added layer of complexity of language, right?

And it's not in one language, you know, as India is.

So is that a good thing?

Because, you know, in a way, this entire gen AI space could get supercharged because of this new complexity that's coming in.

It's solving a big problem and maybe solving it for the world.

Oh, absolutely.

So.

So the complexity is a problem and an opportunity.

Okay.

Now, again, let me give you, and the, and the, and the, and the, and the, and the, and the, and the, and the and the and the and the and the and the and the and the and the and the and the and the and the and the and the and the and the and the. Yeah.

And so, you know, if you have like a chat GPT which is trained on a, you know, largely western, uh, uh, database, you know, it's useless for most Indians or Indian, yeah.

You know, let's say the Indian farmer or whatever.

And so there are these complexities, okay, which, uh, are there.

And then there's obviously language, uh, you know, 25 official, almost official languages, hundreds of thousand and thousands of dialects.

And it's going to take time to kind of, you know, bring them in, etc.

But once you have that, think about the benefits.

And let me just throw again, three quick benefits at you.

You, you might have been exposed to an app called jugalbandi, you know, which was, well, which has been created by Microsoft Thoughtworks and a couple of other partners on generative AI, which is actually Satya Nadella himself kind of put it out on LinkedIn, etcetera.

Because he was thrilled with it.

It basically is an a generative AI based app, uh, which can tell any poor underprivileged Indian about the schemes available for her, you know, if she's a 80 year old widow living in Bihar somewhere, you know, she doesn't know there are 300, 500 government schemes.

So in her natural language, she can ask an app, I'm so and so and so merrily, yeah, have what is available for me.

And, you know, it kind of then you interact. Yeah, yeah, it's back in that language, etc, etc.