FDE Episode 1 : AI Reshaping Global Research & Infrastructures | Weekly Tech Update

Welcome to the Code Ninja Spaces. Today we are breaking down some major updates in AI. So Omar, I want your take on this first story. There's an AI system called Aleph just solved famous several math problems and the entire cost of running it was only around $5,000. So what does that mean for the future of scientific research?

So there's a big shift. Historically solving a major math problem required lifetime of expertise, sometimes decades of human effort. And what we are currently learning is that a machine has now solved multiple open problems in 48 hours using commodity cloud compute for less than the cost of a MacBook. And that's very significant, especially for regions like Pakistan or I mean in South Asia for example. It means scientific progress is no longer limited by how many PhDs a country has young researcher with access to cloud compute can contribute to a frontier research and it's. I think this is incredibly exciting.

True. So Alex, the AWG call, this is the beginning of theorem factories. What's your take on this?

So I mean what I can explain is, what he means is math becomes scalable industry rather than an individual sport. So traditionally math breakthroughs were rare events tied to exceptional individuals. I mean we all know those great researchers in our universities that are basically that have spent their lives in probably generating proofs for some theorems. But that's now what AWG is saying is that's what commoditized and I think there is truth to that. And I mean imagine, I mean there are students and researchers in our universities here in this part of the world and they can now contribute at the same level as let's say by the students at mit if we deploy AI systems. And then also adding on top of it is like OpenAI has invested a lot into this how AI can help scientific research and they have invested in company called STEAM Science I guess. So that's one of the biggest areas where AI is going to create a lot of value.

So interesting there are some talks about AI moving from one big AI model, I mean monolithic AI to diversified cognitive ecosystems. So what's actually happening here? The Omni models and stuff like these.

So Omni model like. So I think what you're referring to is like what AWS has released. They have released a new Omni model. So what we are seeing is that the world is shifting from one model dominates like ChatGPT to a world where many specialized models would coexist. I mean AWS for example has launched this Omni model that can process text, video, sensor data in one go. Mistral, Even, even launched Lars 3 which is extremely good at reasoning under low latency. I mean we all talk about model performance and how good model is. But if you bring latency into the equation then what really matters is like, which can perform better in the scenarios where for example you have, you have to perform some compute at the edge. So for example Mistral large tree model, that's incredibly good in reasoning under low latency. So these are the things that we're seeing like when we think about AI, we always think about these 3, 4 models, Cloud OpenAI. But when we think about how can we deploy these systems into environments, let's say where most of the compute is required at the edge, then most of the performance is required at the edge. But then we are limited with compute and all there. We need very specific systems. So I think that's what we're seeing here with the Large 3 and models.

Like that shaping AI like in what kind of benefits we going to have to it.

So if clearly AI is not just a software anymore, I think it's more of like industrial infrastructure. Every data center needs enormous amounts of e.g. copper wiring, transformers, cooling systems and power generations. And as AI work, workload grows, countries need stronger grids, more fiber networks, more solar, more chips and more cooling. So it's a, it's not just AI, it's not just a software everything as, as for example Jensen Wong said, like Jensen Wong said like it's a, it's a 5k piece at the bottom there's a chip, then there is, I think infrastructure at the top is application, that model. So it's not just one thing. So for example, countries like Pakistan, Saudi Arabia, uae, they can participate through manufacturing, energy and data center infrastructure. We don't, I mean historically we've always thought of, when it comes to software, we always thought of like okay, we need to think about how to build software. What it means is they have to build applications, software application. But I think now we need to also think about how, I mean software is not just an application layer. Software starts from now, for example, energy. So I can see software companies would be investing a lot in data center infrastructure, in energy infrastructure down the line because that is now software that's part of software. Historically, I think we've missed many revolutions, but I think this revolution can't be one because the reason for that is this revolution AI revolution is incredibly dependent on energy, real estate. And I think in the Global south we have energy surplus, we have real estate, there's lots of sunshine, we can use lots of renewable energies. And the west needs it. So yeah, I mean I think that these are the things that we need to think about when it comes to software. Not just software as if like application.

There definitely AI is reshaping the global supply chains and affecting the physical world as well. So in your opinion, how does this create opportunities for Pakistan and specifically for gccs?

Okay, so as I said, like let's imagine you want to set up a large scale data center. So data center, I mean when we talk about data center, I mean at the macro it looks extremely complex. Yes, and it is complex, but it doesn't start from just from like we need lots of chips. It starts from you need to acquire real estate. So it's a, it's a huge real estate business. When you talk about data centers then it, it involves manufacturing. If you see in America there's a huge uprise in people who are basically electrical or mechanical, I want to say engineers, diploma holders. So they're, they're getting lots of jobs. Plumbers, electricians, these are the kind of people that's, that are required. So for example here in our part of the world, when for example think about real estate, they currently I think are not involved very much into that. This software boom is pretty much real estate boom. It's pretty much energy boom. And then it require, then it's requires lots of electrical technicians and engineers who set up those infrastructure. So for in terms of opportunities, think about how many electrical engineers did this part of the world produce? A lot. How many mechanical engines this part of the world produce? A lot. So the, when we think about software, we need to think as a whole. We don't, we can't think of software just as a computer science anymore. We need to think about the intelligence as a whole which includes all these fields what we've talked about and I'm actually in talks with some of the, some of the people who, I mean they never spoke a word about computer science in our earlier conversations. They all, they're focusing on real estate, data center, electricians, mechanical engineers, manufacturing capacity that we have. So these are the things that's required to build a great software.

Now both of us are in the trenches of software engineering and AI, so and we talk a lot about energy efficient software engineering as well. So what do you think how it will lead us towards the energy efficient software engineering?

So I know what you're referring to. So energy efficiency software engineering for in our case is like because we're building this platform hyper and there our optimization function is energy efficient software engineering. What it means is like what's possible through train automation. Don't consume tokens for that, let's make it token efficient. So as I said, like I was in the podcast with this Martin Flower and said like he wanted to change a class name through cursor. Then it took 30 minutes, the name of the class not changed, but then it came back and eaten all of the tokens that he had. So I mean this could have done, I mean, you know that this could have been done I mean just like in a split second by an engineer. So what it means is like how can we make software engineering where we are, where just drive efficiency in terms of energy too? I mean at the end of the day I think like for example, we've been talking when we talk about information, we talk about bits and bytes, but I think in the next four or five years we'd see there's a shift from bits and bytes to tokens. And tokens is basically energy as Elon actually said. Like it's basically the equation is intelligence produced per gigawatt spent. So intelligence produced. I mean the best abstraction of intelligence at this stage what I can think of is like tokens. So how can we make sure we can consume intelligence in a way that we have to spend extreme, we have to spend less amount of energy to produce a lot of intelligence. So that's what energy efficient software engineering means is like don't consume tokens where you're not required, where you can do for example to automate simple crud, you don't need, you don't need LLM RAW code, you can simply write a framework or platform do it for you. So and you would save a lot of tokens for that. So that's, that's what energy FM software engineering means.

Yeah, I do agree. So that's it for the today.

Thank you.