#240 Manos Koukoumidis: Why The Future of AI is Open-Source

This episode is brought to you by Sonar, the creators of SonarQube Server, Cloud, IDE, and the open source Community Build. Sonar unlocks actionable code intelligence, helping to redefine the software development lifecycle by use of AI and AI agentic systems, to continuously improve quality and security while reducing developer toil. By analyzing all code, regardless of who writes it—your internal team or genAI—Sonar enables more secure, reliable, and maintainable software. Join the over 7 million developers from organizations like the DoD, Microsoft, NASA, MasterCard, Siemens, and T-Mobile, who use Sonar. Visit http://sonarsource.com/eyeonai to try SonarQube for free today. ———————————————————————————————————————— The Future of AI is Open-Source | Manos Koukoumidis on UMI & The AI Revolution Is closed AI holding back innovation? In this episode, Manos Koukoumidis, CEO of Oumi https://www.oumi.ai/, makes the case for why the future of AI must be open-source. OUMI (Open Universal Machine Intelligence) is redefining how AI is built—offering fully open models, open data, and open collaboration to make AI development more transparent, accessible, and community-driven. Big Tech has dominated AI, but UMI is challenging the status quo by creating a platform where anyone can train, fine-tune, and deploy AI models with just a few commands. Could this be the Linux moment for AI? What You'll Learn in This Episode: Why open-source AI is the only sustainable path forward The difference between "open-source" AI and true open AI How OUMI enables researchers and enterprises to build better AI models Why Big Tech's closed AI systems are losing their competitive edge The impact of open AI on healthcare, science, and enterprise innovation The future of AI models—will proprietary AI survive? The AI revolution is happening—and it's open-source. If you care about the future of AI, innovation, and ethical tech development, this episode is a must-watch. ———————————————————————————————————————— This episode is sponsored by Thuma. Thuma is a modern design company that specializes in timeless home essentials that are mindfully made with premium materials and intentional details. To get $100 towards your first bed purchase, go to http://thuma.co/eyeonai ———————————————————————————————————————— (00:00) The True Meaning of Open-Source AI (02:15) The Open vs. Closed AI Debate (07:54) Why Open AI Models Are Safer (10:34) Defining Open Data (13:21)Beating GPT-4-O with an Open AI Model (16:36) Open AI in Healthcare (19:31) Why Open Models Will Dominate (23:07) How OUMI Makes AI Training Fully Accessible & Reproducible (28:44) UMI's Collaboration with Universities (32:29) The Shift Toward Open A (36:41) Can We Build Truly Open AI Models from Scratch? (40:20) The Role of Open AI in Eliminating Bias (45:02) Will Open AI Replace Proprietary AI Models? (50:19) How OUMI Works (54:44) The Open AI Revolution Has Begun

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Speaker 00

  For something to be open source, it needs to be open data, open code, and open models or open weights. This means that it should be possible for somebody to have the ingredients and the recipe to be able to reproduce the model. Yeah, it's good if you're giving them the model, the weights, but also they should be able to reproduce it. I think this is a very, very good definition. However, what we say is that we want to have open code, open data, open code, open models. And what we say open collaboration, what this means by this, which may be a little bit less clear, is that something that is open source by the OSI definition, it's not sufficient. If, yeah, the code, everything, it's there, but it's too hard for anybody else to reproduce and build upon. Because then it is the whole purpose of open source being a community effort where it's

• 0:44Speaker 00

  easy for people to build upon and extend somebody else's work.

• 0:48Craig S. Smith?

  AI coding assistance like GitHub Copilot, Google Gemini Code Assist, and Amazon Q Developer have quickly become essential tools for developers. They generate code with remarkable efficiency, significantly boosting developer productivity. But the widespread use of AI-generated code brings its own set of challenges. Bugs, vulnerabilities, and suboptimal code can inadvertently enter production, leading to issues with maintainability, stability, even costly outages. How can organizations minimize disruption and risk while maximizing productivity and

• 1:32Craig S. Smith?

  innovation with these AI tools? With Sonar, the code quality and code security leader, organizations can amplify developer productivity in concert with AI assistance. They can improve developer experience with streamlined workflows and prevent quality and security issues in all code from reaching production. Sonar's AI code assurance feature found in their Sonar Cube solution offers a thorough validation process for AI-generated code, including the

• 2:10Craig S. Smith?

  First in its industry to automatically detect and review AI-generated code from GitHub Copilot. The AI code assurance workflow helps to build trust in AI-generated code, assuring companies that proper due diligence has been performed and code is ready for production. Their AI code fix feature speeds up productivity and issue resolution with instant AI-generated fixes for issues discovered by their code analysis that developers can review and apply directly within their workflow. Sonar is helping redefine the software development lifecycle

• 2:54Craig S. Smith?

  by use of AI and AI agentic systems with tools that streamline the verification of AI code and provide guided remediation of issues, supercharging developers to build better applications faster. Join over seven million developers from organizations like the DOD, Microsoft, NASA, and Mastercard who use Sonar. Visit sonarsource.com/eyeonai to request a free demo and see how SonarCube can help your team with machine-generated software development.

• 3:36Craig S. Smith?

  That's sonarsource, S-O-N-A-R-S-O-U-R-C-E dot com slash eyeonai. That's E-Y-E-O-N-A-I, all run together. And request a free demo. Sonarsource.com/eyeonai.

• 3:58Speaker 00

  Very nice to be here, Craig. I'm Manos Kukumiris, I'm the CEO of Umi. Until about nine months ago, I was at Google Cloud. I was supporting the science and engineering for all the natural language AI services. I built also some of the multimodal ones at Cloud AI at Google. And about three months before GPT was coming out, I was bootstrapping the Efforts to productionize CloudPalm and uh drove this effort almost until the general availability launch in May of twenty three. Before it was called DeepMind and uh sorry called Gemini and moved to DeepMind. Uh before that, I spent some time at a startup. I was at Meta working on conversational AI. Uh also before that at Microsoft, where I was building something like JGPT in two thousand and sixteen.

• 4:41Speaker 00

  A multi-model, open-ended chatbot, even build what now we call embedding retrieval and RAG back then. And before that, I was doing a PhD. I spent my first three years at Princeton and the last two at MIT. And before that, I was doing my undergrad in Grace.

• 4:57Craig S. Smith?

  Okay, uh and UMI, what does UMI mean?

• 5:02Speaker 00

  Uh UMI stands for Open Universal Machine Intelligence. Uh we wanted to pick a name for the effort that for one clearly denotes that it's open. We want to make sure that this is very clear. Uh and also we use the term uh appropriately, unlike perhaps what's happening in other cases. Uh and uh we didn't want to say AI because we think it's a little bit overused term. Because where everything started was machine learning, uh machine intelligence. So that's why we call this Open Universal Machine Intelligence. The universal means is we wanted something that open machine intelligence that is democratized, that people can use everywhere and anywhere.

• 5:45Speaker 00

  And that's why we ended up with the name.

• 5:46Craig S. Smith?

  Yeah, and and UMI is a is a platform, right? It's uh uh so I I mean, I I was interested in talking to you because open source. Uh there's been a long-running debate uh in um in the artificial intelligence world, certainly. Uh in the generative AI space about open source versus closed source. There were a lot of hand-waving in the beginning about, my God, what if this stuff is open source then? All the bad actors in the world are gonna have the most powerful AIs.

• 6:27Craig S. Smith?

  People like yourself and Yan La Kun argued that open source is the way to go. And that if anything, having more eyes on the model creates a safer model.

• 6:41Craig S. Smith?

  And since certainly, uh, I mean, there's been a growth in open source, you know, starting with Lama. By Meta and then Mistral with its models and then DeepSeek sort of came along and woke people up. Uh, if they weren't paying attention to China, that open source is a big part of China's AI strategy. Uh, so I wanted to hear from your perspective, just first generally. Uh, how you feel about the open source, closed source debate?

• 7:17Craig S. Smith?

  And then my second question is, open source is a an overused and under defined term. And what exactly does it mean? There's open weight, there's open code, there's open training data. Uh, and and no one seems uh to agree on what open source really means. So, uh, let's start with just the open source closed uh model debate. How, where do you stand on that and and how do you see it developing?

• 7:54Speaker 00

  Yeah, I agree. Craig, if that's okay with you, how about we start from the definition? Just make sure we're on the same page about when I say about open source, what do I mean about open source?

• 8:02Craig S. Smith?

  Great.

• 8:02Speaker 00

  Once it's clearer, you know, when I argue that that's the way forward, you know, it's clear what exactly I mean by this. So, um, as you mentioned before, there's a lot of efforts, quite often, many of them uh being dubbed called open source when they're not really.

• 8:18Speaker 00

  I think the OSI, the Open Source Initiative definition for Open Source AI, it's pretty on point. It says for something to be open source, it needs to be open data, open code, and open models or open weights. This means that it should be possible for somebody to have the ingredients and the recipe to be able to reproduce the model. Yeah, it's good if you're giving them the model, the weights, but also they should be able to reproduce it. I think this is a very, very good definition. However, what we say is that we want to have open code, open data, open code, open models. And when we say open collaboration, what this means by this, which may be a little bit less clear, is that something that is open source

• 9:01Speaker 00

  by the OSI definition, it's not sufficient. If, yeah, the code, everything, it's there, but it's too hard for anybody else to reproduce and build upon. Because then it is the whole purpose of open source being a community effort where it's easy for people to build upon and extend somebody else's work. So that's what I mean by open collaboration that for one. We have the principle that any of these AI technologies that we also are building, that are easy for anybody else to use in a fully record, sorry, very easy for somebody to reuse, because what somebody else did is fully recordable, and the platform has developer ease of use as a first principle. And also to extend a little bit also the collaboration

• 9:42Speaker 00

  aspect is that we already have and we plan to have even broader open collaboration efforts where we'd say, you know what, here's the goal, the problem we plan to solve, anybody in the community is invited to help us make it better. So this can be more inclusively developed. It's almost similar to what was happening in some ways with Linux compared to Unix, where it was collaboratively advanced. But again, to do this, it needs to be open code, open data, open models, but also in a way that's easy for everybody else to reuse, build upon, and contribute. So that's our definition of open source. We think it should be nothing less, because anything less is doing a disservice to the community and not helping us as humanity advance this technology in the best way possible. Let's now move to the debate, you know, why is this way possible?

• 10:24Craig S. Smith?

  Actually, before you do, open data is the one that's a little confusing to me. Open model means the source code is published. Open weights mean the weights are published.

• 10:37Craig S. Smith?

  I at the the models I've looked at, people describe the source material for the data, but no one that I've seen really has, you know, a database that you can look into and see exactly what data is in there. And I presume because there's a lot of concern about

• 11:04Craig S. Smith?

  copyright liability and that sort of thing. But but can you talk about what open data means? And are there any models, quote unquote, open source models today that have truly open data?

• 11:19Speaker 00

  Yeah, yeah, very good question. So, you know, for an open data means that they're going to be used analogy of, you know, baking a cake, you need to have the ingredients, not just the recipe or the oven, but you also need to have the ingredients themselves to to to build it. Now, in terms of, as you mentioned before, quite a few companies, often they may say we have an open source model, we can open source the model, but usually they only mean the weights, not the data, not even the code. There is very few truly open source efforts right now. And it's mostly by universities or nonprofits, like for example, AI too. Everything else, it's even though they may say they're open source, they're not really open source, which means it's very hard for anybody else to

• 12:04Speaker 00

  reproduce and build upon it.

• 12:07Craig S. Smith?

  Okay, and somebody like DeepSeek, who claimed when they came out that they're fully open source, including the data, they described the data. Is is that helpful at all? The, you know?

• 12:25Speaker 00

  So I'm looking forward to all the announcements that DeepSeek is going to be making this week. They say it's going to be announcing one more thing every day. But as it stands right now, the models they have are like Arwan and others, they are open weights. They have released some code, like yesterday, they released some inference code that helps do inference more efficiently. But still, nobody can fully reproduce their work. They had a report that detailed some of the steps. But if anything, that left a lot of questions across the industry about how exactly did they do it? And what exactly data did they use? And how exactly did they produce the data? So yeah, nobody can still reproduce, for example, Deepsec R1, because they're lacking the actual data.

• 13:08Craig S. Smith?

  Yeah.

• 13:08Speaker 00

  And some of the deals or the outcomes that it's not clear if everything is in the report.

• 13:12Craig S. Smith?

  Yeah. But but you guys are promoting full open data. And what does that mean?

• 13:20Speaker 00

  Exactly. So one good example is a an agentic model that we released just about 10 days ago on a Friday. It was called CoALM, C-O-A-L-M.

• 13:33Speaker 00

  And actually, it's an agentic conversational, an agentic model that even beats GPT-4O in some of the Berkeley benchmarks in the leaderboard. And to give an idea for this, any of the data that we curated in collaboration, actually, let's say that it was U I U C, Emre and Gökhan and Delectur from U I U C that was leading this effort. Any of the data they generated and any of the training that we also helped them perform and scale up to 70B or even 405B, which I'm not sure if any other university has trained the 405B model so far.

• 14:07Speaker 00

  All of that is open, which means right now somebody can go to the UMI repository and see where we have the the pointer to Hikinface, where we have uploaded the data sets. And it takes right now two commands for somebody to reproduce. They can do umi install to install the solution. And then umi train with the name of the configuration file if they want the the smaller 8B models or the 70B or the 405B. That's what takes two commands. And as I tell people, it shouldn't be any harder than that for anybody to reproduce somebody else's work end to end with the data, the code and everything. Of course, the model itself is based on Llama, but at least all the post-training and everything else is fully open source.

• 14:43Craig S. Smith?

  Yeah. And then presumably there are, even if it's fully open source, there are licenses attached. And is there a concern that if you fully open source in that regard, open data, that people will just take the model and not license it?

• 15:07Speaker 00

  Yeah, so there is that concern, but at the same time, this should be almost, you know, also if I understand the question correctly, you know, by design, the desired benefits of open source, which is, you're saying, you know what, I'm gonna help contribute to the common efforts as research community, as humanity, to make this technology better. And I'm gonna tell somebody else all the critical ingredients, so that they can also help me make it better. I think one good example here is, I was seeing this just a couple, maybe weeks ago. It was the Human Genome Project. That it was, I don't remember how many, over a hundred institutions, thousands of researchers that participated.

• 15:49Speaker 00

  And somebody could have said, like, the same thing now with frontier foundation models. Hey, we're gonna give it a black box, just so that we can monetize freely. Or we're gonna make it open and share the knowledge with everybody. And, I mean, I don't know what people stand in terms of politics, you know, with, I think it was Bill Clinton. But regardless of politics, you know, he made a statement that resonated with me, that said that, I think he said it would be virtual criminal if they were to withhold all these critical discoveries that are almost about life and death. Just so that, as opposed to advancing humanity and helping everybody advance it, instead they close in a box so that some companies can monetize out of it. I think the same thing for foundation models.

• 16:31Speaker 00

  The link may be less intuitive to life and death, but actually it's there. Where using foundation models, not just for technology and the industry, but for all the sciences, from material science to climate science to healthcare. I was actually, maybe not to disclose specific names here, in a discussion with quite a few healthcare leaders from some of the biggest institutions we have in the US. And what he was telling me, that they use right now these models, they adapt them to detect, I think, polyps, I think, colonoscopy, if I'm correct. Or some of these procedures. And what they said for these procedures is that,

• 17:12Speaker 00

  soon enough, it's gonna be malpractice if somebody doesn't use AI, not the other way around. Because AI was like five times, if I can recall, better than humans. Not 20% or 30%, but five times or something better. Which means, again, and then you get almost the question about, it's almost like a matter of life and death. I mean, it's, these are critical technologies, it's a disservice to humanity if they're not open. And they're not advanced. And even if you keep those open, there's still enough opportunities for many enterprises, including ourselves, to monetize. So that's, that's, you know, that's why I think things should be. And we're going to also be touching on the earlier debate.

• 17:50Craig S. Smith?

  Yeah. Okay. And then more generally, how do you think this is, this is going to develop? Do you think, and then we'll talk about UMI, but the, the, do you think that open data will increase? What will happen to the closed source operators? I mean, already DeepSeek has put pressure on them with its pricing.

• 18:14Speaker 00

  Yeah, yeah. I think eventually things should be more open because if anything, if anybody, even close providers are using data that they shouldn't be using, there should be a little bit more transparency over that or make sure that everybody's being a good player. So if anything, eventually things should be more open. But yeah, for pre-training, massive amounts of data, it's the whole web, sometimes even accounting for everything, maybe practically hard in many ways to be very thorough. I think for post-training, a lot of the research, for example, that we recently did or we're doing with the university, it's, it's, it's a lot easier to do. And yeah, I think now your second part of your question was, okay,

• 18:58Speaker 00

  what will happen in the future, right? In terms of open versus closed and the pressure that open models are putting. That was actually our strong conviction about a year ago, or more than that, when we were starting this effort. So you know what? It's clear that the future of enterprise AI and beyond just enterprises is going to be open. The same reasons why Linux became better than Unix. And I could elaborate a little bit on the history and the parallels that are happening with AI. And once this start happening, which if anything, maybe with DeepSeek could accelerate a little bit more, even perhaps a little bit more than what I was expecting. And if anything, then all the pressure is going to be a lot of pressure to the closed.

• 19:42Speaker 00

  Source companies and they're gonna start feeling a lot of pressure because they I mean arguably you want people to justify the greatly inflated valuations anymore or for them to be able to continue raising at this high valuation so they can keep executing on the same, in my opinion, bad strategy because it will become clearer and clearer to more people and their investors that hey this is just doesn't make sense. There is a better way to do this. If anything, history has taught us that there's better way usually for this complex technologies to be developed and they may stop funding them or be less interested.

• 20:19Craig S. Smith?

  Yeah. Okay, well, let's look at Oumi. So tell us about the genesis and exactly what Oumi's platform does and how it's different from last time we spoke from Hugging Face or Lightning Studios, one I'm familiar with, the other platforms out there.

• 20:44Speaker 00

  Yeah. So the reason why we started Oumi is actually somebody, sometimes people ask me, you know, what was the one thing that made you start Oumi? I say, no, actually it was not one thing. I not just want to do the right thing by humanity, but also I'm a pragmatic person. I want to build the conviction that something makes sense before I start doing it. But in terms of the right thing, that goes a little bit back to the debate, right, that we're discussing before. I, it was very clear in my mind that for this generic technology should be like electricity, should be a common utility. It's gonna be powering everything. For sciences industry to advance, it needs to be an impeded access and also to advance it faster, safer, and most efficiently. This can only be done if you have all hands

• 21:28Speaker 00

  on deck, all, you know, all eyeballs on this technology helping to make it safer and advance it. So philosophically, you know, it was very clear that, you know what, this is the right thing for humanity. But then increasingly, I kept thinking, okay, but this may be the right technology, but will it happen in this way or will unfortunately go in a different path just because, you know, it's not practical for open source AI to succeed. And then it was a sequence of things that made me realize, actually, you know what? Actually, this could work, and not only it could work, it's actually the most likely scenario, the most likely outcome. The first one was, increasingly, I was realizing how, especially for post-training, and there's so many tasks, so many capabilities that these foundation models have across so

• 22:13Speaker 00

  many modalities that we are already working on and many more that we haven't worked on yet, that it's not necessarily bottlenecked by the amount of compute you have, but they are a matter of people, creativity, and people that you have to go and innovate in all these different directions and across all these different tasks and modalities to advance them, which means that we, an open community that is not just 1,000, 2,000, 3,000 people, but orders of mind larger, can do a much better job if they have the right tools, which is what we're helping them, to me, to give them. At the same time, besides the fact that, again, an open community could do better than a closed community, I was increasingly realizing that the current status quo is only serving a few companies.

• 22:59Speaker 00

  Nobody else wants it. So as I tell people, unless you are an aspiring AI oligarch, and the handful ones that they are, they know who they are, nobody else wants this. Because imagine, for example, the outcome where, let's say, you know, a specific closed source model, like Gemini, or OpenAI's models dominate, everybody's gonna go use those specific models on the very specific cloud provider, on the preferred accelerators that this cloud provider has, which means other cloud providers, they won't have a competitive model to serve, other accelerator providers, including Nvidia, AMD, all the major ones, they may be out of the loop because they know that the closed model providers, they may have their own preferred accelerators.

• 23:43Speaker 00

  Where because the model is closed, they can more freely optimize across software and hardware. Which means they can offer a better experience end to end. So anyway, and besides the cloud providers and accelerated providers, then I started realizing that actually it's not just them, even consumer companies don't want AI, closed AI, closed source AI to dominate. Mark Zuckerberg has said it publicly recently quite a few times that even for a company of the size of Meta, it would be problematic if AI one day is controlled by a single company, let's say OpenAI or Google, and then they are at the mercy of getting permission from them to use the AI they need for their consumer products, and the AI is going to be everywhere, right? So this is extremely critical. So given all that, I was realizing that there's so many tailwinds.

• 24:28Speaker 00

  Again, most of the players in the ecosystem, they don't want to succeed. And if you help channel the efforts across the community and across them in a more meaningful way, at the end, what is the most likely outcome? That open source would succeed? Or closed source that is developed in a silo by its company individually? To solve all the economic and human costs to make it happen. So for me, it's becoming clear again, as history has taught us with Linux versus Unix, that the open source will be actually the most viable path. And I think this is becoming increasingly clear to more and more people, and that's why we started Umi about a year ago. It was a combination of, you know what? This is the right thing for enterprise to succeed. This is the right thing for humanity, and at the same time, it's possible. So it just makes sense to go build it.

• 25:12Craig S. Smith?

  Yeah.

• 25:14Craig S. Smith?

  Well, tell us how Umi works. I mean, how does somebody use the platform? Create an oasis with Thuma, a modern design company that specializes in furniture and home goods. By stripping away everything but the essential, Thuma makes elevated beds with premium materials and intentional details. I'm in the process of reorganizing my house, and I'm giving Thuma a serious look for help in renovating and redesigning. Thuma combines the perfect balance of form, craftsmanship, and functionality. With over seventeen thousand five-star reviews,

• 25:58Craig S. Smith?

  the Thuma bed collection is proof that simplicity is the truest form of sophistication. Using the technique of Japanese joinery, pieces are crafted from solid wood and precision cut for a silent, stable foundation. With clean lines, subtle curves, and minimalist style, the Thuma bed collection is available in four signature finishes to match any design aesthetic. Headboard upgrades are available for customization as desired. To get a hundred dollars toward your first bed purchase,

• 26:39Craig S. Smith?

  go to thuma, that's T-H-U-M-A dot C-O slash IONAI. IONAI all run together, E-Y-E-O-N-A-I. So for a hundred dollars off your first purchase, go to thuma dot C-O slash IONAI. That's T-H-U-M-A dot C-O slash IONAI to receive a hundred dollars off your first bed purchase.

• 27:13Speaker 00

  Yeah, so Umi, as we launched it about three and a half weeks ago, so it's very, very recent. It's a fully open source platform on GitHub. It provides all the capabilities that somebody needs in enterprise or academia to advance foundation models, anything from pre-training, all the different post-training techniques, full fine-tuning, parameter percent fine-tuning, RL techniques, evaluation with all the most common bed marks and a growing list of them, data curation, auto-evaluation with LLM judges, and many more. And it is all the tools that somebody needs to do research in academia. They say you want to develop the next TIPSEC RL, the next TIPSEC R1. You should be able to have all these steps in a single platform

• 27:57Speaker 00

  so that your whole recipe can be recorded fully. It's not you combine this with that and there's dark knowledge and scripts and... So anything should be recorded in the same platform. First of all, for convenience, you have all the tools you need. The whole recipe can be fully recorded, which means it's easy for somebody else to go to produce it, cite your paper and continue advancing it. It shouldn't be any harder. So that was part of the goal. Now, besides providing all these capabilities, you know, we support both text and multimodal models, over hundred open models, pretty much all the ones that matter right now, that are relevant for enterprises or researchers, they're supported, from hundreds of million parameters to all the way four or five B Lama that somebody can train.

• 28:40Speaker 00

  And also as a solution, we wanted to make it fully accessible, which means that, and seamlessly accessible, which means we want to make sure that somebody, whether they're getting their compute, let's say they have a, just a MacBook with a couple GPUs, or they have, they're getting their compute from some cloud provider, W A H C P, Azure, Lambda, Rampo, Together, or anywhere else, or they have a big HPC that they can still run the same experiments, of course, assuming they have the necessary compute, but all the change is that some simple deployment configuration says, how many GPUs do you have and how many do you want to use? But it is that the same recipe should be reproducible easy by somebody, regardless of where they get their compute.

• 29:25Speaker 00

  So that's kind of what the platform provides right now. Again, all the capabilities that somebody needs to build foundational models or adapt them to specific tasks and domain, whether it's pre-training or post-training, even evaluating, data curation, all text and multimodal models, and also running across any platform and across any scale. Which, as I mentioned before, I'm not sure how many universities have been able to produce models at this scale. And that's why those researchers from E I U C, and we had a long relation with them, they said, you know what? We really need your help to scale them up, because we have been using other technologies, but it's just impossible for us to do this. And yeah, the good thing was very easy, and it's also very easy for anybody else now to reproduce.

• 30:07Speaker 00

  So this is, this is the, the fully open source solution. We're starting to develop also enterprise solutions on top of that. Happy to discuss about it as well.

• 30:16Craig S. Smith?

  Yeah, on building models, how many,

• 30:22Craig S. Smith?

  you know, on hugging face, there was, I don't know how many models are now transformer-based models.

• 30:31Craig S. Smith?

  How many people are really building models from scratch? I mean, they, they write their first line of code and then, you know, code for however long it takes and then

• 30:48Craig S. Smith?

  train that on, on data that they've collected versus how many people are taking Llama or some other open source and tweaking it or forking it or whatever they're doing, fine tuning.

• 31:08Speaker 00

  Yeah, so the vast majority of the people are getting an existing open model because to get to that point takes significant cost. It doesn't make sense for them to produce it. They get an existing open model and they continue post-training it over different techniques to continue making it better for specific research or if they're enterprise for a specific problem they're trying to solve. Very few are working on pre-training or training from scratch, actually to be more precise, to training from scratch those foundational models. But the good news is that there is huge innovation that's still required, not just on the pre-training, but across all the other stages. And as I was mentioning before, an open community, especially when it comes to post-training or finding creative ways to get existing open models

• 31:53Speaker 00

  after somebody like Meera has incurred all these massive costs to train them and make them better across all these different domains and modalities and capabilities. And with creative approaches like Deep6, GRPO and any new ones that they're gonna invent, that's something that you don't need massive compute and there's huge opportunities, us as a research community, as humanity, to continue to advance them. Sure, it would be great if it was easy for more people to do pre-training, but even on the latter stage where it's more of a, how many people, how many creative minds do you have to advance this? And less about the compute constraints. So there's huge opportunities again to advance even. On getting an existing model and curating or finding creative ways to make it better.

• 32:38Speaker 00

  And actually, just something to add to my previous, the previous question I forgot to mention is that besides providing the platform itself, we already have also open collaboration with many universities and we plan to launch even broader collaborations where we say, you know what? Here's the problem that we as a community makes sense to solve. Let's all open and collaborate and make it better.

• 32:59Craig S. Smith?

  And does the platform require open code, open

• 33:09Craig S. Smith?

  data as well as open weights?

• 33:12Speaker 00

  Yes, exactly. So for any of these collaborations that we currently have to advance these foundation models, we say, you know what? Everything needs to be open. And if it's built on the OMI platform, for one, it's easy for you to do your experiments end to end and then record the full recipe. So that anybody else can easily reuse your data, your code, everything and reproduce this work and then make it better.

• 33:35Craig S. Smith?

  Yeah.

• 33:37Speaker 00

  Yeah, it's, it's, it's, it's, um, it's an absolute necessary requirement. You know, if you want open source to thrive, that's how we need to function. Anything less will just impede the success of open source.

• 33:50Craig S. Smith?

  Yeah. And, and you think that people will, will start building truly open source models if they have, have this, uh, this platform or, or, yeah.

• 34:07Speaker 00

  Yeah, I think for pre-training, it may become, my hope is eventually we'll be able to do more and more open source, even though now the truly open source on the pre-training states are very, very limited. My hope is that in the future we may do more and more there as well. Uh, but for post-training and or at least using existing pre-trained models and with different, continue to pre-train them or post-train them in different ways. That is already happening. There's huge appetite from researchers in academia to do this. And by the way, to your earlier question, that's how even we started with OMI. This started with a Sunday discussion with Ruslan Salegudino from CMU where he told me, you know what? Uh, we are.

• 34:50Speaker 00

  I mean, he didn't say that, but they are the top researchers, among the top researchers in academia. And then he said that, uh, that was the top researchers, it's my assessment. And he said, you know, even for us, it's very hard to do this type of research. It takes a lot of effort for my students to put together all the pieces that they need to do this type of research. And then, uh, none of them has scaled and done multi-node distributed training to scale up and train bigger models. But it's too hard for any student to, to do this. So if you build a platform, we'll support you. And then he introduced me also to Amit Alvarcar, who is also from CMU. And actually, in his words, he said, uh, we are an untapped resource. We want to contribute, but those labs don't make it easy for us, or possible to be precise.

• 35:36Speaker 00

  Uh, and also, you know, if you had the platform we described, that would be really helpful. So yeah, please build it, we're gonna help you. It was the same response from many other academics. So that's why we started to build this, so it's easy for people to do the research, build upon it, and then give the, as we say, open source, the kind of Linux moment that it was missing.

• 35:52Craig S. Smith?

  Yeah, uh, one question I've, I've had for a long time. Uh, if you take an open model. Well, let's, let's take Llama as an example, it's been pre-trained. Right? So if, if, uh, are there models out there?

• 36:15Craig S. Smith?

  The code out, the source code is out there that you can pre, pre-train from scratch?

• 36:21Speaker 00

  Yes, there, there is a few, very few examples. The, one of the most notable is the efforts from AI2. Where, for example, for the OMO paper, they said, okay, here's the data that we used. Uh, and here's, uh, the code, and then the resulting model. So we need more of that. Uh, and my take was this is, you know, but such models should not be developed by almost a single organization, or small collaborations should be a much more open collaboration across the community. And that's what we're trying to, to help do. So I, it's amazing to have such efforts from Meta or AI2. But there is, uh, was a missing tool, a missing capability, the Linux of AI, as we call it, that was missing for the community. And that's what we hope to help provide.

• 37:07Speaker 00

  And as a platform, it's something that can work with the models developed by AI2, by Meta, or any new models that the community develops. Uh, any, any of the best tools. So that's why, you know, we're very. Grateful and excited for all these efforts that all these other organizations are doing, including Meta, even though it's not fully open source, open weights, it's still a huge asset for the community.

• 37:26Craig S. Smith?

  Yeah, so if someone, because I talk to people all the time who say they're building on an open source model and they're fine tuning it for one particular use case or another. But if you're doing that, if there are biases or frankly, bad data in the original training data, it's baked into the model, right? It's encoded in the weights. So it seems like that's a disadvantage.

• 38:08Craig S. Smith?

  You're not able to fully control how the model is going to react. Or is it that taking source code, as you said, some of these models that, and then pre-training it yourself is just too expensive?

• 38:29Speaker 00

  Yeah, yeah, yeah. So that's a very good question, Craig. So what is definitely clear is that if one had access to all the data and the procedure in which a model was pre-trained, it would be easier for them to reason about why the model is behaving in a certain way or what could I change as a researcher to make it better? This is something actually that some very notable researchers have mentioned in the presentations that I attend in the past. Say, you know, we want to be able to do this research, but you know, if we had access, if we knew these details, it would be a lot easier for us. So yeah, that's definitely unfortunate. That being said, if a model like, for example, what Meta does with the Lama models, they're open and say,

• 39:14Speaker 00

  and says, you know what? Here, as a community, you can get it, deploy it wherever you want, test it however you want. I'm not going to control or limit you if you issue requests to my API that limit you in certain ways. Or because maybe you don't want to reverse engineer or do whatever you may want to, you're trying to do with the model. Then it still leaves some good amount of flexibility for the community to identify the issues. And say, you know what, I think it's demonstrating this issue. And here, and here's how by continuing to pre-train it or to be usually with some alignment technique, mostly as a post-training technique, I can align the model to do less of that bias

• 39:57Speaker 00

  and behave more in the expected way. So you could still rectify and fix some of this in the post-training. But yeah, to do fundamental research that helps us understand better how these big models behave. I mean, yeah, it would be really helpful to have more access to the whole recipe. That would help to do more foundational research in how they behave.

• 40:21Craig S. Smith?

  Yeah.

• 40:23Craig S. Smith?

  I was talking to somebody recently about a project that they're working on. It's a fascinating idea is to take a model, an open source model, and train it on all the original religious texts in the world. But if you do that with a pre-trained model, if you do that in the fine tuning or with RAG, you've, the, the pre-training model already has a lot of that text in it. And depending, and you know, I mean, this is something

• 41:08Craig S. Smith?

  that bothers me about, about these public models that have been trained on the internet. As the internet fills up with marketing material, for example, you know, if you ask about a company, the answer you get back is not objective. It's skewed by the marketing, by the marketing content that's, that's in the, and I would guess that companies with enough budget realize this and are pumping out a marketing content onto the internet so it gets.

• 41:47Craig S. Smith?

  vacuumed up and and biases the model in their favor. And and so if you're working with a pre-trained model on something like this religious text, presumably, there's going to be some bias in the underlying model depending on on the distribution and the training data. So how do you how do you adjust for that? Or or is is something like all the religious texts in the world, is that enough of a corpus to pre-train a model from scratch? Or or is it? Yeah.

• 42:25Speaker 00

  Yeah, so I would conjecture here that likely the all religious content may be may not be enough, or maybe a lot of, I guess, knowledge, like science and many others, that if you want to have a very comprehensive model, you need to expose it to all the knowledge, not just on our religion.

• 42:42Speaker 00

  That being said, you know, I think that's also one of the benefits of open source is that.

• 42:50Speaker 00

  You know, it's hard to say, you know, what is the right religion, or what is the right way to respond about a certain religion, you know, do you take a position or do you not? And different nations, religions should have the control if they desire, you know, to develop their own model, let's say, that answers questions about their own religion. And it shouldn't be a single owner, like I say, one specific big tech company that decides what everybody's religion or what's the, you know, the right answer to many of these topics. So that's why, and arguably, it may be even hard for them, because there may be even some small religions that are not represented on the web very well. That's why, for the sake of inclusion, the best way is if those technologies are open

• 43:35Speaker 00

  and you democratize and make it easy for anybody to build the tools for their own language, dialects, for their own religions, for their own, you know, a more specific context. I think Jan Le Cun had mentioned in the past, actually, quite a few months, maybe it may be a year already, that That's why he thought also open source is better than closed source, which is like the equivalent of Wikipedia, that the best way to capture all of humanity's knowledge and inclusively, without having single person decide what is the right thing and then leaving out everybody, everything else that he doesn't consider important enough, is to do it in a Wikipedia style, to do it in an open way. So I think it was a very, very thoughtful comment he had made.

• 44:19Speaker 00

  So anyway, yeah, I see it the same way. And again, that's going back to me, that's what we call the democratize, should be easy for anybody, if they have access to their own data, to continue pre-train, post-train, get the model that is desired for their domain. Because yeah, for some of these things, it shouldn't be a single tech company deciding, you know, for everybody else, what the right answer should be.

• 44:42Craig S. Smith?

  Yeah. Is it possible to to train a foundation model to purely on the linguistic structure of language, without any knowledge? So that then you can add the knowledge in the fine-tuning stage?

• 45:03Speaker 00

  Yeah. You could, I mean, there's been quite a few discussions and research lately that, you know, what would be best ways to, I don't know, maybe your question is getting that way, to decouple the reasoning of a model from the knowledge. Because model that can reason, but is liking the knowledge could be smaller, more compressed. And then the knowledge could be external. And then if you're smart enough, you're able to reason about stuff, but then you can access the knowledge when you need it. You could have smaller models that can tap into specific model, sorry, specific knowledge, you know, when desired. So to some extent, that is possible, and arguably, you mentioned before, RAG, that's almost what people do with RAG. I mean, they,

• 45:46Speaker 00

  you train a model that has some basic understanding and knowledge, but when it needs more specific knowledge, it knows how to understand text, it knows how to understand concepts, and can access this external knowledge and bring it in and use it to respond to something, almost pretending to be an expert in this specific domain. To be honest, it's not as effective as fine-tuning, that's something sometimes people don't understand, they think that RAG solves all problems. And what I tell them, you know, RAG is like having a high school student that has access to Google Search. You know, they can quickly go find information, process it. Yeah, they have some common knowledge, respond to you. But it's not the same. For example, let's say you are a legal company, an attorney, having somebody who has the experience

• 46:29Speaker 00

  and they're able to reason for how people should reason the specific discipline

• 46:36Speaker 00

  while still accessing the knowledge to respond. For example, if you're an attorney, you don't just respond based on some information, but I would argue again, hopefully not attorneys get offense here that usually they're risk averse. That's my understanding. Again, I hope I'm not ignorant. That they may respond in a way to help cover their clients across all corners or all cases. And there's some reasoning, there's some specific

• 47:02Speaker 00

  rules perhaps that go into their training about how they think about the information to craft the recommendation. That's why, and that's something that models can learn with pre-training or post-training. And that's why, for example, just RAG, just accessing knowledge may not be sufficient. Because like giving a high school student access to their book, he can read it, but doesn't mean that he has the, he has built the right intuition.

• 47:25Speaker 00

  In some cases, you need that to be able to give the best possible answers.

• 47:28Craig S. Smith?

  Yeah. And, and is anybody, well, this is getting off topic frankly.

• 47:36Craig S. Smith?

  But, but I'll ask this last question and then we'll get back to Umi. But

• 47:42Craig S. Smith?

  are there efforts

• 47:45Craig S. Smith?

  to train a foundation model more carefully with more highly curated data? Rather than just, you know, willy nilly scraping the internet?

• 48:00Speaker 00

  Yeah. Yeah, I was gonna, I was gonna say that even Meta, they, they have, even though, you know, they only have open way models, their report for Lama 3 was actually reasonably detailed and helpful. And for some, they make it very clear that they put a lot of effort in cleaning up the data on the web. But actually, one thing that may be not clear to many people is that when you say cleaning up, it doesn't mean that you necessarily need to remove all the bad data, like offensive data. Because for somebody to understand what is bad and what is good, see both the bad and the offensive and the good. So it stops them to see all aspects. But anyway, it's still, there's still a lot of careful data curation so that we can give all the most useful tokens, the most useful data to those models to be trained. So yeah, data curation is already happening, it's very important.

• 48:47Speaker 00

  And arguably, one of the best ways to continue improving these models is to be more careful and thoughtful, both about the type of data you expose them to, but also in what sequence, you know, something that's called curriculum training. How do you start training a model on some basic data? And then you, perhaps you move it to more harder domains? Or then you introduce later some capabilities? Like being able to understand longer context and things like that.

• 49:11Craig S. Smith?

  Yeah, okay, so Umi,

• 49:16Craig S. Smith?

  you, you know, a model builder can

• 49:22Craig S. Smith?

  go on to Umi and connect to data sets to, to train a model, either write code or, or, or, you know, pull code in from another model. Can you also connect to, you were talking about the cloud.

• 49:46Craig S. Smith?

  You can connect, is, is, is all of that in there? That then you can train right in Umi? Or, or do you have to export the code?

• 49:56Speaker 00

  Yeah, so, um, so, yeah, as you mentioned before, let's say somebody wants to train a new model, like the agentic model we released or some other model. Many of the most common data sets that are currently in hiking phase, we provide adapters so they can all be standardized and brought into the same right format. And existing recipes that should work very well already out of the box, so that you can almost run out with, again, like almost a single command, you have already installed Umi, reproduce training Llama on some common benchmarks of your own choosing, almost. You know, it's a single command, a single configuration file you create. So the the the basic functionality is also already very, very easy to do.

• 50:38Speaker 00

  Now, if you have some reasonably capable laptop, you could do some of that on your own compute. And then it just takes some changes to what we call a deployment configuration. Not writing code, not writing these things where you just say, here's my credentials, let's say, for GCP. And here's my credentials and here's how many GPUs I need that I have already created the VMs. Go run it there. So it doesn't take even any code to run it, just change the configuration for how

• 51:14Speaker 00

  UMI can deploy while running the commands on your laptop, they can deploy to the VMs you have allocated on GCP. So it's it's it's it's very easy. And again, I tell people, you know, it shouldn't be any harder than that. And again, if you have an HPC, again, it takes again a single command. You can run the UMI code on the HPC itself and you can deploy to hundreds, thousands of nodes, GPUs, if you have access to.

• 51:41Craig S. Smith?

  Can you talk about the, you said there was an UMI powered model that was published on Hugging Face recently? And outperformed some benchmarks. Can you talk about that? That model?

• 51:55Speaker 00

  Yeah, yeah. So there was a, you could say a gap in the existing models. And Embra, working with Delik and Gekan from YU, see like, you know what? It would be good if we have an agenting model that is very good at doing multi-terne conversations. And at the same time, is very good at tool use, which means with like simple terms, you know, you can call external services to access information and then use that to perform a task.

• 52:24Speaker 00

  And through research that they did and the data curation, they managed to create a recipe that it showed that actually you can develop a model that beats GPT-4O in the in many.

• 52:38Speaker 00

  Actually, it stacks higher. In some benchmarks, it beats GPT-4O. And in the leaderboard, it currently, I think it's above Mistral, Gemini, and many other models across the board, across all benchmarks. And in some, it's even better than GPT-4O. And I think that was actually the thing that I'm very excited about this because I still people, you know, I don't want to diminish this research, excellent research done by the research team. In some ways, it's a drop in the bucket. It's like all these little drops that everybody, one, everybody of us in the community needs to do and will do because there's so much interest

• 53:20Speaker 00

  to fill up the bucket of innovation, to make those open models better. And the good thing again is because they're already in the same standardized platform. For somebody else to combine Emre's work with somebody else's work, it's very easy. And then they can again go beyond it and make it better and better. And then the next person can reuse their work and make it one step further.

• 53:43Craig S. Smith?

  Yeah, that's exciting. And that model's been up there for how long?

• 53:48Speaker 00

  I think it's now 10 days.

• 53:51Craig S. Smith?

  And are people downloading it? Are there any metrics around how the community is using it?

• 53:59Speaker 00

  I think it's been already, at least in hiking phase, over 100 downloads, maybe many more. I haven't looked at it recently, to be honest. But overall, the reception was very good. One way how we know this is the day the model was published, we got quite a few people reach out to us and say, "Hey, I want to volunteer and help in Umi's research efforts. How can I do that?" And it was a very big optic, which means it resonated with a lot of people.

• 54:24Craig S. Smith?

  Yeah. And the licenses attached, does Umi offer a selection of licenses? How do you?

• 54:35Speaker 00

  Yeah, so quite often for a lot of research work that happens because people, they may use Llama, or they may, which has some limited constraints, but I think still doesn't block for the most part, research or enterprise use. Unless you're hitting a specific scale.

• 54:56Speaker 00

  And quite often, the researchers, when they produce databases or advanced research, sometimes they may use data sets that come from sources that restrict them. From them releasing their own models or improvements in a fully permissive licenses. But by far, our preference and what we tell to many people in the research community is that if you need to do that because that's the only way for you to advance the research, do it because it helps, still educates the community. But the best way is to release things with very permissive licenses. Because then not only fully enables the research community, but also the enterprise to use them. And then you can have a material impact to help enterprise, give them a better alternative compared to a black box. Uh, that, you know, you would get from all the other closed source providers.

• 55:39Craig S. Smith?

  Yeah.

• 55:39Speaker 00

  So yeah, permissive licenses are definitely preferred wherever possible.

• 55:42Craig S. Smith?

  I mean, before I ask the, the, the next question, is there anything I've missed? In talking about Umi, I mean, uh, that, that, that you're waiting for me to ask or that you want listeners to know?

• 55:57Speaker 00

  I think we covered a lot of ground. The only thing I would say is that uh, I'm very hopeful and optimistic that as a platform, it should help people do this work end to end. And also help other people build upon their work, which again, as I mentioned before, it's critical for open source succeed. Uh, but the best way, you know, for any of this to succeed, you know, for the benefit of all of us, for the benefit of the sciences, enterprises, humanity overall, uh, is if we all work together to make this happen. Uh, if people say, you know what, I know I'm supposed to doing my own small thing. I really want to call it something bigger. Pursue my own independent research. But also in a way that helps everybody else build upon it. And the whole community to move forward. That's the, that's the key thing. I mean, we all need to work together. That's the best way to make this work.

• 56:42Speaker 00

  And I do think that we can make it happen. It has happened more with many complex technologies like Linux and Unix. We can make it happen.

• 56:49Craig S. Smith?

  Yeah. Uh, the, is this only for, is Umi only for foundation models or can you?

• 56:56Craig S. Smith?

  Can you build other kinds of software, open source software?

• 57:00Speaker 00

  Yeah, great question. So currently it's mostly around foundation models. It's sometimes people they think about GNI, generative models. We plan to introduce capabilities, which we have internally, but we haven't introduced in the platform yet, to train classifiers and to perform more classic machine learning tasks. Because foundation models are great tools to even perform these more classic tasks. So we plan to do more and more of that. There's plans that we have to develop tools across the whole ecosystem. How can you better review and improve data or contribute to data improvements? There's all across the ecosystem that we plan to do

• 57:39Speaker 00

  more work to improve overall the tooling that you need across Antwen.

• 57:43Craig S. Smith?

  Yeah.

• 57:44Speaker 00

  And of course, contributions from the community are more than welcome. We want these tools that, if there's anything that helps the core community to advance foundation models for the benefit of all of us, we want this to be fully open and uninhibited.

• 57:57Craig S. Smith?

  Yeah, yeah, and I can see that increasingly there's, that's why I was asking about, are people building models from scratch? Because there really is a need for a completely open source, open data foundation model that's high performing, that that people can work with. So that, you know, there aren't all these questions about data bias or whatever. Where do you see the open source landscape in, say, five years?

• 58:38Craig S. Smith?

  And presumably, you're hoping that Umi will will play a big role in shaping that. But but do you do you see that there will be these fully open source foundation models? Meaning open data so that?

• 58:59Craig S. Smith?

  People can can work with them. And you, I mean, as you said,

• 59:06Craig S. Smith?

  and a lot of people have been saying there, there are going to be these foundation models all over the place, or at least applications built on foundation models. Do you see that that they're going to be a handful of of foundation models, open source foundation models that people work with? Or are there going to be thousands or tens of thousands of differentiated foundation models out there?

• 59:40Speaker 00

  Yeah, so I think the, I was gonna say the foundation of foundation models, that general foundation model, who's going to be the the starting point for many people to start building more domain specific or task specific models, that general foundation, the general foundation models, I think that will increasingly be open and I would for sure hope, but I would also expect, you know, not just wishful thinking, but I would expect that the majority of those use cases in the future are going to be powered by open models. I have seen that, I was expecting that trend over a year ago. I have seen it accelerating the last months, especially weeks. Deepsec had a big impact on this. Where more and more enterprises are reaching out, say, you know what?

• 1:00:21Speaker 00

  Six months ago, when we started, I wasn't sure if we want to use open models, but now I see what you were saying. I want to use more open models. The quality is almost there. Many of them are already telling us, the moment we customize them, they're way better than the models that comes out of the box from OpenAI or Google or anybody else or Anthropic. So it makes sense for us to use an open model and customize it and get better quality with a full privacy, security, flexibility, a lower cost. It's just a no brainer. Yeah, it takes a little bit of cost to train our own model, but it's not as hard as it was before. And that's why we built Ome. So given that, again, I would expect that the the the core, the general foundation models are increasingly open. If anything, the existing closed source AI, closed source model providers, they're gonna have, as we discussed before, increasing

• 1:01:07Speaker 00

  pressure and less and less innovation and less and less lead. And for them to remain relevant as a viable business, they may have to move more and more to add-on applications as opposed to Core differentiation on the base foundation model. The thing is easily that benefit is gonna diminish, actually it's already happening. Uh, and open is gonna be just a better alternative overall for most enterprises. Uh, so that's, that's the future I foresee. And as you mentioned before, I surely hope uh, that Umi plays a key role with this. As I was telling people early on, I was telling them, you know what? Even if Umi fails, I really hope somebody else with the same strategy as we have right now succeeds. Because it needs movement for the benefit of the sciences, enterprise, humanity overall.

• 1:01:52Speaker 00

  That being said, I surely hope that uh, uh, you know, uh, Umi is the one that helps achieve this, but uh, again, uh, the, the biggest thing is, you know, we make sure that this happens for the benefit of all of us.

• 1:02:05Craig S. Smith?

  Yeah. Okay, well, that's, uh, that's all fascinating and uh, if somebody wants to find Umi, uh, give us the URL or, or where do they go on GitHub?

• 1:02:19Speaker 00

  Yeah, so somebody can go to www.umi, O U M I, dot A I. Uh, and then we have also the links to our GitHub and our Discord where they can, uh, join and uh, help contribute. Uh, we already have quite a few GitHub issues as they're called and efforts and also some projects we have started in our Discord. Um, so yeah, uh, we would more than welcome for real contributions. As I mentioned before, the best way to make this happen is, uh, if we all work together. So I'm looking forward to seeing more and more people, uh, coming aboard.

• 1:02:52Craig S. Smith?

  AI coding assistance like GitHub Copilot, Google Gemini Code Assist, and Amazon Q Developer have quickly become essential tools for developers. They generate code with remarkable efficiency, significantly boosting developer productivity. But the widespread use of AI-generated code brings its own set of challenges. Bugs, vulnerabilities, and suboptimal code can inadvertently enter production, leading to issues with maintainability, stability, even costly outages. How can organizations minimize disruption and risk

• 1:03:34Craig S. Smith?

  while maximizing productivity and innovation with these AI tools? With Sonar, the code quality and code security leader.

• 1:03:44Craig S. Smith?

  Organizations can amplify developer productivity in concert with AI assistance. They can improve developer experience with streamlined workflows and prevent quality and security issues in all code from reaching production. Sonar's AI Code Assurance feature, found in their Sonar Cube solution, offers a thorough validation process for AI-generated code, including the first in its industry to automatically detect and review AI-generated code from GitHub Copilot. The AI Code Assurance workflow helps to build trust in AI-generated code, assuring companies

• 1:04:29Craig S. Smith?

  that proper due diligence has been performed and code is ready for production. Their AI Code Fix feature speeds up productivity and issue resolution with instant AI-generated fixes for issues discovered by their code analysis that developers can review and apply directly within their workflow. Sonar is helping redefine the software development lifecycle by use of AI and AI agentic systems with tools that streamline the verification of AI code and provide guided

• 1:05:08Craig S. Smith?

  remediation of issues, supercharging developers to build better applications faster. Join over 7 million developers from organizations like the DOD, Microsoft, NASA, and Mastercard who use Sonar. Visit sonarsource.com/eyeonai to request a free demo and see how Sonar Cube can help your team with machine-generated software development. That's sonarsource, S-O-N-A-R-S-O-U-R-C-E dot com slash eyeonai.

• 1:05:51Craig S. Smith?

  That's E-Y-E-O-N-A-I, all run together, and request a free demo. Sonarsource.com/eyeonai.