00:00You have raised $53 million in five weeks for Hangten
00:05on the pitch that AI can replace a 30-person engineering team
00:09with two to four people,
00:11and that the build part has become close to zero marginal cost.
00:18That's the disruption story investors want to hear.
00:21Does it worry you that the same week you are selling that story,
00:26the men building the underlying models are warning it
00:30that it could end humanity?
00:35No, no, no.
00:37These are two entirely different things.
00:39The models, like if they were all development in AI,
00:42not just slowed down but stopped completely,
00:45what has happened already can produce enormous value.
00:51See, one of the great applications of this wave of AI,
00:55generative AI, is software development.
00:58And partly, for various reasons,
01:00partly because there is so much software available publicly
01:03for these models to munch on,
01:07there is a software languages tend to be easier
01:10than, you know, human languages.
01:13Python code for a Python grammar, for example,
01:16is like a page and a half of grammar.
01:18And then verification and all of these ways are built in to these.
01:22So we have always had a QA industry, you know, and so on.
01:26So for all those reasons,
01:28software development is a very powerful application of this AI.
01:33But people have to understand what it is that you are building.
01:36They have to be able to precisely guide AI
01:38into building exactly what they need
01:40and to avoid building the things that they don't.
01:43And then once it has been built,
01:45you have to verify that indeed what they wanted got built.
01:50So that definition step, the what step, is still there.
01:54It is still human-guided, human-centric.
01:58Arguably, it is even more rigorous than it used to be before
02:01because you could talk to the programmer before
02:03and, you know, come to the same page and things like that.
02:06And now, because that construction time has basically become zero,
02:11so you define the thing,
02:12you take your time making sure that the definition is right,
02:15and then you build it, and maybe you iterate this a few times.
02:19And then once you have built it,
02:20you have to check that this is indeed what you wanted
02:23and it doesn't have any untoward or negative consequences in it
02:29and no issues, security issues, reliability issues,
02:33and things like that.
02:34And that entire cycle becomes much more compressed.
02:37It becomes much cheaper, much faster.
02:43And, you know, it is possible to build a lot more,
02:46to build it even arguably if you know what you are doing,
02:50you can even build software better than we could build before.
02:53That's what we found.
02:54For example, you know, one of the quotes that we had
02:57from our customer in our press release
03:00actually says that,
03:02that the quality and the efficiency
03:04with which the software got built,
03:06and this is in an incredibly demanding environment
03:09where you are building software
03:10that controls mechanical systems.
03:13So the AI is not running on the mechanical system.
03:15The software that was generated with the help of AI
03:18is running on the mechanical system
03:21before people, you know, start to freak out about this.
03:24So I think that's a very powerful application.
03:26And yes, in the short term,
03:29it might displace the way we used to build software in the past,
03:32but there is a tremendous need to build this kind of a thing
03:35and there is a great demand for it.
03:38And, you know, I am hiring.
03:40I cannot hire fast enough for my company.
03:43So I think I see this as an extremely positive development.
03:48Now let me bring you to the regulation question.
03:51At the BRICS summit here in New Delhi,
03:53Prime Minister Modi had warned about what he said,
03:57weaponization of technology and critical minerals.
03:59that technology itself is now a geopolitical lever,
04:03not just a tool.
04:04What role should nations play now?
04:11No, I think that call that he made,
04:13the Prime Minister made,
04:14is extremely, is absolutely correct.
04:17And it is extremely necessary
04:18that this is the time where this technology
04:21has to be much more broadly and universally available.
04:24It has to be accessible to people everywhere.
04:28We have to endeavor to do that.
04:30And that's why I'm very happy that in India,
04:32we are doing our own efforts towards building these models
04:35and, of course, applications on the models,
04:39environments to run the models.
04:43The way this ecosystem works is you have a model.
04:49So the next generation model is built from there.
04:53There is a ton of data.
04:55The, you know,
04:56the frontier companies have more or less run out of public data.
04:59So they all get synthetic, synthesized data
05:04that train the next generation models on.
05:06Vast amounts of synthesized data
05:08that are made by companies and other organizations
05:11who actually produce the data for them.
05:14And then because the frontier labs have the deepest pockets
05:17and other factors,
05:19they tend to get the, you know, best quality data first.
05:23And then a few months later, the other companies do.
05:26And then you see the second generation,
05:28the wave of next wave of models and things like this.
05:32So this is sort of how this ecosystem is evolving.
05:34And once you try to understand it,
05:36it's not that mysterious.
05:37These are a model, an LLM,
05:40a reasoning model is a very powerful thing,
05:42but it is basically a mechanical system.
05:46It's an extremely predictable mechanical system
05:49that is given vast amounts of data
05:51and it produces outputs based on the prompts
05:53that you give it in a very predictable way.
05:57And so this is the ecosystem.
06:00Now, people who have the better models
06:02clearly have an advantage.
06:04And so if there was a weaponization of that
06:07that were to happen,
06:08that would clearly disadvantage people
06:10who don't have access to these models.
06:12So the prime minister's call is exactly right.
06:14It is necessary in this time.
06:19Then how do you look at
06:21what US President Donald Trump has said?
06:24He has dismissed calls for caution as negative forces
06:27and has in fact doubled down on the idea
06:29that whoever wins the AI race wins geopolitically.
06:33In a world where leaders like Prime Minister Modi
06:37are talking about weaponization of technology,
06:40you have President Trump
06:41who is speaking a different language.
06:44Then how will regulation really come through?
06:47Because it is about immense collaboration
06:50between various nations.
06:54I think that the way I see it,
06:58the more, first of all, I don't see it.
07:01There is a lot of interdependency already.
07:05The AI labs are very diverse.
07:08There are people across the entire ecosystem
07:12from the chips made in Taiwan,
07:16designed here in the US by NVIDIA designers,
07:19and worldwide,
07:20and we have a lot of chip design work in India as well.
07:23And then the manufacturing of it,
07:26then you have memory and other components and networks.
07:29So it's a global kind of ecosystem,
07:31a global supply chain
07:33that comes together to make this real.
07:37And it is true that models are,
07:40the frontier labs are,
07:41the three frontier labs are American,
07:43and then you have a bunch of,
07:45I think, five major Chinese model makers.
07:49These are sort of at the frontier,
07:50and then there are several other companies
07:52that are doing this.
07:54But I think we should keep in mind a few things.
07:58One, there is a long journey ahead.
08:02These models, this is not the end-all of AI.
08:06People are already working on next-generation models,
08:08world models, models of the physical world,
08:11which are much more efficient and reliable and so forth.
08:15So that technological progress
08:16is going to continue for a long, long time.
08:20Second, I think the more that we are aware of this,
08:23the more that we are educated on what it is,
08:27what it is not,
08:28what it is capable of,
08:30the better off we will be.
08:31Especially, you know, for us in India,
08:34the more that we are able to use this.
08:38And in this time, you know,
08:41one of the things that the Prime Minister once told me
08:43was this idea of a billion entrepreneurs.
08:49One of the things that AI does,
08:51this generation of AI,
08:53is it makes it possible for people to do things
08:56by themselves that were not possible to do before.
09:01You can do things that would have taken very long times
09:05much more quickly now.
09:07You can do them in minutes or hours,
09:09what would have taken weeks or months to do.
09:12And so in this situation, what would you do?
09:14So I think that the way I see it is that
09:19the entrepreneurial possibility with AI is enormous.
09:23It is, individuals can do things
09:27that teams and organizations used to do before.
09:30So what could we do with that?
09:32You know, I think there is also an imagination deficit
09:35that we have to some extent
09:37that we are not able to see the possibilities
09:39of what this could do.
09:41My wife, Vandana, you know,
09:42she has a beautiful app that she has built called Wasi.
09:45And her idea was actually,
09:48part of the idea occurred to her
09:50during the India AI Summit in February,
09:52that in this time of AI,
09:54what is the uniquely human thing?
09:55And so this experience, right?
09:57Like you and I are talking live right now.
09:58There is no AI in this conversation.
10:00Later on, AI can transcribe it
10:03or make, write a book about this conversation
10:05or make a comic book or whatever.
10:07But this conversation itself is live.
10:09It is an experience.
10:10And so she has these live experiences
10:13where people can come and teach things.
10:15You can sit in with 10, 20, 50 other people
10:18and participate in something
10:19that is truly authentic and live,
10:21that is being done by some creator
10:23who is an expert in something.
10:25These kinds of things, you know,
10:27are much easier to do now because of AI.
10:30So I think that the biggest gap,
10:35you earlier used the phrase gap.
10:37The biggest gap that I see
10:38is the gap between our ears, you know.
10:41I think the more that we focus on
10:43what is it that I can do with this?
10:45How do I understand it?
10:46How do I build it?
10:47And how do I use it?
10:48What can I do with it?
10:50The better off everybody is.
10:52And I think then this whole,
10:53all the drama around it,
10:55it will play itself out
10:58like other technological waves have in the past.
11:01I think the important thing to focus on is
11:04how do we develop the awareness
11:07and the skill of what do we do with it?
11:09And I think that's a much more promising path.
11:12That brings me to my last question, Dr. Sikka.
11:15for countries like India,
11:17which are not yet at the frontier,
11:19but are major adopters.
11:22How do you see this debate even translate?
11:25Can we afford to slow down
11:27when access capability and scale
11:30are still being built?
11:34Oh, no, absolutely not.
11:35In fact, we need to accelerate.
11:37We need to embrace this technology.
11:40It is already here, you know.
11:41It is, what, we are at version 6 of ChatGPT,
11:45version 5 of Opus and Sonnet
11:48and from Claude and Fable and all these.
11:52No, this is, it's a powerful technology.
11:55Yes, it is very expensive to run.
11:57It needs specialized computers and all of that.
11:59But all that will get better over time.
12:01No, this is not the time to slow down.
12:03This is the time to understand what it is,
12:05what can we do with it
12:06and bring that to a massive scale.
12:08Dr. Vishal Sikka, always a pleasure
12:10listening to you and your wisdom on AI.
12:13Thank you for this absolutely fascinating conversation.
12:17Thank you for your time, sir.
Comments