Speaker 1: And joining me live to throw some light on some of the questions that DeepSeek seems to throw up is Ganesh Gaur, he's a well-known tech expert. He's a founder of the India Future Foundation. Ganesh, welcome, good to see you. You know, DeepSeek is the top trending word right now across all social media. It's all everyone is talking about, you know, everywhere in the world. You know, a few basic questions, because everyone is sort of familiar with ChatGPT. You know, how do you see DeepSeek? You know, because the word out there is that it's been made faster, leaner. It's been made by China, which was perceived to have a big gap with the United States when it comes to AI models. How do you see DeepSeek? What would you say about it?

Speaker 2: DeepSeek has been built on an open source-based model. So far we have seen OpenAI came into the picture, whereas it is having a commercial model. It is a commercial model where we have seen Microsoft that has contributed to it. And primarily we have seen open source models being more transparent and enabling more innovation. The key difference with DeepSeek is the Virgin V.3 adopts multi-head latent attention and an architecture which allows better loss of frequency and better load balancing. Hence the predicting capability of this model is well-developed. Again, the pre-trained models, which ultimately have the efficiency built up, are far better and have basically more precision. So when it comes to first-time validation or training on extremely large language models, DeepSeek is far better as compared to some of the existing models that exist. You would also see the co-design of algorithm frameworks and the hardware through which they communicate is way better than OpenAI or some of the existing AI platforms like Cloud, Microsoft ProPilot. Hence it enhancing the training efficiency and reduces the training cost. So what you could do is it can scale up really fast and models could still provide support without having any overload. Hence its ability to provide output with the economic cost really changes the dimensions completely. Hence the V3 model with a 14.8 T tokens would produce strong amount of open source base models that would allow retraining of 0.1 million GPUs per hour, which is enormous. So that's one of the key differentiator and no one has got that level of efficiency

Speaker 1: built in so far. And since this is from China, obviously there are all manner of questions. I put out a tweet today talking about privacy and many people laughed saying it's open source and you don't understand it. And I represent millions of people out there who are still sort of searching in the dark and trying to understand many of these things. And that's the reason why I'm doing a show on it and trying to get some of those questions answered. But stay with me, Kanishk. Vivek Mehra, who's co-founder and Chief Sustainability Officer at Only Good AI is also with me. Vivek, welcome. You know, the other question that a lot of people are asking, Vivek, is, you know, which is a basic question beyond what DeepSeek is and what it can do is, you know, where is India's chat GPT or India's own, you know, AI model? You know, why does a country which has such a large IT base and so much engineering talent not have some, you know, big global AI splash the way chat GPT and DeepSeek do? What's the answer to that question, Vivek?

Speaker 3: Yeah, so it's obviously not an easy answer. I think one of the challenges is the availability of data and the training models. So it's not for lack of expertise, for sure, on the algorithmic side, in terms of design, in terms of building algorithms. It's really having data which is so heterogeneous, which allows for reinforcement learning where the data can interact with the environment, create more synthetic data that the machine can train on, and then improve the models. So India's AI market is much more fragmented. There is no concerted push towards creating an India-specific language model, which caters to India use cases, caters to the heterogeneous nature of society in India and the industry base. So there's no clear reason, but I think it's not for lack of expertise, for sure.

Speaker 1: Okay, it's not a lack of expertise. Ganesh, could you weigh in on that? Is India going to make some kind of a big global splash? Do you see that on the horizon? Because this appears to be the biggest game in town. India shouldn't be missing out.

Speaker 2: See, for India to make a global splash, we really have to bump up the algorithms and the speed of our Bharat GPT, which, you know, IIT Bombay is working on. Or let another institution in India rack up something which could be really quick. But again, it depends on the GPUs that we have the hardware, apart from the open source models. So how do you train them? How do you build out more efficiency is again going to be the key. And how large language models will work enabling better communication to happen and better relay to happen. That's going to be one of the key differentiators we have to look at. And, you know, it will take time. They started, you know, the DeepSeek started building it one and a half year back. Now they have built up something and now they have launched it. That's why the adoption is quick. If you look at it, some of its key benchmark parameters, it is far better than existing models, whether it is its architecture, whether it is, you know, its file test speed, whether it is, you know, the drop which is allowed, everything is far better. And the user experience is also so great that you really don't need a large infrastructure. We have so far been debating on infrastructure, saying India doesn't have the GPU or the infrastructure. It is more to do with who's willing to do the investment in India. Is it going to be the government? Is it going to be the industry? So far, we haven't seen large investments being committed from businesses on this, apart from Tech Mahindra, which said they will build something very similar to what OpenAI is doing. We haven't seen anything being committed from companies, which are the tech giants from India. They haven't really gone into the race. So today, what is needed is that some of these tech companies to build up institutional partnership with academy institution of eminence or startups. We also need venture capitalists who are willing to pour in the money. We have seen OpenAI raise tremendous amount of money, but where are the Indian venture capitalists? A startup called CoreOver has just been able to raise only 5 million. They are trying to build out something on AI. So the funding ecosystem is also lacking. So we will require government to really rack up and make heavy investments

Speaker 1: if India wants to continue to be in this race. Let's hope we see something of that kind. This is a technology, this is a world that has changed the world completely. It has made a new normal as far as our life and our work is concerned. And as both Vivek and Kanishk probably know better, our lives are all going to change very, very rapidly, exponentially, really in the weeks, months, and the years ahead because of AI. So this is something we need to take very seriously. Appreciate your time, Vivek and Kanishk, for being with me and helping us understand what DeepSeek, the new kid in town, really means for us all. Thank you.