What Open Source Adoption Looks Like in 2025 with Manos Koukoumidis

Hey everyone. Welcome back here to Techstrong tv. I'm really happy to introduce you to a first time guest here on Techstrong TV and a new kind of a company and movement to talk about.

Let me introduce you to Manos Ku COIs. Uh, he is the CEO and co-founder of something called, excuse me, OMI, OUMI, open Universal Machine Intelligence. He's gonna tell us all about it, but Manos, welcome to Tech Drug tv.

It's great to have you on here. Thank you very much for having me, Alan. It's my great pleasure to be here and talk more about, uh, OMI and as you mentioned, both OMI as a company and the bigger mission that we're pushing forward.

Absolutely. So, Manos, um, let, before we get into umi and everything, you are the co-founder and CEO I've in, I've interviewed literally hundreds of founders, co-founders. Every single person has founded the company because they buy into the mission.

They feel in some way it's going to help change the world, help make the world better, but they, no one, no one is born and co-found a company. Do you know what I mean? There's always a life before.

Let's hear a little bit about your life before co-founding omi. Absolutely, absolutely. So maybe I may take you a little bit, uh, very far back, but I started wanting to become pretty much like my father to become an electrician.

And luckily the best school in Greece was the electoral and computer engineering school. So, as by accident, I got into computer engineering, quickly fell in love with computer programming. Uh, and then when I was going to my PhD not knowing what AI really is, I kind of almost bumped into it.

It was when the first iPhone was coming out and I was like, you know, there has all these sensors, all this data you need to do something useful with it. And that's how I bumped into ai. And way before any company said, we are an AI first company.

That was back in 2007, 2008. I'm like, you know what? AI is gonna be big.

You need this machine learning. You need this technology to do something useful with all this data. And that's how I got and started getting deeper and deeper into ai.

Then, then moved into Microsoft, started working a lot with natural language ai. I even built something like in 2016. I was a little bit premature, uh, and then, yeah, continued on with that at Meta, working on conversational ai, uh, then a startup, then Google where I was leading all the natural language AI services, uh, including also bootstrapping the efforts for pal.

That was the model before Gemini. Yeah, Uhhuh, very cool stuff, really. So you, you've literally had a, you know, been right in the middle of it here as we've launched into this, you know, era of ai.

It's funny, I have a friend, John Willis, he has a new book coming out this week or this month, kind of the history of AI and the people behind it. I don't remember the name of it, but if you go on Amazon, look up John Willis, it's his newest book. What I think a lot of people don't realize is this whole AI thing that burst on the scene, what, two and a half years ago, maybe with chat GPT Open AI has really been something that's 50 years or more in the making, right?

And the idea of what we now call ai, whether it's a form of machine learning, pattern matching, et cetera, right? All the way through the latest and greatest agent AI and LLMs and training models and all of that. You know, it didn't just spring forth.

It's been a gradual buildup over all these years, but there's always some mono, there's always some ignition point or something that lights it up, right? That, that makes it, uh, just go viral, so to speak. Obviously, open AI's chat, GPT was that event here, but why, look, you were working on this, as you said, in 2017, 16.

Why then why, why not in 2017, what was missing then? That is now. Yeah, f first of all, Alan fully, fully agree with the points, all the points you made, uh, ai, machine, machine learning, AI has been in the making for quite a bit of time now.

Um, and as you said, you know, back in 2016, we built something like ZGBT. Even back then, I could see people were amazed how there was a chatbot like ZGBT, uh, that could respond to any question you would throw at it. But the technologies are very premature at the time.

In other words, not as advanced at the time. For example, what we now, we call, we use Transformers, it hadn't been invented yet. We're using their predecessor called Ltms.

So even though people were amazed, the quality was just not there yet, uh, to be coming the product. But lemme tell you, I mean, there was quite a few times I could see people being amazed at the time, you know, almost the reaction we have with ZZPT. It just, you know, every now and then the system would say something that doesn't make sense, and then people will lose interest.

And, um, but I think that's, that's, you know, that's, uh, I think what open I did, they did manage to scale up those models and some continue making them better and better. And because, as you said, they make it, they made it so accessible to people. It, they got this viral moment that, uh, finally, you know, they took, uh, the world by, you know, uh, the huge excitement around the world about what they built.

But there was definitely something in the making with many research labs, not just what we did at Microsoft, but even after that building such technologies. But I think they managed to give it a good delta in the quality improvement and then make it so accessible, building the hands of people that, uh, created that earthquake in the industry in, uh, anywhere in the world. Absolutely.

Absolutely. Let's turn to Umi. What, what drove, you know, you, you worked at Microsoft and Google and Meta Giants, you know, the, the hyperscalers as we call them.

What made you, you know, people don't do this like, like, you know, uh, without good reason. What, what drove you to co-found omi? Yeah, that's a, that's exactly right.

Uh, the, the, as you were also discussing earlier, the main thing was the mission, uh, starting from the problem that we wanted to solve. The more, uh, as I was mentioning earlier, I bootstrapped the efforts of production. I found working with, I don't know, 20, 30 teams all across Google to make it happen, until that effort moved to deep mine.

But even as I was continuing to work on Gemini, when I moved to deep mine, and I was increasingly realizing how more and more sciences, you know, for not just healthcare and the tech industry, but climate science, material science, and many, many more robots, all science going forward, gonna be powered by ai. I was increasingly getting worried about the future I was contributing to. 'cause I was thinking it can't be that what's gonna be the foundation for all, both tech industry and all the science.

It can be a black box that is owned by open AI or Anthropic, or let's say Google or somebody like that. It needs to be, we need all the science need to have more free, uh, access to this AI technology. It needs to be a glass box, something they can easily look at and adapt so they can promote the sciences.

So there was that, uh, you know, philosophical concern. And then I started also realizing something that was, I would say, not clear to many people. I would say still not clear to most people, which is that while these large tech giants, they may brag about the number of GPUs that they have to build these models.

This only part of the story, the thing that most people don't know is that, uh, they're actually themselves very constrained in terms of the human capacity. Those technologies are extremely complex. They don't take a lot of GPUs to train.

They take also human amount, humans, huge amount of human resource and human ingenuity to continue moving them forward and advancing them. And that's where I realized that actually AI is the most prime technology to be advanced in open source compared to operating systems like Linux, or compared to database like PostgreSQL, even more than those technologies. And the best way to advance it, both faster and safer and most cost efficiently, is to do it in the open, to do it collaboratively with all the open community that is orders and orders of magnitude larger than the couple thousand people that are in Deep Mind or in open ai.

Uh, and that's actually, again, the best, the best way to advance ai, the safest way to advance ai, and, um, uh, the best way also to make it accessible then to all the sciences and enterprises. And that was the kind of the, the most important, I would say, motivations that led us to the fund domain. Yeah, fair.

You know, you, you, you mentioned, uh, or in, in some of the OMI stuff, is kind of trying to pattern OMI after sort of the Linux movement as it eventually triumphed over the, the many flavors of Unix, if you will. Yes. Right.

But to be fair, when Linux did that, we didn't have what I call the foundational era of open source, right? We have the Linux Foundation now, which is of course, a lot more than just Linux, as you know. Um, you know, open source now is a much more defined, uh, method of go to market.

However, there's the flip side, right? Where we, we, we've seen a lot of companies struggle with an open source model, right? It, you know, it used to be a Red Hat was the big success story, obviously, but we've seen many companies succeed with open source, but we've also seen many companies not how, how can OMI be successful with this open source model?

Yeah, lots of lots of good questions there. So the first one was that, uh, if you come to think about it, I, I, I, I, I, the, the, actually, the more I was thinking about, the more I was getting convinced that the best way to develop these technologies and compete even as an enterprise, would be to do it in open source. Because currently what's happening with Open AI and Tropic and Google and all these companies, it's, uh, it, it's, it's like, uh, about who has the deepest podcast to keep investing in AI and keep draining all this money that if I were a shareholder of this companies, I would be very frustrated how this money is being spent hoping that they win this, and hopefully they make, make it big, and that recoup all the money they have lost across the years with some getting, uh, less and less optimistic it's gonna be happening for any of them.

Um, and that was actually the key conviction that the best way to develop this technology is not to try to spend more money than open ai, but to say, you know what? No, as a community, we're all gonna contribute to this boat because there is all the open community, all the academia, all accelerator providers like Nvidia, a m, the oldest companies, all cloud providers that are not one of the few aspiring AI guards, all these companies that want open source succeed. So the best way to compete as OMI or as any company, is to bet on open source as opposed to trying to outspend open ai.

Uh, it just, it's the best thing to do for the world, the best thing to do for humanity. And I think the most viable threat is to compete. And given that there was this gap, as we mentioned, the example of Linux and Unix, there was this gap of there was no Linux of ai, there was no platform that democratized this frontier, frontier AI issues and development.

Uh, that's why we said also, you know, we're gonna build this because it will help. It's, first of all, it's gonna be a positive thing to do for the world, and we think it's gonna be the MVP is gonna be the thing that will, uh, resolve the initial friction that the community has to help it then to continue advancing and doing their research to advance Frontier ai. Because if you can unlock and enable this community, and the more they innovate on Frontier ai, then it, these are gonna be for Rumi and any other company to stand against this, uh, you know, as a David stand against those colias and compete with them.

Uh, so that's why that open source strategy, you know, I'll be very candid. I think it's, it's both self-serving for omi, but at the same time, I think it's the thing that benefits humanity and every other small enterprise that wants to be able to compete with open AI or anybody else. Fair enough.

It's gonna be interesting, you know, what else? So, open AI and, and our, our audience is familiar with this. It's not open source open ai, but it is sort of this foundational not-for-profit.

You know, there's all of this tied up in there, and of course, Elon Musk is suing over and what he's doing and everything else. Ask what makes us, you know, there's a lot of flavors of Linux, right? There's SUSE Linux, there's Red Hat Linux, there's Rocky Linux, there's what makes us think Umi is the one, you know, to put our efforts behind versus Yeah.

Yeah. Others, first of all, just, just a quick comment. As you mentioned, OpenAI is nothing but open.

It's as closed as it gets. Yeah. Um, and omi, the way sometimes I describe OMI to people is that the polar of OpenAI, you know, OpenAI is nothing but open.

It's fully closed. I normally would try to make truly open criteria, but as opposed to building a model and giving it to people and say, you know, here's a great model we developed, now you can use it. Instead, we're starting from empowering the community and everybody to contribute to making AI better.

I think that's the key thing for, to unlock open source, not develop a model and give it to everybody, but empower everybody to come on board. I think that's a big, I would say, philosophical or strategy difference for omi that I think is gonna be key to our strategy. And I think it's key for OMI to win this key for open source to win this, uh, which is that you bet on the platform, you start by enabling the community to advance criteria.

So Manos for people out here who say, Hey, this is the thing I've been waiting for. I want to get involved in ai, but frankly, I, you know, I was waiting for truly an open, open source community to come in and do it. How can they get involved with umi?

Yeah. So I would say they definitely can and a lot more, and a lot more easily than they could until recently. Uh, and that goes again to the design of, and the goals with whom it was, was to create a platform to create the lineage of ai that would make it easy for anybody.

Definitely for researchers and ML engineers. It makes it a lot easier even for them. But even for people who are more like application developers, uh, to experiment with, uh, ai, and they can go incrementally deeper and deeper.

We say that even though OMI is open source, it's as easy, if not easier to use than the API is built by open ai or that my team was building for MI. Um, but at the same time, because open source, you can experiment and stay at the level you're comfortable because we make sure it's very easy with a lot of examples to follow. But then as you get more and more proficient, you can get deeper and deeper.

And I think that's key to any individual person or any enterprise to start building the up the AI muscle, which I think in this time and age, it's gonna be very critical for any individual or enterprise. Agreed. So for people here who want to get involved, what should they do?

So we have, uh, the only GitHub that's be the best place to start your, I think if you Google search, uh, only GitHub, uh, I'm sorry, it'll come up as the first entry and yeah, sorry. From our GitHub, somebody can find all the information, all the documentation we have about how to install umi a library, how to get started. Again, we have a lot of examples and we got a lot of positive feedback about the quality of the documentation, because again, that goes back to our design principles, which was we want to start by enabling the community, which means we give them a great platform, great documentation, all that they need to get going easily and without friction.

Any plans for, like, is there a Discord server or anything like that that we can, people can get onto? Yes, absolutely. ai.

There's a link both to our GitHub and also a link to our Discord. Uh, we have an active Discord. It's over a thousand members right now, uh, with a lot of discussions happening, and even some research efforts that are being organized through Discord, uh, on our website.

Also, we have a form that people can fill if they want to contribute to those research efforts because we're spinning up more and more such research efforts. So yeah, definitely. Um, check out GitHub, if you want to get started, discord, please join for discussions or if you want to join some of the research projects or fill up the format, uh, OMI ai and, uh, we can make sure to include you in any of the future research efforts.

I'll plan. ai, open Universal Machine Intelligence. Manos, I wish you guys much success in this.

It would, it's gonna be great to see how this develops. You know, clearly a couple months ago with the Chinese deep seek, one of the, one of the selling points, I don't know if it's the word selling, but one of the bright spots on it was that it was open source. And people really liked that idea, though.

How much of it is open is another story. So having one here, umi, it'll be good. It'll be interesting to watch how the community comes around on it.

And we do have the GitHub and Discord, uh, URLs. We'll, we'll include those in the notes. Thanks for being on Textron tv, man.

Good luck. Thank You very much, Alan. It was my pleasure.

Thank you. Manos Kois, uh, CEO co-founder Uni here on Text Drunk tv. We'll take a break.

We'll be right back.