TRANSCRIPT 🔗
Rachel Roumeliotis: Hi there, this is Rachel Rameliotis from Punch Tape, and we are here with the latest episode of Contribute Beyond the Code. This week we have an amazing all about how open source and AI are fitting together. Nithya Ruff is gonna join us. She is amazing. She is gonna be speaking a lot on these topics over 2026 and I imagine 2027. She's currently chair of the Linux Foundation Board of Directors. which is awesome. She was previously with Amazon, Comcast, and a few other companies. And so she will be joining us right after I touch upon two stories that I feel like are at the opposite ends of the spectrum of one actual story. Maintainer burnout, which we all know is an issue, is a major reason that NGINX is ingress. is not going to be receiving security patches after March 2026. While this, I'm sure this is not the only reason, I think it seems like it is true. And so this is something that people are talking about already and have been talking about for years, that there are critical pieces of infrastructure that are open source. And it's just because people in some cases are taking care of them that they're able to stick around. And so what do we do when that person can't or doesn't want to anymore? How do we solve for that? I do not have that answer. Now, on the other hand, another news story was that there was some new funding for critical infrastructure that is going to help out the OpenSSF and it's going to be millions of dollars and it's from a bunch of different companies. Among them are, let me see here, sorry, Anthropic, AWS, GitHub, Google, Microsoft, OpenAI. And it is going to make sure, again, as we know, an ongoing issue. security. We just talked about that with that Ingress open source project. And also talking about security in the AI systems, which is obviously a major thing too. So seeing it's interesting. It's just something to think about. I don't have a definitive answer on this or anything, but is money at the core of it? You know, it's that why things are able to move forward or not? I don't know. And where does that money come from? And with money comes influence, I imagine. So that's interesting too. And, you know, making sure that we have a well-rounded reason and group of people guiding where open source is going. So just something to think about. Wanted to point those two out. I'll add them to the details or the show notes, whatever they're called. And now we'll head over to the interview. And I want to thank Nithya. It was educational and inspirational as all of my conversations are with her. Thanks again. Hello. Thank you so much for joining me today, Nithya. So good to be here, Rachel. Thank you for inviting me. Yes. So we are going to talk mostly about open source and AI, which I know that you're focusing on, especially now, but have been. But let's start sort of general, right? So you've talked about how open source is cultural infrastructure. Totally agree. Tell for someone who might not be as like
Load full transcript
on the inside of open source what you mean by that and why it's so. Cultural infrastructure is such a great word because we often think about open source as code and as technology and has projects. And so we most often talk that way, or we talk about licenses and you know, how, that means and how you comply, et cetera. But at the heart, heart of open source is really processes and practices that we have as a community that make open source what it is, whether it's transparency or whether it's. common values or it is the governance that we put on projects so that there's multi-party involvement, there's checks and balances in terms of not having the power vested in one place and really valuing people who actually participate and contribute and show up to do the work every day rather than people who appoint themselves, if you will, as leaders. So I think open source is one of the few places where you can have those types of collaborations and that kind of meritocracy, if you will. I know it's a controversial term, but it's worth protecting. Yeah. No, I mean, I completely agree with you. It's one of the reasons we started the podcast. Open source is so much more than the code. So I love that you're spreading that word. And there's lots of people involved in the cultural side. doing all good work out there and just trying to give it a platform amongst many that are out there to do this. But I think that's a very clear outline for anybody who has not yet been initiated in. So thank you for that. Okay. So I actually want you to do some basics on this one too. So some companies you've called out for open weighting without releasing training data and model weights. One, can you just explain that before you get into your answer on? How do we make sure that, you know, what does open actually mean in AI? So kind of a big question. Yeah, it is a big question. You know, it was, the world was pretty clear when it was just software. And then we kind of knew what open source software meant that, know, the four freedoms of open source that you could actually see it, you could distribute it, you could modify it, you could use it for any purpose. And when you could read the source code. So it was actually very straightforward. I think AI makes it even more complex because when you think about models, models are really composed of lots of different components. There's some software, there's some data, there's weights, which is the training weights and algorithms. And then there could be some documentation, et cetera. And we have to then say. Do all of these components need to be readable, usable, open sourceable? Do the freedoms apply to all four? All of these components and then only do we declare it open source. And the community has been debating this for quite some time because the practicality of the matter is data is hard to open sometimes. There's copyright involved, there's some data is proprietary, some data is PPI. ⁓ Etc. And so you really don't want to or
have the rights to share this data. And so some companies have erred on the side of not sharing the data, but sharing only the weights and also adding some restrictive practices in their licenses in terms of where you can use it, who can use it, and yet calling themselves open source, which is, which is why I think a lot of people feel that it. does not create trust when you're misrepresenting what it actually is. But to come to a more pragmatic place, I think things like the model openness framework from the Linux Foundation have tried to say, there's no one standard for open. We can define a few tiers. The most open being open science where all of the components of the model are open and examinable, modifiable, et cetera. And there will be a few use cases where organizations can satisfy that and want to, because for the sake of science, for example, or reproducibility, they may want to do that. But a majority of organizations may stick to something called, I'm trying to get to that. There's an open model, I think it's called, which is the entry. point where not everything needs to be open. And I think it's acknowledging that we're not quite there yet. We still have to resolve some of the data issues and other, what license do you attach to these different components, et cetera. Yeah, it's definitely complex, as you say. And I always think with open source or in any area, like regulated industries, how does that play a role in it? So you're starting to talk about, in your answer, their governance, right? How does open source fit into that conversation? mean, again, I feel like that's a difficult conversation without open source. So I want to hear how you think they can blend together. think open source brings a lot to the party. And that's why I'm such an intense student right now of open source and AI. And what lessons do we take from open source into AI? And how do we move forward? How do we create more openness in? AI, cetera. I think there are three important things open source brings to the table. One is the transparency of open source allows us to audit models, correct for bias and other issues. really transparency creates trust. The second thing about open source that you and I know very well is that it distributes power across the world and it doesn't concentrate power in the hands of a few. Lastly, Open source historically has created access to a broader and broader set of people. It has brought curves, cost codes down. has created open source on technology that everyone can afford and provides access, you know, broadly across the world. So I think those are all three big things that we can bring to AI. But open-on-openness alone is not enough. But you had a question, Rachel. No, I was just going to say. It's not interesting. mean, what I've heard you say, I mean, is trust is the thing that rings out to me. And I think that's so important because I don't think anybody that knows anything about AI feels like trust is trustworthy at this
point. And I don't mean just like terrible. mean, like, is it giving you the right answer all the way up through, you know, scary stuff? But it's just interesting that the solution there is, part of the solution. is most likely, I would think, open source. Yes. The opacity of models, I think, and not knowing what they're going to do next, how they behave. Even the creators of the model sometimes don't know how they will behave and what answers they will give because it's non-deterministic. so hopefully, if we can do more open source models or introduce more openness in how we create models through data, et cetera, we'll better understand them and humans will be able to use them in a more intentional and understandable way versus just being subject to, you know, whatever they do and drive it, right? And the thing that openness does not bring, unfortunately, to the table is there aren't enough policies that we can bring to the table for AI from a governance perspective in terms of. How do you prevent harm? How do you prevent AI from going rogue? What accountability frameworks do we need for agentic AI? Right? Do we need a human in the loop? Can we have agents go off and execute on certain things by themselves? How do we inspect how they make decisions and things of that nature? And so I think we have a lot we can bring to the table, but we have to also modify it for this world of AI and say, how do we work on this together? For example, I'm part of a working group at the Adjantic AI Foundation and it's on regulation, security and governance. And so we are looking at exactly this in the world of Adjantic, what kinds of new policies, new processes, new tools do we need to make sure that things go well? That's great to hear. Yeah, it's interesting times, I think. I'm glad that you're thinking. policy and hopefully feeding that up to governments across the world. I think that that's going to need a lot of help because I don't think they're the most technical group. So glad to hear you're working on that. Yes, yes. And we have a huge role to play. And we've learned to do that in open source over the last decade, which is how do we educate policymakers? How do we collaborate with policymakers to make sure that laws are well-informed, regulations are well-informed? the intentions are good, but can we understand how this actually works and make the regulations not burdensome, but there to protect, you know, citizenry? absolutely. And so something else that's come up too that I've been reading a lot about and that people mention is sovereign AI. I was a little surprised by it. Maybe I was being naive, but I think there's good reasons for it. You have said that countries need AI that reflects their own languages and values, but there is definitely a real tension between that sovereignty and fragmentation. So how do you see that resolving? It was at FOSDOM a year or so ago. I was in ⁓ the AI dev room and I think a French model company or AI company was talking about the fact that most AI today, you know, is trained either in the West, mostly the US, speaking, trained on, you know, a lot of North American data, or it is, you know, done in China. So they're really two big centers of where models are, you know, being created. And so they're influenced and created by the language. the values, the norms of those areas. And this person was giving an example of asking, I forget which model, the question, can you give us the recipe for beef bourguignon? And lo and behold, it gave Julia Child's recipe, Julia Child, the American chef, as opposed to a French chef who had created these recipes and they were saying, hey, we are being disaggregated, know, our culture, our language, not... in the system because of where it's trained and how it's trained. So how do you bring up a really good tension, which is should each country create its own stack? And if so, then you have a severe fragmentation, right? Or should we go with these two centers that exist today, which is not fair to these countries because it doesn't reflect their values. So I think open source may be an answer here. because if we can create the substrate, open standards, common protocols, et cetera, tooling and other things that everyone can use to create models. And then over and above that, they can create their own custom language models and other things. It still becomes interoperable. has common standards and governed globally by an open governance kind of framework. So maybe that is the answer. And we have a few components in that stack. Now we have, you know, standards like MCP and agent.md. And then we have the technology stacks like Kubernetes and PyTorch and Ray and Lang chain and Jupiter, et cetera. But there are still lots of holes in creating that common framework, if you will, that everyone across the world can use in an open way to build. their values and their standards. Yeah, I think that's really well said that it's sort of the substrate. I think that's the perfect word for it. And yeah, I do worry that without that, things may go awry. I think one of the strengths of open source, wherever it is, and this is something I think that we have to be vigilant about, is that there are different types of voices, right? All different places in the world, young and old, from different experience, you know what I mean? And so I think... I just think that's important and I'm glad to see that people are already thinking about it. So that's great. So you have also been talking about AI bias, which you kind of talked about in the previous question. There is a sequence to these. So as a training data problem, not just a fairness problem, right? Is there a version of open source AI, and you kind of talked about stacks already, that fixes that? Or does the openness of everything just showcase, which is a good thing in and of itself, that type of bias? I think it's more... that it creates the condition for bias to be discovered, named and
addressed by people who can do it. Transparency is an important prerequisite, but I think it's not enough. ⁓ And we do need ways to inform the data that we use to train these models. That's where the bias comes from, because the existing data out there is highly biased by us humans and the way we think over many, years. And so we have to actively work on cleansing the data or catching it from a transparency perspective. And that's one of the reasons we say it would be good to have open data. Waits alone don't kind of let us do that. And so I would say it's part of the solution. It sheds light, but it's not the whole solution. sure. Yeah, no, I think it's great. So you have been a force. in open source government forever. You're so gracious with everyone with me, with everyone that I see. So from in the 90s with SGI, you've been Linux Foundation chair. How much of what made open source government work in software do you think really is going to apply to governing AI? And you talked about sort of the space level. I think there's lots of lessons to be learned and shared. Can you just kind of that together for us and like summarize? think over the last 30 years we've learned a lot as an open source community and we have succeeded in working together and also transcending from one industry to another while open source started you know in the systems and operating system type of area. Next and... GNU and compilers, et cetera. It has consistently shown that it is valid and it works across different verticals, different transitions of technology we've had. We've gone from mainframes and minis to cloud, and we've gone to embedded now AI and it's still relevant. And I think what makes it relevant is the process and not just the code, right? As we've talked about. It's the transparent decision-making, it's peer review, these organizations, foundations like the Apache Foundation, Linux Foundation, where people can come together and work on common projects and programs together. Even competing companies, you know, coming together. ⁓ I think all these principles that we've learned will absolutely transfer to AI governance. The transparency that we can bring, the fairness. and the collaboration, et cetera. But sadly, AI is so complicated and is moving so fast and is so opaque right now. And we depend upon in open source transparency of these components to really ⁓ apply our rigor and process to. And in AI, so many of the things are opaque. The other problem is it is so compute intensive and only a few players in the industry have that kind of resourcing to work on AI. is capital intensive, it's opaque and it's governed by just a few actors and the government and policymakers, as we said, know, struggled to keep up with this technology and how to regulate it. So I think it's architecture is probably the best available template we have. but it needs to be adapted for this kind of a situation where the artifacts are not readable and the window is closing quite fast in terms of how AI is moving and we need to move fast. Yeah, think open source is the tenants of open source are so strong to have lasted so long in an industry where things do move so fast and they're like going hyper speed now with AI as you well know. And so I think exactly what you said where we have to, first of all, think open source always has to be looking at itself and sort of making sure it's balanced. And to your point, it's not like three companies that are doing this, right? So however we can make that happen and bringing in the different voices, I just think that's so important. but the speed, but yeah, so I mean, it's ongoing again. That's one of the reasons I am also trying to. get more people involved that are outside of software development? Because I just think it touches upon so many things and more and more, right, with AI and everything. So that actually leads me to my next question, which is a little bit different, but is, of course, like everything, touched by AI now. So contributing is, I think, something that everyone should be doing, right? And oftentimes that is not the case, unfortunately. But hopefully we can fight the good fight. But I do believe, like I was just saying, that contributing can mean different things. And I think AI has had an influence certainly on that. what do you see as, what does contributing mean in 2026? Is it about the code or has AI really shifted that? definition of contributing when we first started, you know, typically always meant code and I contributed so many patches, so many life codes. I contributed a project and it was always, that was what was measured. That was what was celebrated. And hence, you know, people who did coding and who worked on projects in that way were celebrated and contributing was considered code. I think over the last 15 years, at least, I know many people like you, Rachel, and myself, and others have been saying, wait a minute. These projects are socio-technological elements and they cannot function without community management or marketing or someone managing the finances behind the scenes or legal or, ⁓ you know, governance and people who contribute their time towards all these things, documentation, translation, et cetera. And in fact, Those are the reason why a certain project either succeeds or fails. sure. create code and put it out there on GitHub, but if you don't reach out to the right people, have the right processes, translate, document, et cetera, the project goes nowhere. And so unfortunately there haven't been good measurements for those types of activities as opposed to code. And when you go to GitHub, it'll tell you how many patches you've contributed, how much work you've done, but there isn't a count of. how many events I organized, how many people I brought to the table, or how big the project got because of the governance or contribution I made. And so we need to correct the measurements and metrics around this. To me, contribution is a
very broad word. is. who makes it successful and who creates a successful community and takes it forward. You make a really good point about the metrics because that's a way to look back to and see what people have done. I also actually really like metrics and goals because I think they set up, they can signal success. It isn't like they're there to like, we didn't get there. Like we did get there, right? So I think that's a really, really good point. So final question I ask everyone, and I hope to continue to do this, is that I want to give a shout out to people. You know, because as we've said throughout this whole interview, you know, it takes basically a global village, right? And so I'd love for you to call out a contributor, a team, program, anything you want that you think people might not either be aware of or they're not getting, you know, the shine that they should be. think the person that I wanted to call out was someone who's often behind the scenes and someone who, yes, does coding, but does a lot more than coding. The other things that this person does is really what makes them shine and have an incredibly broad impact on the world. So this is Shawa Khan. Shawa Khan is the first female Linux Foundation fellow. And when you think about the kernel Linux kernel, you always think about, you know, Linus and Greg and others, but Shawa has been a force of nature in the Linux kernel. And not just for the code, yes, she maintains the kernel regression testing framework and the USB over IP driver and the CPU monitoring, power monitoring system, et cetera. But her most important contribution in my mind is the non-code contribution. She leads the kernel mentorship program. And if you know this, it's a six month mentorship, which connects aspiring kernel developers to maintainers. And Colonel is the notorious, it's notoriously difficult and one of the pinnacles diet of getting involved from a technology perspective. And she does the unglamorous task of keeping the community safe through her code of conduct work, but also through the mentoring. And she's been a consistent voice for all these years. And I consider her a friend. and someone I admire and respect. There are a few others that I would say are amazing. Like give them a call out. I say, not? Let people know about them. And you know, frankly, when I started in the industry and I wanted to in open source rather, and when I wanted to do more work in open source, it was people like Jeremy Allison from the Samba Project who was my guide and who taught me. Dave McAllister at SGI. And then there were many, many women along the way. We became a sisterhood in open source because it's a smaller group of women, Supporting each other, reviewing each other's abstracts, encouraging each other to speak and do talks. I remember Deb Nicholson, who now is the head of Python Foundation. Don Foster. ⁓ We're friends. Ricky Hensley, Ruth Seeley. know, these are the... The women that I always seek out at an event and, you know, share talks, support each other, reach out and have a meal together. That's my community. These are the people who I think have been consistent voices in open source for many, many years. And we are really much richer for their contributions. Excellent, excellent list. I agree, like education and mentorship. is at the core of open source and not just to people necessarily that are new, but I'm totally for mentorship and support for people, colleagues maybe at the same level, telling somebody else that's been around for a long time what I think, or people new to it. So I think that's really, really great. Thank you for giving a shout out to all those people. Thank you. Well, thank you. That is it for today. That is amazing and I want to thank you for taking the time. I really appreciate it. I you're very busy and I'm sure we'll see you down the way a little bit. Yeah, I look forward to it, Rachel. Thanks for the opportunity to share some of my thoughts today and I look forward to seeing you soon. Take care. Take care. you