The Iferia Techcast
00:15:16 · Ezekiel Iferia
[00:15:16] S0: Alright. Alright. That was a really, you know, eye opening response. So, Krishna, you know, you you talk about the world moving toward sovereign AI, you know. You know, What what, you know, does that mean for a country's security? You know? Why why are nations starting to treat AI like they treat nuclear technology or intelligence?
00:15:45 · Krishna Kumar Sharma
[00:15:45] S1: Yeah. Basically, like, this is a very good question. And in terms of, like, why AI is getting treated as nuclear technology, the answer is that, like, AI is, like, one of those technology which is, like, revolutionizing the way we perceive the world. The the major thing which AI is able to do is that it is able to scale things beyond what we were able to scale before. Example, like, recently when Israel started attacks in Lebanon or Iran, they were able to attack at multiple locations at the same time with very pinpoint precision, like such a precision that I'm sitting here in my bed, and then a missile came. It did not destroy the entire building. It just dug a hole in my wall and only blasted my bed. So imagine this is the level of precision at which you are able to control weapons with the help of AI. And of course, when something is too powerful, you do not want other nations to have it. The same thing happened with nuclear weapons because they were able to do mass destruction. And now with the AI, you are able to do the similar level of destruction. So if it is not a real weapon, then using AI, you can do cyber warfare, you can do propaganda, you can do a lot of different dangerous things, right? And then every nation would want that they have the most powerful weapon so that they have the biggest stake to scare everyone away. This is the reason why sovereignty comes into picture because like seventy years ago, every nation wanted a nuclear power, nuclear weapon, so that maybe they don't want to attack, but they want to defend themselves or not get pushed around. The same way now AI is important that if you don't have this power, you will get pushed around by powers who are really controlling the AI.
00:17:50 · Ezekiel Iferia
[00:17:50] S0: Okay. So, you know, you are building for defense use cases. You know? So what... What's does that world look like to someone on the inside? You know? What are the ethical lines you think about when you are... You know, when when you build for that space?
00:18:07 · Krishna Kumar Sharma
[00:18:07] S1: Yeah. In terms of ethics, I would say that war in general is unethical. Right? Like, I I think, like, wars are bad for all of us. Even if you are not directly involved in a war, you will... You or your country will somehow be affected negatively because of the way the world is operating. But at the same time, you need to have defensive capabilities to protect yourself because the way our today's world is quite chaotic. In terms of ethics, I would say that what we are building, I can comment on that because the way current wars are happening, there is no ethics involved over there. But in Europe, the guardrail which they put in terms of ethics is that an AI should not be attacking a human by itself. Like you need to have AI can detect a target, and then a human needs to decide if the human wants to attack the target or not. This is something which Europe is doing. From my understanding, other countries, they don't have such guardrails. Like, there, the AI can detect the target, attack automatically. And the problem here can be that while AI is amazing, but AI is still like a pattern matching machine, and sometimes it can actually do mistakes. And so we don't even know a clear picture, like how many misfires would have happened or what boundaries has been crossed because of using AI because this information does not go openly. But I definitely feel that the way we are currently approaching AI, there can be some serious consequences in future.
00:20:04 · Ezekiel Iferia
[00:20:04] S0: Alright. So... No. Krishna, you you you you say that you've... No. You say that you discover the way to achieve fable five level quality, you know, using multiple AIs working together. So tell us about this. You know, for someone who is not very, you know, updated with the terminologies in recent AI development, you know, what does that mean? You know? Just walk us through it.
00:20:35 · Krishna Kumar Sharma
[00:20:35] S1: Yeah. So basically, like... So Fable five is, like, one of the best models out here in AI. Yeah. And the major use case where actually AI currently getting used is coding on problem solving. Like defense, maybe AI is like maybe 2% of the total defense spend. Like, 7%, 98% of AI expenditure is happening on either coding or trying to solve some... Help people with productivity. And the idea is that each model in the world, like, you know, let it be Anthropic, OpenAI, Gemini, or the Chinese models, they all are built with their own philosophy, with built with their own biases. They try to put their own special mechanics. And the way these models work is that they all hate each other. So when you have these kind of models which hate each other, and then you ask them to review each other's work, they really go and they really try to find real bottlenecks, like in a code written by one model and the other. And basically, what happens in this disagreement war is that they continue to raise the quality of your work. And for this to happen, you have to use at least three different AIs, not just one. And when you use three different AIs, example, in my case, I use, let's say, Claude Opus, and then I use GLM 5.2, like a Chinese AI, and then I use a codex from ChatGPT OpenAI. So here, let's say ChatGPT created some response. Then I ask ChatGPT itself to review the response. I ask GLM to review the response, and then I ask Claude Opus to review the response. And then the fascinating thing is that they all find totally different 50% of their findings in reviews are very different from each other. And when you bring them all together and you try to solve them, you will see that the level of code, the quality of the output goes considerably up. This is something that I'm saying. This is already... While I identified it last year, currently, as for the latest research, example, OpenRouter has the OpenRouter Fusion API. They release this same thesis that if you use three different AI models, you are able to get Fable five like quality. Similarly, in Japan, there is a company called Sakana AI. They got really hyped up a few weeks ago. They are also saying the same thing, that they are using multiple models in solving a single task, which they are able to get Fable five like quality.