Will OSS Models Take Over?

0:07Aaron: Good morning, good evening, WebYarn. Welcome back to the Enterprise AI show. This is your host, Aaron. And if you're following along at home on the last intro, I was in Utah and now I'm in Maryland, but I'm heading back to the massive studios later today. For today's show, we have the second in a two-part series talking about OSS models, both a potentially positive outlook and a negative outlook. And on the last episode, we discussed the potential decline. And on this episode, we'll be talking about potential positive futures for OSS and open weights models. Brandon and I explore the future of open source models, potential role of industry consortia, and how major tech companies like NVIDIA, Google, and others are shaping the AI landscape.

0:53Aaron: We discuss the potential for open models to become industry standards and the strategic motivations behind these moves. And all of that is coming up right after this break. Today's show is sponsored by Nasuni. There's a growing gap in AI right now between what's possible in theory and what successfully works at scale inside an enterprise. The difference comes down to unstructured file data. Many AI initiatives struggle because the file data they depend on is scattered, unstructured, and disconnected from where and how work actually happens. Nasuni changes that. It brings your unstructured file data into a single, secure foundation so AI, both generative and agentic, can access it with the context, governance, and performance it needs in production.

1:39Bring AI to where your unstructured data lives. See what it takes to activate your data for AI and request a demo at nasuni.com slash AI.

1:49Aaron: And we're back. And Brandon is here taking over the podcast once again. And for those that listened to the last episode, this is the second part of our OSS Models kind of mini series, if you will. And this is the positive take on all of this. We went full Ed Zitron on the last one and took the negative sign. And we were the bears to it. Now we're going to do the bulls. So, Brandon, why don't we kind of lay this out here quickly? Okay, OSS models, do we think they're going to take over the world? And I'm going to point to the why now. There was the NVIDIA Open Secure AI Alliance, which, by the way, that's a mouthful.

2:36Aaron: And everyone but Anthropic joined it. Mm-hmm. And then they even handed it over to the Linux Foundation to manage it and manage the proposals. Could this be the catalyst of what the industry needs? Well, I don't know if that is the catalyst because they need somebody on the marketing team to rename that thing. I don't know if that's going anywhere. But I think an analogy we should draw that many of our listeners will be familiar with would be the old Kubernetes, right? Right. And the case to be like, why is it open source is that if there are enough stakeholders that kind of, you know, agree with oligarchy, if you will.

3:17But if, you know, to say that more simply, it's, you know, if a bunch of people agree that like, you know, it makes more sense for this to be open because then we all make money rather than, if you will, playing a game of like winner take all. Right. And I think you could kind of maybe make the argument. I mean, that's really what happened with Kubernetes. It's not so much, you can listen to my other podcast, my other co-host has a whole long rant about this, but you could say like Kubernetes was maybe not the best solution, but it was the solution that the industry was like, hey, it's all better if we all just use this and we compete this and we can all kind of protect our domain.

3:50So taking that and applying here, I think something along the lines of like NVIDIA, you mentioned, I think we've talked about in previous shows, we talked about Azure and AWS, not building a frontier model, wanting a frontier model, talking about companies that are controversial like Palantir wanting an open model. A company we haven't talked about as much that's in this world, Apple. I think Apple would be very content for there to be like a model out there. So, and then of course, NVIDIA, they just want people to keep buying their chips. So if NVIDIA believes that like an open model leads to more people buying GPUs, then you will.

4:23So you kind of have this consortium, poorly named, you saw it Open Secure AI Alliance. You know, maybe if that group of people really do get together and say like, we all benefit from that and we would like to put together a group that's going to put a foundation model out and it's going to make it available either open source or for very friendly terms, I think that's the reason that you would be, if you will, bullish, the anti-Zitron take. It's like, that's what people are going to want to do going forward. And of course, maybe the government even steps in or governments around the world step in.

4:54It's like, listen, this technology, it is just too important to allow any one company or even a few companies dominate. So let's make sure that we always have an open source model. So that would be, to me, the best case of why it gets created. What do you think?

5:10Aaron: So first of all, I agree. I feel like we're just starting to see the beginnings of this as an industry, but I feel like very few things follow new patterns. And I feel like Kubernetes is the exact pattern to follow here of like, Like, okay, you know, company develops it. It gets out into the foundations. Yes, it may not be the best or it may not be like, I love the Kubernetes as a platform to build platforms kind of analogy. And so the idea here is, okay, great. You have something out there that is the least common denominator. And, you know, if we look at where we are today, like, okay, the models, like they keep increasing in size and then the training costs keep going up and up and up.

5:58Aaron: And then, oh, by the way, the shelf life of them we've already talked about isn't that great. And we've talked about parallels with the database market. But I think in this instance, Kubernetes might be even better of like, okay, I could see, and we've hinted at this, and this is where we need to record some future shows about this. How important is the model going forward? How important is the harness going forward? And if we get to a point where the harness and everything else that wraps around that model becomes the area of customization, the area of specialization, does it matter the model that's running underneath?

6:43Aaron: and the idea here i think the industry is really coming to the conclusion i think man if i go gosh i would say beginning this year i think the industry was very bullish on customizing your models take models you know fine-tune them do post training on them make them your own but then then what then you got to make the next version of that model you to make the next version of that model. And as, as we've talked about, there's not a lot of data scientists in the world that can take a trillion plus parameter model and make it special. And so I'm not sure how well that goes going forward. Now, if there was tooling in the industry to, to do that, great.

7:26Aaron: But I think at the end of the day, the entire industry is moving towards models being a foundation product and i think you put it best of why did nvidia do something like this well because at the end of the day nvidia wants to sell more chips and by making all of these models and maybe we can maybe we do get to the point where there's a world in which there is a a linux foundation llm that is co-maintained by a bunch of folks and you know i mean there was like gosh i forget what the company was but there was a company that was kicking in for kubernetes and was doing all the testing of kubernetes for a long time there and was funding everything and like nvidia starts to fund things like i'm playing a you know the wouldn't it be awesome f world game right now but like i can see a scenario in which we absolutely have something like that going forward because and and what's in it for nvidia i think what's really in it for nvidia in this instance is if this market ever becomes a open ai and anthropic take over the world well again who owns the account who owns the customer and oh by the way they could very easily go train it on tpus they could go train it on their own custom chips they're developing everyone said they're going to do their own custom chips so nvidia needs to do something to keep the world interested in their layer yeah and of course they have a lot invested you know obviously the cuda and they have the the whole proprietary stack and things like that so you kind of think about it for them it's like well, let's keep the model open source, but let's keep the tools and how to do it and how to utilize GPUs in the most effective way.

9:21Well, let's keep that as our proprietary features that differentiate us. So that'll be interesting going forward. And then of course, I just think Apple is just such an interesting wildcard here, right? Because after essentially just completely missing on AI completely, and I've used the recent beta of the new stuff, it's not so much that the new Siri is like great, but it's sort of like table stakes. And it's just like, but when you can't, when you think of the scale of Apple and like local inference and running these models and like, you know, I don't think they have any aspirations to like, you know, if you will, build and sell models for like the enterprise.

9:57I think they just want to make sure that we keep buying these thousand dollar iPhones that are going to probably be 1500 or$2 ,000 going forward. So I just think there's, there's almost like this sleeping giant that's just trying to figure out like what it wants to do. And then if you look at most of like Apple's foundational stuff, a lot of it is open source. Like obviously, you know, we talked about that on previous shows, right? You know, a lot of this is just Unix and stuff and they kind of bring their own special, you know, flavor to it. So, you know, whether it's that or, and of course, then we have the whole Android ecosystems and we have other ecosystems out there.

10:27So there is, you know, there's just going to be a lot of momentum to be like, where do these models go? And as people that have different business models selling a device, they may have different motivations and their motivation may very well be to like, let's keep it open source. Just like the motivation today, for sure, for Azure and AWS is like, they would like nothing more than great open source models that they can host. And then, you know, and then all these things, and you kind of touched on it before around the harness. It's like one thing that we know, it makes me a little sad, but you know, this is the world, this is the life I've chosen in enterprise software.

10:58It's like, you know, all these companies want to grab the model and they want to like put all the enterprise stuff around it. They want to put up observability and security and scalability. And then you got all these consoles and it's just like, part of me is like, I guess it's a necessary evil, but it's sort of like, that's what they do. Right. That's all the stuff they do. And no one wants to do that in open source. No one's like, I want to work on the identity access management for an enterprise, for an open source product. No one does that. Right. They leave that to the vendor. So there's a lot of motivation out there to actually do that.

11:26And then, you know, like I said before, it's like, you know, you have Anthropic, you have OpenAI. and it's just like, I always, this is the part where I always acknowledge that we just don't know. I don't know. Like I've been, it's been explained to me and I've watched a lot of the videos about the mathematics, about how they create all the models. And it's very interesting. And it's just always like, who knows? It seems to be, it's not like someone just creating an equation. It's all these super smart people taking information from just different disciplines. And they kind of like the scientists kind of just wrap it together and then boom, something great comes out of it.

11:56So that's the part, like, I just never know. Maybe that happens somewhere else and maybe that continues the proprietary stuff that goes on for a long ways. But if it's just what we know right now and it's just about getting better at that, then that really does lend itself for something that would be open sourced or trending more towards commoditization than anything else.

12:15Aaron: Yeah. And so I want to go back to Apple because the Apple thing is super fascinating to me for actually a couple of different reasons. I mean, it's semi well-known, although they never talk about it, Apple really uses, I mean, most of the major OSS projects that are out there. I mean, they are really big, both consumer and oftentimes quiet contributor to a lot of the open source. Absolutely. Today. And I could see a world in which they absolutely become a player in this quietly behind the scenes never publicly and it would be interesting too and maybe i'll ask you since you've played with the series stuff like i'm super interested too is like okay are they going to do small models on device are they going to do big models in the cloud are they just going to farm that out to gemini like i'm so interested in the potential architecture around something like that?

13:20And will OSS models play

13:23Aaron: a part in that? If there was a company that ever did kind of lead the way, actually, I could be them. Yeah. And I think the answer to that today, I mean, we should maybe take a little time shot or a snapshot rather of like what they're doing today. So today the answer is kind of both, right? And then the new one, they've got a local model and that's usually what's doing like the text message, auto correction stuff. And then when they use their world model, they call it, when you go into Surrey, which just looks like any other chat bot, that is then going back to their data centers, right? And then you can actually use them just like you would use ChatGPT.

13:57And all indications are, I mean, this is not breaking any news, that is some version of the Google model, Gemini, that they have highly customized for their own use, right? And then they've got that. And then at some point, it sounds like they're going to have some paid tier, or maybe it'll just be rolled into Apple subscription for those that are heavy users. So that's sort of like the ecosystem that exists now. But you can kind of see, it's like, okay, let's assume the rollout goes well. We're all going to know here soon enough in September. And if they start to see an uptick in it, right, and especially if they can then say, like, actually, it turns out we can add to the Apple subscription$5 more, right, for, you know, if you will, unlimited AI usage.

14:35and we can start to do things on the device that are very interesting, right? Then I think suddenly, right, they're going to go back and figure out, it's like, well, maybe we should invest in specific models that we need. And the thing that I think will be different for them is like, it will be very clear where they need to invest. I think where they got caught up was they sort of just sat in a room, made up a bunch of use cases, didn't have a lot of usage and then promised things that either no one can build or they couldn't certainly build. But this time it would be much different. It'd be very driven by usage.

15:04And I think that's where you can see them going forward to say, well, listen, we don't necessarily want to use the Gemini model going forward. Just like, you know, go back in time. Like they decided, like, we're going to write a new programming language. We believe Objective-C is dead. We can't go any further. We'll build Swift. And maybe the same thing with Gemini. Or maybe it just kind of goes back to, like, we'll call it the natural state of things. Like you like to talk about, everything is just like there's a natural state. And the natural state maybe is always, if you will, Google and Apple having this like behind the scenes relationship where Google pays them to use the search engine.

15:35And maybe this just gets rolled into that. But, you know, the thing and I think, you know, we, everyone listening to this podcast is an expert at AI relative to the rest of the world. And remember, most people, probably their first interaction with potentially AI is probably going to be an Apple device. It's probably going to be this mobile device when it comes out. because a few billion people have used it and they may not even know they're doing it. So just a whole nother world could open up and a lot of it could drive the use or the creation of open source models.

16:06Aaron: Yeah, agreed, agreed. And I think at the end of the day, the future is open source models or open weight models. I don't know that, I think we're going through that difficult growing pains stage right now where everyone is fearful everyone is is trying to figure out what this future means and everyone is like worried especially all these you know companies that are burning vc money they're trying to figure out their place in all of this and we're all going to figure it out it's going to be a pretty interesting six to twelve months absolutely so all right brendan well i appreciate your time as always and for everyone out there listening thank you very much for listening this week as well If you enjoy the show, please, wherever you get your podcasts, leave us a rating if you can.

16:56Aaron: And of course, we welcome your feedback for both shows and guests and show topics as well. So for that, everyone, thanks for listening. And we will talk to everyone next week. Thanks for listening. Check us out at theenterpriseaishow.com for past shows, newsletters, and all things Enterprise AI.