01:32:39 · Jordi Hays, Jeffrey Morgan, John Coogan
[01:32:39] S0: What set of models are you most excited about for the back half of this year in the open weights world? [01:32:47] S3: I mean, I think with DLM five point we just had another massive moment in open models. And, you know, Olana by far, at least from what we know publicly, is the highest token volume of accessing DLM 5.2. And so I'm excited for that. I think there's going be a series of new models that are long horizon. They're focused on these really hard agentic use cases, and there's going to unlock so many use cases in enterprise that the prior generation of open models couldn't. The gap between open models and the frontier models is shrinking, And so, you know, I I think at that point, you we're able to get to these incredible use cases that just weren't there, you know, three or four months ago. [01:33:23] S1: Take me through some of the game theory in the open source community around those rumors that we heard that there might be export controls on open weights models coming out of China soon. If if if we stop getting frontier or or or near frontier open source models from China for free. Is the next step that you would see an American company step up, NVIDIA, or maybe Meta changes their strategy? How are you thinking the open source ecosystem would evolve if China changes their strategy?
01:34:02 · Jeffrey Morgan
[01:34:02] S3: Yeah. We like to work backwards from our customers. What are they trying to do? By and large, they may have preference on specific geographies where the models are from, but by and large, they're adopting both. And some mix of open models and frontier models as well. And to your point, I think The US models are absolutely stepping up. They're incredible. The Nemotron three Ultra model is just amazing and is able to accomplish some of these long running agent tasks. And then also, you know, one of the most downloaded models on the llama is a US model. It's the Gemma models. You know, this is like a super amazing team at at DeepMind that's bringing them out. The new ones are, you know, agent ready. Like, can run coding agent loops. They can accomplish much harder tasks. And so, look, I think it's really up to the customer. If they want a US entirely US built model designed from scratch, that's there. If they want a Chinese model, which is often the case, they it's less about where the model's from. It's like, where does it run? And is it running next to your data, which, you know, a lot you can deploy it locally and then are you able to deploy with safeguards? And ends up a lot of customers, they're not looking for, like, where the model's from. They just want make sure that they're running it properly and safely so I think, you know, they can have a a understanding of what's gonna you know, what could go wrong, but what could go right. And and there's a lot of safety tooling that can be deployed to help with that. They have tons of appetite for that.
01:35:19 · John Coogan, Jeffrey Morgan, Jordi Hays
[01:35:19] S1: What do you see your role as in terms of benchmarking reality checking, vibe checking, different models, helping enterprises that work with you to make the right decision, pick the right tool for the job? [01:35:34] S3: You know, our job fundamentally is to connect the 9,000,000 developers on Ollama to the right model for the right task. And that's that's step one. And so just by having that sheer volume and that this critical mass of devs, we're able to already understand just from, you know, our community which models are performing right for the right tasks. That's a starting point. I think from there, it's really collaborating with the Model Labs and we're launch partners with every major Model Lab and just making sure that the best parts of the model are shining through through Alama, including what are they capable for, what are their benchmarks, how can customer customers benchmark it for their own use cases. Sure. It it all comes down to a lot of software tooling and and and and and, you know, a community in a in a network. And that's Mhmm. Kind of what we built and and it's what makes Ollama special for developers. [01:36:20] S1: Got it. $65,000,000 raised. Is this like, what are you using the money for? Because you don't have the crazy training costs because you're more of a gateway. Is this head count off [01:36:31] S0: 2,000 BDRs. [01:36:33] S1: Is that what banging down here. Wait. Wait.
01:36:33 · John Coogan, Jordi Hays, Jeffrey Morgan
[01:36:33] S1: Is that what banging down here. Wait. Wait. [01:36:35] S0: No. It feels like you have also, like, bottoms up adoption with developers. Yeah. So you maybe need a lighter sort of, like, Salesforce to to actually capitalize on that. [01:36:44] S1: Yeah. What what yeah. What do the next, twelve to eighteen months look like for you? [01:36:47] S3: You hit the nail on the head. Look, we've we've put out a you know, on our site, hey. We're launching a Teams plan. We were inundated with thousands of teams that wanna use Olemma. [01:36:55] S1: Yeah. [01:36:55] S3: And, you know, that's gonna that that's the core mission. Like, look. We've got this critical mass of devs. How do we go solve problems for businesses back to what we were just talking about? And that takes a team. So obviously, we're expanding. We got here with 14 people to to company of this magnitude. Mhmm. But there's a much bigger team to be around the market. [01:37:13] S1: Very [01:37:14] S3: good. And then, of course, you know, one thing Alignment does very special for the larger open models is we host it on US and European servers. [01:37:21] S1: Okay. [01:37:21] S3: A lot of the consumption of open models is going to China or is going to servers where [01:37:26] S1: Sure. [01:37:26] S3: It's there's there's no data retention guarantees. And that's so important for companies. And so that's a compute investment we're making and really enabling, you know, every business in the world to access the most powerful models on compute that's secure and safe in The US or Europe.
01:37:26 · Jeffrey Morgan, Jordi Hays
[01:37:26] S3: It's there's there's no data retention guarantees. And that's so important for companies. And so that's a compute investment we're making and really enabling, you know, every business in the world to access the most powerful models on compute that's secure and safe in The US or Europe. [01:37:41] S0: Will we ever settle the debate on whether the gap between open and frontier models is closing or widening? Because depending on what sort of group somebody is a part of, they tend to have one one view or or the other. But it but I think in in reality, it's probably always kind of going like going like this to some degree. But what's your view?
01:38:06 · Jeffrey Morgan, Jordi Hays, John Coogan
[01:38:06] S3: Yeah. I think you're right. It's oscillating. I mean, I'm a daily GLM five two user through Alomma right now and it's replaced 80% of my coding work. And I think a lot of that's gonna be true of a lot of customers. As for the gap, like to your point, I think it it may widen, it may shrink. I think overall it's shrinking. But but ultimately customers are gonna use a mix. And, you know, for the bulk of their use cases, they're gonna reach for these open models because they can tune them to be much faster. They're obviously much cheaper. And there's always gonna be use cases where you need the frontier. I don't if we'll ever settle the debate. I think ultimately the gap will will continue to shift. I think that's what makes it exciting. Right? It's like every every three months we're able to do something new. We're able to run better agents and quickly open models will catch up and really enable a whole wave of customers that wanna run open models to do that, you know, in their own environment or to customize it to the point where, like, they can even make it more more powerful. The last thing I'll say too is, you know, customers are readily taking these open models and customizing them, and they're actually getting better results often than [01:39:04] S0: Yeah. [01:39:04] S3: You know, just a a stock frontier model. Yeah. And, you know, I think we're just at the beginning of that transformation. [01:39:10] S1: Very cool. Well, congratulations on the progress [01:39:12] S0: Awesome to meet you. [01:39:13] S3: With the round. [01:39:13] S0: Congrats to