← Latent Space: The AI Engineer Podcast
Loading summary
Transcript267 lines
- [00:04]
A
All right, we're here in the studio with ISO Khan from poolside, together with Fibu. Welcome.
- [00:09]
B
Thanks. Thanks for having me, guys.
- [00:10]
C
Good to be.
- [00:11]
A
Yeah. Fresh on the plane. You texted me, you're like, hey, I'm on my way to sf. I was like, you're on the plane right now, right? Like, hey, you know, after I texted
- [00:18]
B
you, I realized that probably coming in with major jet lag, offer some fun
- [00:22]
A
experiences today, but let's do it. I mean, I think the thing I would tell guests is that they don't actually have to prepare that much because if you're trying to truly working on this every single day, then even what you hazily remember is going to be new for a lot of the audience that don't live in your world every day. Right. So 10 years ago you did a talk at Google Slush talking about the democratization of AI. And now here you are open sourcing an incredible new model that we're going to talk about. But I guess what got you into democratization of AI? It's not obvious from your LinkedIn or something?
- [00:58]
B
No, it's not at all.
- [00:59]
A
Actually.
- [00:59]
B
I don't think it's obvious how I got in this space place. I owe getting into this space to Andrej Karpathy. In 2015, he wrote an article called the Unreasonable Effectiveness of Recurrent Learnings. And that article, I read it and I pivoted my startup at the time overnight to working on RNNs and later LSTMs and transformer models to be able to write code. If you go to this article and you scroll down, you can kind of start seeing like this was the precursor to what ended up becoming language models. So at least what he was character level language models that we're starting to actually predict letters. He has an example out here. There's a little Paul Graham generator and you can kind of read it and the text kind of makes sense, but it doesn't. And there's a little. There's an example of code a little bit further down Shakespeare. And for some reason I read this and I went down the rabbit hole of learning everything I could about RNNs and LSTMs. Right. This is pre transformer paper. And I had built a completely unreasonable belief that neural nets should be able to generalize to anything and everything and that language should be able to generalize, you know, to a lot of things that are intelligent and the ability to write code. And so I started building Sourced, which was a fully open source company trying to build what we used to call Machine learning on code, language models on code. And we spent about four or five years on this till the end of 2019. And that sounds really cool today, but back then no one cared. No one cared. We were in the dark. We did things along the way. We tried applying convolutional neural nets to the structure of code. When attention came out, we were applying it to lstms. And then the transformer paper came out and it wasn't obvious and what we missed throughout that entire journey that we were on the right track, but we should have just kept scaling up. And today to all of us, the scaling laws and scaling up seems like the most obvious thing. But having spent four or five years of my life on working on language models on code, it wasn't obvious. I have a lot of respect to folks at Google and OpenAI and others who kind of took that confidence and kept going. We failed ultimately at the time and it kind of was like, biggest failure of my career. Right? You blew $12 million of investors money, which was a lot back then. You spent still a lot. And you spent years with a group of 40 people just obsessing over this problem. And life took a different turn and family kind of became a focus. And I kind of kept my heads down and really, frankly didn't really look at language models for the following two years. Kind of big mistake considering following years. Gonna be really interesting. And then ChatGPT came out and it was kind of like a vindication. It's like people started texting me. I found like my old work decks and these old talks. And throughout that whole journey, we kind of really had a strong point of view at the time that as you're building more capable intelligence, it should be open, open source. When we started poolside, that actually wasn't the case at all. And I want to be very open about it. When we started poolside, we were like, there was a premise of two things. One is this technology is not going to stop compounding in capabilities. I think to most people obvious today, but three plus years ago when we started, most people were still arguing if these were stochastic parrots or not. And the second was that reinforcement learning was going to be the biggest driver for LLM capabilities today. Very obvious. Three years ago was not an opinion held or direction held at either OpenAI or Google or Anthropic or others. And so people kind of looked down on us a little bit or like, you know, is this really going to work? And so we just started working the problem and we never really thought about open source again, we just kept our heads down and we built our knowledge understanding from scratch, right? We didn't roll out of an existing lab, so we picked up the papers and started writing code and figuring things out. And it wasn't until the beginning of this year that me and my co founder Jason picked up the open source conversation again. And if you go back to some of the early things on our website, it was very straightforward. It was, we want to get to AGI, we want to support a world of abundance, and we want to be the first company that gets there. But we started talking at the beginning of this year because it became kind of obvious that the world was going in a direction that was starting to kind of like, kind of like pick at us a little bit. Like it didn't, this didn't happen overnight. It was like a little bit. We were seeing this and we're like, okay, the world's going down a path. And throughout this journey there was something that I kind of used as a, as an analogy or things I said kind of, well, if I go back to back in those days, kind of 2015 or 2016, we're working on this. And I picked up a sci fi book off the shelf and I was reading the book about 2035, AGI is achieved and the story would be over the following, you know, decades. And I would have that first chapter where everyone's trying to figure things out. You'd get the chapter of ChatGPT coming out and then you would get to the chapter where the world was at a fork in the road. And the one that it picked was one where three or four or a handful of companies were going to create all of intelligence moving forward. And when I thought about that story, it felt like a dystopian sci fi book, not a utopian sci fi book. And the reality is I'm a utopian sci fi guy. And so we kind of took a step back and said, hey, can we play a role here now? It was easy for us to do so because we were not at the frontier, that we were at the frontier. I don't think we could have changed our mind. And I don't mean this like it's when the moment there's too much capital involved, too much expectations, you've built up things, right? We're small team, you know, just improving and improving. And so we knew that we could make that decision now, but it would be a lot harder to make as we got closer and closer to the frontier and caught up to others and did A lot of soul searching, a lot of conversations and said no, this makes sense. Even if there's big unanswered questions like how the hell do you build a business model with foundation models about open source, Big open ended question that we do not fully have the answer to yet. Right. At what point do you no longer want to release open source models? Because misuse of models has real potential risks associated with it. How is the government going to respond to open source? But I think it all just came down to one thing, and I'll stop the monologue, is the fact that I rather live in a world that has 100 foundation model companies than a world that has five. Even if I was one out of five. And the smallest and most meaningful contribution we can make for a hundred to exist is to open up our research and open up like our weights right now and kind of figure out along the way how we can do more.
- [08:01]
A
Yeah, I think if anything over the past three years that has become a bit more true. You are one of a cohort of NeoLabs that people are now calling that and we're doing this on the day that Thinky launched their new model and you are outperforming Dev on some benchmarks that they released. Right. They just don't have it yet. So it goes to show that I think actually this is one of those things where there actually is room for multiple players and you are seeing a little bit more in the future. Maybe more like 20, not 100. But you are one of the 20.
- [08:36]
B
I really hope so. Right. I think we are. I'm excited about their release. I'm excited about everyone kind of releasing because ultimately choice competition is both going to drive progress in the right direction, but the fact that, you know, we, we create models and while we all, you know, drink out of the same well of, of data effectively, we do introduce very different behaviors and biases in our models. Some are intended biases, some are completely unintended biases. And if we shape up in an ecosystem in the world where open models are going to be a part of the token economy, like I don't think there's any question about it anymore, then we want to be able to live in a world where companies, countries, people can choose and say, hey, I am most aligned and I trust most this provider for these kind of things.
- [09:25]
C
Yeah, I think more than just one of the 20 Neo Labs. Up until recently, most of open source innovation was coming from the Chinese labs. Right. So there's the deep sea of the West. Is it today? Okay. Maybe it's thinking machines, reflection, but there aren't many. Right. So one of the things you guys started sort of in France, Europe, but very much now you're kind of taking that American standpoint. And more than just that, the point is the Chinese models that we see, they're not super open research. The work you put out is, I think, some of the best. So every few months you get not only frontier models, but also here's a sort of breakdown, blog paper, technical report of here's everything for state of whenever to build frontier intelligence. And you're filling that gap too. Right. So not just only open weight, not just Western, but also pretty open research.
- [10:20]
B
No, I appreciate it. Look, I think it's actually the most meaningful contribution. Right. Weights are a binary. Let's call them what they are. Yes. We can modify them, we can change them, but giving someone the weights does not allow them ultimately to recreate what you're doing.
- [10:34]
A
Right.
- [10:34]
B
And so now there's challenges around releasing data sets. Challenges around, like, release certain things, but being able to share your research, like, right, how did we do it? What are the lessons we learned that we spent, you know, tens of thousands of experiments of compute on? I think very much so. One correction, ovibu and I say it's because it's kind of been haunting us for quite a few years. We actually, from day zero, were an American company.
- [10:55]
A
Yeah, they moved to France.
- [10:57]
B
So the story once and for all is very. We started as American company. We have always been an American company. And early on we made a very conscious decision. We said, we're not going to hire any researchers in the Bay Area. We're going to actually look for talent everywhere else in the world. And that is everything from Middle America, Seattle to Serbia and to Taiwan and Singapore and other places. And it was because we kind of took a view that this was going to become a talent war for this. And I think it has over the years. Now, three years ago, that wasn't fully obvious yet. I think today it very much is. And we also realized that some of the world's most capable people with the most interesting, innovative ideas were not just going to be here. And so it led us to create a fully remote company. And we ended up opening an office in Paris and London and different places. And we have a lot of team in the US And a lot of team outside. But we kind of always took this view of we're an American company, but if we want the best of the best to work with us, we take a global view. Now we do also have people here in Silicon Valley, like, the company's grown and others. But I think one of the things that it slowed us down at the beginning, but it has sped us up now. And it's why you're seeing like the progress, I think, on our models and the cadence, that which we release is because we didn't roll out of an existing lab. Right? We didn't. We didn't actually have a lot of the information that's freely flowing around here at the time. We just kind of took this point of view. It's like, okay, well, let's just work the problem. Let's just go and like read the few papers that are out there and let's just figure this stuff out. And we made some hilarious mistakes in model training because of that over the years, like especially in the first 12 months, there's a few that I think still haunt me and scare me. We're going to talk about them later. But it kind of created like a resiliency and persistency in the team. Right. We've extremely few people have left us over the years that like told us, okay, we can do this. When we first wrote our first training code base completely from scratch, it wasn't a fork of any open source. It was just like, okay, let's build it from scratch. I remember we had this one moment where we spent three weeks working on an optimizer bug. Like, it was like training just couldn't get stable. We like obsessed over it and we thought like, maybe we were wrong, maybe we should have just forked this repo or should have done. But then when we solved it, I still remember at the time we were like five people in the company when we solved it. We're like, oh, we can do things if we're just willing to work hard. And I think that culture with a very strong engineering bias has kind of helped us get to where we were. And so there's this kind of notion of open source and talent and these things. I think we just took different decisions from a different starting point and I think we are lucky. I do want to definitely call it lucky, but also a lot of hard work at the team that now that's starting to kind of show up in
- [13:51]
A
results just because we probably won't revisit this again. And this is a fun recruiting challenge. If someone knows the answer, what was the bug? And then we won't tell the solution, but also the bug.
- [14:03]
B
You're going to test my memory here, but I think I can recall. So if you look at so if you take Adam as an optimizer, you have epsilon, which is in the denominator. Exactly in the denominator. And at the time, if I recall, you looked at, like, the early llama papers and things like that. People were juicing epsilon, like, quite a bit. Like, they were, like, adding. I don't know if it was E minus 4 or whatever, like a high value for epsilon. And if you think about this during training, it's kind of like a bit weird and counterintuitive that we're adding noise to our optimizer by just adding effectively, like a random number in the denominator. Right. Like behind the decimal point. And I don't recall the exact bug, but what I remember is once we solved it, we no longer had to choose epsilon as much as was happening in the llama paper and other places. And it was one of those fundamental moments where we had trusted this paper that was out there, and we're like, oh, no, it has to be this way. It has to have this high value of epsilon. But it made no sense to us intuitively. Why do you have to have this so high? If you're just trying to avoid division by zero, why can't the value be extremely small? Uh, and. And. And that was kind of like one of those moments where you realize, like, okay, finding things out from scratch yourself builds a better intuition. Because the one thing you learn very quickly with model building is that your intuitions that you start with are gonna get beaten up so hard. Right? Like, the. It's such an experimental science that the things that seem obvious, you very quickly get to learn, like, you were wrong, and hopefully you figure out why. And sometimes you don't even.
- [15:45]
C
Yeah.
- [15:47]
A
So one of the reasons that when you released your new models, VIBU got really excited. I mean, everyone got really excited, but VIBU led our paper club on it. You guys saw it, obviously. Maybe talk through some lessons learned in that whatever you can sort of disclose, we can focus on the model factory stuff. Whatever you think is a good starting point.
- [16:08]
B
So I would say that our view from very early on in the company was that model building is ultimately 90% engineering. And I think we all know it in the industry because if you look at where every researcher is spending their time, they're spending their time writing code, right? Looking at data and writing code. And so we kind of said, okay, the state at the moment, like, three years ago was bash scripts and Slurm and spaghetti code, basis for training and, like, data pipelines that were kind of patched together. And we kind of looked at this and said, well, ultimately model building is a process. You're going from raw data, right? Like pre training, raw material, the web, et cetera, you're doing a whole bunch of filtering, cleaning up transformations, analyzing these days that's, you know, far more complex than it was three years ago. Then you're training a model which is effectively a large distributed systems problem, right across hardware that has still it's become a lot more reliable. Was extremely flaky back then, and now with every new generation we get our new sets of challenges and then you go into the next stages, right? There was no mid training back then, but like you got your post training and then your reinforcement learning. And so we kind of looked at this and we said, well, this looks like an industrialized process, this looks like an end to end process, that every single part of it kind of has its machinery, right? If it's your big data pipelines, if it's your crawling ingestion of the web, if it's your large scale distributed training, and then you've got your reliability. And we said, well, why don't we take some of the world's smartest distributed systems engineers that we knew and make them part of the process of research from day zero, not retrofitting it later on, but like really from the beginning. And that became our model factory. And so our model factory started with a handful of components. Today it's thousands of components. And I kind of try to equate it to, if you think about like someone who was at the very early days of Foxconn, if they had been there for the following decade, they would be able to rebuild Foxconn because they saw every decision that led to building that system and all the complexity. If you and I walk on a Foxconn today, no chance, right? Because we don't have the lineage and history of decisions that led to that. And so we kind of built early on from the beginning with a team that really understood that, well, the metric that we are optimizing for is the speed of an idea from a researcher to an experimental result that we can trust to then being part of the next model training. And because it's such an experimental science, ultimately in the beginning, when it wasn't that complex, you could kind of patch your way around it, right? But now at any foundation model company you are running, I mean we're small team, right? We're less than 70 researchers, another 35 engineers, and we were running. I haven't checked the latest count, but far more than 10,000, maybe 10 to 20,000 experiments a month don't wake up. And so if you look at that scale of every model run that is like it's ultimately it's, you need to be able to trust it, it's an infra problem. And so what we have now done over the years has gotten really good at that and just by working it and improving it and obsessing over those kind of end to end decisions. So now what that means is that you looked up Laguna XS2 that we launched. It was five weeks from the beginning of pre training to launch. The model that we're going to talk about today was eight weeks from start of pre training to launch. We started the next model literally yesterday because we now finished the post training required for the model we're launching next week or by the time this comes out today. And we moved that compute to the much larger Laguna M model that we're now training. And so the model should be an artifact of someone's process. It shouldn't be really a thing in itself. Like, and we kind of treat this like the way you would look at like a SpaceX factory where yes, the first rocket really hard to build, but the much harder challenge was building the factory. And now they're rolling off and no one is really thinking about the next launch anymore. So it's just another launch, it's another launch, another rocket comes off. And that's what we're trying to do with model building. And what has been which was not planned from day zero, it was kind of in the back of our mind like this will happen one day is that, well, when you build a really good end to end model Factory, really good APIs and really good engineering systems. Well, what is it perfect for? It's perfect for agents because agents are now starting to take over more and more work in our model factory. Yeah, so I look at the screens when I walk, like when we're really come together in our monthly, we do monthly on sites and I kind of walk behind people's screens and I stop by and I talk to our researchers. And the default is all of these different agents running on their screen that are writing the code, they're launching the jobs, they're evaluating the results that are coming back from the model runs. They are, you know, making the changes. And we're still in the driver's seat, we're still coming up with the ideas, we're still helping with the debugging, but more and more, and this is right now very profound on the data side of our, of our pipelines in both pre and post and the synthetic data pipelines, it's starting to become more on the architecture side as well. You're starting to kind of see these twinklings of what RSI is going to look like. And that's frankly. So when we talk about like to your question about our models, I really talk about the model factory. And my coolest example of these things is always that when we kick off a new run, doesn't matter if it's a pre training, like big run or if it's now a post training, like one of 10 post training versions we do for like pre release or many experiments, is that at any given moment the changes that somebody made that they had experimental results on from the day before make it into that run. So there's not like a cutoff 90 days before or like no, it's like literally from that moment because we can now trust the machine enough. And then you also have to invest in a reliability. So one of my favorite metrics about like Laguna S is that there was no on call events, right? Like completely zero. And actually we haven't had a meaningful on call event like someone had to wake up for as far as I recall, this entire year. Now there is one asterisk to that. In usually the first six hours of launching a new model run, something breaks because you set a config wrong, you made a small mistake, et cetera. So that's usually there's a little bit of intervention, but that's always within like in call periods, right? Not on call. And I think that's starting to now compound. So the model we're releasing now, I love it, it's amazing. But we're already on to the next one and I think that's the way it should be.
- [22:50]
C
I think I also just want to point out, so for context, this was like a month ago. We found it actually in the tech report. So we just came in with, okay, new models dropped, haven't heard about it. We're very much like, ah, okay, look, it's like, you know, on par with Kimi, Deepseek, whatnot, the small ones, Gemma level. It's a very cool paper on what goes into building. And then we hit this page, right? Like literally page two of tech report is. This process allowed us to build the small model from SC from scratch to delivery within five weeks. Applying the lessons. And then I'm like, oh, this paper is not about. Here's a tech report of benchmarks and here's how many tokens it was trained on like for people that want to dive more from what we're not going to discuss on the podcast. It's all laid out here, right? From custom software that agents can use to interface with training code, pre training data.
- [23:46]
A
We'll link the paperclip.
- [23:47]
C
Yeah, all that stuff. Read the paper here.
- [23:50]
A
But I love principles. I think that is a good starting off point for maybe telling some stories. Maybe we can go one by one past the principles. I'll just call out a Daxter. It's called Bought by a Prefect.
- [24:01]
B
It's kind of fun.
- [24:02]
A
But yes, I'm actually very familiar with Daxter. Just anything where like they trigger some kind of story.
- [24:07]
B
Well, I would say, well, experiments, code is obvious, but I think one of my favorite things is I don't know where it is in here. But early on, and I still think this is the case actually a lot of foundation model companies, people prepare their training data sets, they get packaged up, then they get copied over to a training cluster, distributed across all of the nodes and then training starts. And we looked at this like three years ago and we were like, that makes no sense. You lose so much time because the moment you have to rematerialize the dataset, you have to make a change, you have to fix something, et cetera. You've got all this time of like repackaging it, right? Token, tokenizing it, repacking it, moving it over to a cluster, then distributing it across the nodes. The bigger your clusters are, you start using fancy like torrent like algorithms to like distribute your data. So why aren't we streaming data into training something that's very common in just basic.
- [25:00]
A
Just in time.
- [25:01]
B
Just in time. Good computer science principle. And that was one of the first things that I think unlocked the model factory because the moment you start thinking about, well, a training job, doesn't matter if it's a big hero run or a small post training experiment, consumes a certain number of tokens per second. And it's actually not a lot from a data moving data perspective. So we said, well we have our training cluster and then we've got like our AWS kind of setup where we can build these amazing big data pipelines. We can set things up, we use spark underneath the hood, like all of these things.
- [25:37]
A
But when you say aws, it's not actual aws, it's your internal aws.
- [25:40]
B
No, it's our internal like just running like, like our infrastructure web services. Exactly. Our stuff. Running. Running on like an AWS account or on like any hardware, right?
- [25:48]
A
Yeah.
- [25:48]
B
Yeah. And so once we made that shift into I can stream data into training, all of a sudden you realize a lot of things unlock because now you don't have to wait for the whole data set to materialize. Now all of a sudden, when you're running data experiments about mixing data, it's a config because you've got these data sources that are coming in and you just. We have the circuit called blender that's in the report where we then say, okay, for this run, I want 20% of this source, 10% of this source, I want this much. So many epochs of repetition. I want this to be, you know, shuffled in a certain way. And your training job can start while the rest of the data is even still materializing. Also, what it does is because all of this underneath. So for us, we treated the data layer underneath as like an immutable data layer. And that was really important. Like experiments as code, immutable data later means that you can always go back and understand literally down to the single token at which cursor it went in on which version of the code.
- [26:48]
C
Yeah.
- [26:48]
B
And it took us a while, I have to admit, like the first year of poolside, we understood that engineering had to get great, but we didn't understand yet that this is ultimately in support of like a good rigorous scientific progress. We were quite a few, we were a very small number of people. So a lot of it was YOLO ideas and YOLO runs.
- [27:09]
A
Yeah.
- [27:09]
B
And we built great infra for the YOLO runs. But once we realized that, we treated data as immutable and code as always version, and you could always track and trace every experiment end to end, perfectly. You could repeat everything perfectly. Right. You have perfect reproducibility. I can still reproduce runs from two years ago if I wanted to. It actually enables the scientific progress, like the scientific process. And I think that took us probably about a year, year and a half into the company to figure out. We also had some great hires, like our co head of Applied Research, Nikolai, joined us from Yandex. We've been working on language models since like the early 2000s, I think brought that into the company of like, hey, we want to have even more rigor. And then once we kind of had the combination of like increasingly more capable platform that allowed people to do more, but had this immutability, we were able to start actually. Okay. Every experiment is truly an ablation. We truly need to understand it. And I think we became much more scientifically rigorous in the last couple of years. And the Infra underneath enabled it. And then there's just fun stuff. Like yeah, a lot of it's fun.
- [28:17]
C
Like even just the one, you share all the ablations. Two, picking the data sets, right? There's like a random small paragraph in here where it's just like, oh yeah, pre training data. We have an auto mixer. You know, it trains eight small models, scales them up, picks the pre training data set. We don't even need to look at it. I'm like, wow, a lot of engineering rigor there. And there's just, you know, there's just a lot more. Yeah. And look.
- [28:41]
B
And we want to put out more like actually we treat writing papers as something that we haven't earned the right for yet for a long time. So you earn the right to spend time publishing research once you're at the frontier. Because until then you're catching up. And every minute and hour in this industry matters. I obsess over it. Not just the wall clock time from idea to result, but just general time. Every day that we waste is one that doesn't allow us to catch up. But in this case, we said, okay, we're going to give ourselves. I think we gave the team like three or four days while still doing their work. Like give everything in there. And to your point earlier, if you know your stuff, it's easy to like put it out. And so there's so many more things that we want to talk about over time and we will definitely start doing and as we earn more of the right, but also now have like added to our mission that we want more foundation model companies to exist, you'll see us like be way more proactive and just trying to keep dropping some of those like things that we've learned along the way that can help others like
- [29:39]
C
speed up, which is the other cool side of this. Right. It's not like back to your point, it's not just here's the benchmarks of our training. If you want to replicate here's experiments of optimizers, data sets, post training, you lay out a lot of it here alongside here's your system for how to do it. You know, so it's really like promoting other people can do the same.
- [30:01]
B
And by the way, I also want to make clear, right, we have been incredible. Like we've taken a lot of advantage of the fact of all the open research the others have published, right. And, and, and you mentioned, you know, the, the Chinese labs and we. I think it's important that there's, you know, from every country and every culture and background, including like Western companies like us. There's different models that come out that people can choose to trust. But I think we do have to give credit where credit's due. Right. The, the incredible Chinese lab have done an amazing job at sharing their research and we have definitely like been on the receiving end of taking advantage of that. So when you're on the receiving end of something coming to you, I think it's. You also have kind of an obligation to give back.
- [30:39]
A
Do you have a favorite or underrated Chinese lab that you want to shout out? Everyone shout out sleep C. That's a good question. Muon, obviously for.
- [30:48]
B
Yeah, yeah, look, I think,
- [30:51]
C
I think
- [30:51]
B
obviously everyone's been talking about zifu lately with GLM 5.2. I think what most people don't realize is when they started. Yeah, right. They started years before ChatGPT.
- [31:02]
A
It was just rebranded.
- [31:03]
B
And so I have like, I remember how hard it was to work on these things before the rest of the world got excited about it. And so I have an immense amount of respect for people who were working on improving models when it wasn't the sexy thing to do when believing in LLMs, you know, was going to get you ridiculed. I remember back in 2016 when we were doing what we'd call machine learning on code with some of these models, people would just frankly laugh at us. They'd be like, this makes no sense. Why are you wasting all these millions of dollars on trying to figure this out? And so I would say they're probably the one that I think deserves the shout out. Not just because their latest model is very good, but because they fought to get here. And I think every foundation model company, it takes time to get here. It took us three years to get to the model that we're now going to be releasing. And now the time in between the models is counted in weeks. It's no longer counted in months or years. But this stuff's hard and if we can make it a little bit easier for the next person, we should all do so. Because if we don't do so, we've got a small window before models are really impacting recursive self improvement to a level where catching up otherwise might become unfeasible. And we should try to in that window, encourage as many Neolabs orever we want to call them to start. And so one of my kind of current mission but quality is like I want to encourage whoever is a researcher right now who thinks they can actually tackle this to go and leave and Become my competitor, like start another foundation while coming because I think we need it. I think otherwise we're not going to be in the world where
- [32:53]
A
I don't
- [32:53]
B
want to just be the fifth or the sixth company that wins. I want to look at a world where there's lots of choice.
- [32:58]
C
What else do people not see in starting a foundation model? There's a lot of capital required, a lot of compute, you lay out model factory and how to do the training, but there's a lot there, right?
- [33:10]
B
Well look, this is an oversimplification and I always asterisk it with that because it can land a little bit the wrong way in people's minds. But I actually think you can sum down some 95% of model building to just doing. You're just doing two things. You're improving data or you're improving compute efficiency. And I know that feels like an oversimplification for the incredible gifted and skilled work people do. But if you really look at it like what are we doing? We are looking at data, we're generating new data, we're improving data. And the only way to do that is to look at the data, right? That's a big part of foundation model building. And on the other hand we come up with these incredible breakthroughs in inference, in architecture and new attention mechanisms. But what are they really doing? They're bringing compute efficiency. Now we have definitely had some breakthroughs over the years that allow for more model capabilities, but at the limit. If you could train a large enough model, right, like you had infinite compute, we probably, if you had infinite compute, you'd be at AGI probably already tomorrow, right? Like it's not. And so, and let me say that infinite compute with infinite ability of much faster networking, because networking ends up being more of the bottleneck than compute. But so I do think that those are the main things. And to just realize that this is engineering, I think it's become more obvious. But I think for quite a few years people have held foundation model companies and researchers and others on this pedestal of like you're doing credible magic or rocket science or only like, you know, Nobel laureate physicists can do this. And don't get me wrong, there are some really hard problems that need to be solved. But a lot of the work that all of us are doing on a day to day is not sitting down trying to solve a math theorem. A lot of the work that we're doing is just really doing the basics right? Writing good code, looking at data, improving it, running experiments, looking at plots, trying to See like hey, trying to shape our intuitions and a lot more people could be highly capable researchers. And I think that's. It feels, it feels far for people to do so. But I've seen in our own company we've seen engineers become researchers because the model factory allowed them to be have a much lower hurdle of running experiments and trying things. And one of the guys on our team who started as an engineer building our agents is a legit reinforcement learning researcher now making real progress. And that happened in the span of like six months. That would have not been what I think most people assumed was possible a couple of years ago.
- [35:46]
A
Yeah, I think one of the interesting moments is when you can sort of self host like you know, for the programming language. Like if you can compile the language in the language the equivalent is can you use your own tools? Right. You have to pull cli. You have your own models. Presumably you're not only using your own models. There's no way. But what's that percentage over time?
- [36:11]
B
This is the first model that we're releasing that is starting to meaningfully contribute to our own work. It's not state of the art model yet. Fable and other very capable models. But Laguna S is really interesting. I'm going to actually pull up the quote. Peng Ming, one of our co heads of applied research said something last week as the. As the model came out about 10 days ago, much better than frankly we had hoped for, expected. And he said I have the feeling that a lot of the gains in Laguna S come not from more intelligence but more from different behavior, more verification, less taking things for granted, not declaring victory early and being way more persistent. And to be honest those are more predictive than raw intelligence for success in human also to some degree. And this was. He wrote me this on 5th of July on a Sunday and it's kind of been burned in my brain ever since. Because the Laguna S model as you'll see it and why it does so well in benchmarks and why it does so well in using it on a day to day basis is that it's just incredibly persistent. It reasons a lot. I do call that out. We have work to do on making it more efficient. We have work to do on offering different reasoning modes. But this is the model that has been able to do things that I never thought it could do. 118 billion 8B active model, which is not that large. It fits on a DGX Spark and still runs at you know, 30, 40 tokens a second on a Spark, is able to solve Erdos 397 independently it's able to do complex programming tasks. It's able to. I asked it this morning to make me a WI fi scanner without using any external libraries on my Mac. And it's like figuring out like the core WLAN API by really persistently trying to understand it without access to the Internet and more and more. I love vibe checking. I've probably spent 8 to 10 hours a day with this model for the last 10 days. I'm not exaggerating. I was on my 11 hour flight yesterday. I spent 10 hours reading trajectories and traces and like of the model and what I take away from it is exactly what Peng Ning said. We are going to be able to squeeze so much more out of smaller models than I think we had imagined in the industry. Because yes, there's intelligence and larger models are more intelligent. No doubt about it. We should continue to scale up. But the behaviors of being really persistent, of being able to backtrack when you're wrong, of understanding how to interact with your environment, show us that we can get a lot more out of it. And this for me has created a bit of a question in my mind the last couple of days. If you think about where we're using models today, right, we are using models, say for knowledge work represents 25% of the global economy, you know, $25 trillion of work. As we scale up models and they become more intelligent, we are excited about using them more and more for pushing the frontier of science. And if you look at the frontier of science, like true breakthroughs in science, they are linked to more intelligence in many places. Einstein figuring out general relativity is able to bring ideas together that other people would have not brought together. And I think one of the many dimensions of intelligence is the ability to do that. And it's something we clearly see that as models get larger and more capable, they're able to pull more ideas and threads together that a smaller model wouldn't be able to. And we're starting to see examples of that in medicine and like in bio and other things. But if you think about the majority of knowledge work that we do and it includes building software. I'm a software developer at heart, first and foremost, probably, although I probably can't say it that much anymore because I don't write production code in years, is that what makes us good is actually our persistence. It's our ability to encounter a problem and backtrack and say I need to go figure out this bug, I need to go research this. I need to go look at the documentation. I need to try different, five different ways to see if I can solve it. But it is not necessarily bringing three ideas together from radically different fields. And so if we are now seeing, and I think Laguna S is an example, that we are able to make a relatively small model, much more capable than I had definitely predicted or any previous benchmarks had shown for any model remotely this size or even larger, at least on coding tasks, that it's because of the behaviors. And so now the question I have, and I don't have it answered is I know at the limit. So infinite model size, right? Extremely large model and the cost of that model is going to be very expensive to run, we notice, right? So larger model roi. So I know that at the very limit I'm not going to use the world's largest model one day, quadrillion parameter, whatever. Crazy like skill. We scale up to do a basic coding task. Already today I'm starting to size down for certain tasks. So it means that there is an optimal, it means there's some curve that goes as we go up to model size for knowledge work, at some point we're at the peak and after that the return on the investment of using a bigger model just doesn't make sense. Now I think the question is before I would have thought that peak was extremely, very far away. This model for me is the first sign that maybe that peak is at trillion, 5 trillion, 10 trillion. Maybe we can just squeeze way more out of these models. I'm no longer thinking that we need two or three orders of magnitude on the largest models to be able to, you know, solve knowledge work, the accounting, the legal, the code that we write. And so if, if that holds true, it is an argument for the commoditization of models. It's an argument that open source can win and like succeed in this world. And now it's of course a self serving argument and it's a hopeful argument, but theoretically at the limit it works. We just have to go discover in the next couple of years of how much more we can squeeze out. Now I do want to put a big asterisk. This does not mean I'm against scaling models. I think we ultimately only succeed if we scale our models as large as our competition. I do not like, I think we should not put our head in the sand and say we're going to be king of open source small models. I think that's frankly it's a cop out. It's trying to be king of your own kingdom but not Realizing what the rest of the world's doing. All of us rather use a smarter, faster, you know, more model. But it's kind of a sign of hope. And so I don't want to overly state this is a good model. We have a long way to go to get to the state of the art. But what hopefully people take away when they use this model is that the behaviors inside of it are what push it to be far more capable. Less than necessarily the number of parameters.
- [43:03]
A
Is that mostly post training? Like. Yes.
- [43:06]
C
Right.
- [43:06]
B
So it's entirely post training.
- [43:08]
A
Are we done improving anything on pre training? Is pre training done?
- [43:12]
B
No. Okay.
- [43:14]
A
I just wanted to cover pre training and then we go post training.
- [43:16]
B
Pre training is not done. I mean, look, there's a part of pre training of just dealing with skill. Right. Every new order of magnitude of model skill, you are going to get new things you got to solve for. Those are ultimately engineering challenges. I have, I would say a not commonly held opinion that reinforcement learning will move earlier and earlier into pre training.
- [43:42]
A
Yeah, mid training.
- [43:44]
B
Not even mid training. Like mid training today is, if you look at. So we've been working on this for years already and I think the first time we saw it out in public was the deep seq 0 paper, which is a year and a half ago. I think, if I recall correctly, where you know, you can very early on in a model, as it starts, capable of being able to use language, et cetera, induce reasoning. And so the question that I kind of have is like we have the data set that's the web. And the web I think we could arguably say probably has the totality of humanity's knowledge somewhere encoded in different places. It's a huge variance degree of quality from garbage data. And once you look at pre training data, you really get humbled of what the web is to the most greatest scientific papers and the best blog posts and best transcripts and whatnot. And so now what we are trying to figure out and have been doing a lot of work on and it's a place where maybe not as open as we're on other things, but we will become more over time. We've been spending a couple of years really doing research on how can we turn the web into not just next token prediction, but into a way to teach the model to think earlier in its training. And I think there's a huge amount of gold to be found there. I think we are right now in. We've got some drugs in the industry. One of the drugs is distillation. Another drug is More environments and they're great and they make us feel good and they make the models better and we're all addicted to them and we'll use them right in various different ways. But ultimately I think we are still barely squeezing out of the web what we should be getting out of the web. I think just next token prediction during pre training is not enough. And so I think we'll see some very interesting things still happen. And that RL in post training to induce behaviors to improve things. I think the whole world knows how to do this now. I think we're scaling it up. Everyone is. But I wonder if we need to go as far as we're going today with environments. I'm not sure yet if you mean
- [46:02]
A
you're going too far.
- [46:03]
B
I'm not sure if the path to AGI is just more environmental environments.
- [46:09]
C
It seems like a never ending. Okay, I want instruction manual for this table, right? Am I going to environment out building furniture or are we just going to tail end? Like we need some general solutions.
- [46:19]
B
I think, I think there is, I think there's an ability to generalize more from the web. But I also, I'm very encouraged. Like when I look at Laguna S and which is post training is what is the big impact there. And I see like, oh wait a second. Just by making some of these behaviors much better, we're able to get so much more out of it. It just changes a little bit the way you think about intelligence.
- [46:40]
C
The analogy people draw often is the RL phase is where you don't learn as much new knowledge. You reshift.
- [46:46]
A
Yeah, yeah.
- [46:47]
C
So you know, you reshift distribution and you can have it reason towards what you on your point about mid training, a lot of mid training is still just continue pre training in a domain, say medicine, then you do RL so still just.
- [47:01]
B
It's just better data. Right. Like I mean mid training. I like how we invented this word. Like it's effectively just like you know, second phase. It's a second phase of pre training with like a really dumb way to do a curriculum. But like ultimately what you'd want is a curriculum from token 0 to token 30 whatever or 40 trillion tokens that really truly is the optimal curriculum for the model to learn. But mid training is essentially a two stage curriculum on the web because we do not have to compute effectively to try to ablate the perfect curriculum.
- [47:36]
C
Right.
- [47:37]
B
And so I'm pretty sure that you'll start seeing people talking soon about some other term in a tour because now we do this right we talk stage two and stage three and stage four mid training and like. But ultimately all we're doing is we're trying to assign a curriculum to the web data that we have to allow the model to learn better. I think at some point, as things get compute, as models get cheaper to run as next generations of compute, this will become more of a continuous spectrum. I also think the reason, by the way, you have mid training and like stage two and stage three is organizational, right? This is, I think a thing where that we really try to avoid with the model factory is like mid training exists because there's a MID training team now, right? There's people or like people in between who decide to focus on like a MID training effort. But what you really want is engineering and skill of experiments that allows for a much more continuous spectrum that you don't. You have infinite stages now. We're not there, our compute's not there. Organization design is not there for it yet. But I think we'll get there. We'll look back on a couple of years and be like, oh my God, it was so cute that we did our pre training data like this in such a naive way. Like we barely ordered it. We did really do a good job at like kind of building that curriculum in the industry.
- [48:52]
A
And I'll confirm that, you know, when I talk to some researchers that this is a lot of the focus now is like, how does pre training change and what is the next objective other than next token prediction? I assume you don't have the answers, but you have some ideas.
- [49:03]
B
We have some ideas. We're not ready to talk about it yet. We've been working on them for years. And I think that's the one thing that's also like you asked earlier about like, what's not obvious about building a foundation model company is that you are constantly balancing the table stakes. Work. The recipe, you know, works.
- [49:19]
A
Yeah.
- [49:20]
B
Versus like your breakthrough cured research. And finding that balance and adjusting the percentage to it based on where you are in the race is really important.
- [49:31]
A
I mean so like this is a nice way I was going to bring out autoresearch at some point as another Andre invention, coinage, which is like, honestly, how many objective functions can there be, right? Like just try a thousand of them. Set it running, whatever, you know what you're looking for. You're looking for lost curves like that.
- [49:50]
C
Like it's also a thing people take bets on, right? When you say more Neolabs, you're doing a version of we'll do foundation models, scale them up Next token predictors. A lot of other neolabs that we see want to take a completely different approach. Right. At some level you're right. It's all compute efficiency and that's the net objective. But you know, some are okay. Different architecture, like vastly different amounts of compute spend. So some are different. They're not just. They're like, you know, 99% not balancing. Here's the vanilla and scale up there, 99% on here's novel research that'll change everything.
- [50:27]
B
And I think Luke, I think you. It depends when you started as well.
- [50:30]
A
Right.
- [50:30]
B
When we started like the novel thing we did was reinforcement learning on code. No longer, that's no longer novel by far. But we would like we. That's where we obsessed over when no one believed in rl. So you have to kind of when you start the company, you have to have your own idea. You have to have something that's different that allows you to speed up.
- [50:48]
A
Right.
- [50:48]
B
For us it was RL to LLMs that later became common. Like you know, knowledge. But in the beginning it wasn't.
- [50:53]
C
It's cool. You know, this was like your original 2023 blog purpose. Yeah. And like you do lay it all out here. The dog is pretty underrated, right? The whole RL on code was very, very early on.
- [51:06]
B
Very, very early. And even we had to argue with people. We say here things like to push beyond current capabilities, you need to train your own foundation model. We had to argue with people that it mattered that you had your own base model. You can't fine tune your way to success. Right. Major capabilities emerge from training a base model made accurate and useful during fine tuning.
- [51:24]
C
Which for perspective at the time we knew closed models OpenAI anthropic were huge. The open models we had were like Mistral 7B, a 30B, a 70B.
- [51:35]
B
No, when we actually the date on this thing is wrong. When we published this, it was April, April 2023. I think this is just happened on a migration product on archive.org Mistral had started. We started on the same month. Right. So this wasn't even. There was only I think llama out at the time and that's it. Right. And so but I agree I think we want, we want as many diversity of ideas and I do think if you're starting today, you want something that gives you an edge. Right. And what I do think we sometimes over. I think every architect at the limit, every architecture works and RNN works. It's just not compute efficient. Right. Like if you had infinite compute, you could probably just take a basic RNN from back in the day and you could get pretty far
- [52:27]
C
now.
- [52:28]
B
There have been, you know, meaningful breakthroughs, attention, other things that are there. But you. I think we are, we're still, we're still very early in figuring these down. The things I'm most excited about, I'm most excited about people doing extremely low precision training. Right. So like the ternary stuff that we're seeing and it was very cool. The Banzai stuff yesterday was super cool to see. I think there you can find tweets from me going back to 2023, which is like kind of the notion of like. Well, it's an obvious trade off. Bigger model, lower precision equals, you know, smaller model with higher precision by definition, right. It's just what is like how does that actually play out? Right? What's the actual size limit? So you now have companies that are trying to figure that out. Like those are the kind of things that can change our industry if they're done right. Because ultimately like our bottleneck on compute is, is, is a map mole bottleneck and a networking bottleneck. And the moment you start doing those things. So I'm excited about that. We're not doing any. I mean we're doing the usual like Laguna S was trained in FP8. Only thing that in this run, I have to admit that wasn't FP8 was the all to all. In the new run we just started yesterday, the FP8 was all to all that was just like cut off day. Like ah. We're not perfectly comfortable wanting to do it. You've got amazing work by Nematron and NVF before training. Like I think it's underrated what they've done there. I'm excited to get to NVF before training. Doesn't make sense yet because we're still training on hoppers, right? We're like relatively small. We're a 10k H200 cluster company right now. We'll be scaling to a lot more
- [54:00]
A
soon
- [54:02]
B
and really a lot more. Someone is thinking about applying for a job.
- [54:07]
C
But
- [54:09]
B
yes, I think there's so much more juice to squeeze out of this. And hopefully Laguna ES shows people that a model at this size can get a lot more. And we did this thing in eight weeks. We think there's a lot more juice to squeeze out at any model size. We're now scaling up because it's the most optimal thing to do for us as a company. But if I had infinite time, I would love to push more the capabilities at other Model sizes.
- [54:35]
C
I don't think we've properly announced what your new size is. So we have xs which was 30B ish old medium was 200B which is going to be deprecated it seems so new.
- [54:48]
B
Laguna S, Laguna Small 118, 118 billion total parameters 8B active. So very sparse. It's a scale up of the excess architecture. It's the kind of classic or call it classic these days, like three to one ratio of sliding window attention to global attention. It's a nice size for a couple of reasons. One is just very cost efficient. For us it was a good way to do. We wanted to get our progress out quickly. One of the things that we've seen is that it's a balance inside a foundation model company between focus on releasing and shipping and like your new novel research. But with the model factory we are able to like treat the release of a model as less of a time investment from the team because it's just oh at this moment in time, you know, do the pre training run done, apply the latest post training. And so this is I think a nice weight class. It's one that also will fit on the DGX Spark which I have a small like soft spot for. I love having that little thing like you know, run a good model.
- [55:53]
A
Yeah, we covered it on this pod GTC last year.
- [55:55]
B
Nice.
- [55:56]
C
I think GPT OSS120B was the first because it's a large single GPU which was the H100.
- [56:01]
B
Right.
- [56:03]
C
Rent one H100. Now you've got 128 gig Macs, Mac minis, Sparks. It's the home sweet spot.
- [56:10]
B
But I think what I'm most excited about is that this model hopefully shows people what is possible in this size. Because when you look at the benchmarks and start using it, you'll realize that we are outperforming models two or three times their size.
- [56:24]
A
Yeah. And they think so. For example, today's thinking model is like a trillion params so.
- [56:29]
B
Yeah, exactly. And look and by the way, I'm excited about. I have it just came out. So for those of you listening to this, I saw it on my phone. If you're humming in two seconds. I haven't even had a chance to read the post.
- [56:39]
A
But somehow not only you're better than Thinky, which is one of those benchmarks, but also on certain benchmarks like the Tao bench one, you're actually state of
- [56:50]
B
the art look we're doing. I'm not sure if we're state of the art on, I mean Tau3 banking. I haven't checked where we sit on the leaderboard but I think we are within our weight class. I feel very comfortable to say and even in some weight classes twice larger that we are probably state of the art. I also want to caveat this like you know, best model still in the world right now is definitely, you know, give me a fable, give me a 5.6. To your point earlier, we also use other models.
- [57:15]
C
Yeah, I think that. So the interesting thing you mentioned earlier is you're starting to shift a lot of your actual usage to it, right? Benchmarks are like, they're good to compare but they're not super realistic.
- [57:25]
A
They have to, right? This is how they're going to dog food benchmark.
- [57:28]
B
No, you have to like you have to use your own models and you have to have your own internal evals and benchmarks. And what the funny thing is like within first 30 minutes of a new checkpoint coming out that's you know, the kind of first post train after a pre train you yourself can feel in the first 30 minutes of where this model is going to be. Like you don't know exactly but like when this one came out we were like oh like this is different, like and I think that's, I think that's the best example. But it's a little bit like, you know, like your kids, I don't have kids by you know, parents. Like they see their kid and it's perfect and they love it. And then like you know, they don't see all the rough edges. You always get that when you build your own model. It's the most fun part is that you like you love a little bit. Every model that you do we try to kind of say this thing constantly. It's like it's the worst model we'll ever train. And so I know the team now is like already onto the next one as it should be because this is a race and this model is a moment in time that hopefully shows people that we are serious about this race, that we want to work really hard at it, that we want feedback. Right? Where is it good? Where is it not? Like one of the nice things about having your models out in open weight out in the world is that you get a lot of feedback.
- [58:40]
C
How do you think about building it with like working with a harness, right? So open code codex, you have your own pool cli tool basically getting people to use it. The co design of model harness training it in.
- [58:55]
B
So you need to do some multi harness training like if you, especially at these smaller sizes, like you want to do a little bit of multi harness training for these models to just get the right. And it's very little, like you don't need a lot, but it's just like to get the right behaviors that you see in your harness. Transferring to the harness that like you other people might use it in. We internally have been kind of this calling, this polishing which is like you've got your model and you do a little bit of polishing so that like it's able to work well in other harnesses as it is in your own. No doubt it's going to be better in your own harness. And it's just because of where are you putting your reinforcement learning compute. You're putting your RL and your synthetic data, you're putting it to your own harness because it's the one that you understand the best and you're able to kind of push the most because that end to end control is what allows you to make it better. Then transferring those capabilities is more about just making sure the model induces the right amount of reasoning and understands some of the maybe more complex weird tool call formats that might exist somewhere else. And so we do do some multi harness polishing as we call it. It's not really what drives capabilities, but it does create a better experience. I think, frankly, I think everyone probably does these days. But it is totally fair to see why your own harness is going to still be better than others. And I think we see this with all the foundation model companies. And it's just that when you are pushing capabilities, you don't really want to trade it off by putting 10 harnesses in your RL runs. Because it's just complexity. It's complexity of engineering. Because when you're trying to do good science, when you're trying to really understand what made my model improve, you want to make one variable change to something you understand and a harness from someone else you don't know or understand in the same way as you understand your own.
- [60:46]
C
Right.
- [60:46]
B
They might have different sub agents or different prompts.
- [60:49]
A
If it's open source, you can look at the source.
- [60:51]
B
Yeah, but it's time, right? I really cannot stress it. I know I'm a weird person on this because I have friends like can we meet up or can we do this or can we go? I'm like, no. Because ultimately this is a race and time is the only thing that matters. And if I look at our team and say, okay, what is complexity worth introducing on our general trajectory to building more capable models which generalized to other harnesses quickly. And by the way, our model works well in other harnesses. I really encourage people to do it. It works well. Like I've, we've been testing an open code and kilo code and others and like.
- [61:22]
A
And clothes which just got bought today. Exactly. Everything's getting bought.
- [61:27]
B
Exactly. And it's. And I think part of that is like, you know, and there's some amazing. I'm excited. Like I think her mess is actually a ridiculously cool harness. Like. And so, and you know, part of
- [61:38]
C
the question was actually just like how much effort is model versus model plus harness. Right. So new benchmarks like agents last exam. It's not wanting to just measure the model. Same with models getting more and more agentic. They need a harness to operate in. Right.
- [61:55]
B
So I think for when you're asking that question to a model company, I think you can separate it in two parts which is like the harness. We have a very slimmed down harness. When you look at it, it's like six tools. It's like shell and like shell, kill shell, weight, write, fetch, web and like I don't know, bash. Like I think I'm missing one. But like that's effectively all the tools and it's very simple, it's very lightweight. So it is not a harness that is designed to try to do well on a benchmark or try to do well on a certain subset of things.
- [62:28]
C
Right.
- [62:28]
B
It's not a deep research harness. So I think we see incredible ability for complex harnesses that built lots of prompts around and extra data sources and other tools to really push capabilities of models forward. But our model is still better than some other harnesses who do that in coding like tasks because it was rl'd with it. Now I do encourage people. I think our model by the way is perfectly fine and goodnars that the differences are probably maybe too small for anyone to notice. But we see it ultimately still on benchmarks by a little bit. So I think it's both are true foundation model companies with their harnesses will really push them because it's just operationally frankly the best way to have scientific rigor in improving your models. But also someone who takes our model and really does a lot of work on improving a harness is going to outcompete us as they should. And that's just because the harness is the stopgap between what the model is capable of and what it needs as additional instructions. And what it needs is access to data and tools.
- [63:31]
A
Right.
- [63:31]
B
And that's ultimately, I think what a harness is. It's like, is it able as you build more capable models, you know, you're improving the instruction following the models. And so additional harness is just saying, hey, if you encounter X, Y or Z, behave this way. And so even if you would say that two models with two different harnesses can equally reach the same capability that you care about, a harness that is really tailored towards a capability will do it more efficiently. It's kind of like a person who's getting a manual of how to do the task in the most efficient way with the right tools and the right data sources, versus a really smart person. Go figure it out. They'll both solve the task, but one will do it a lot more efficient. So I'm a big fan of all the harness development that's happening in the world and we want to work with more harness creators to also make sure that if it needs some additional training like polishing, that we will do it.
- [64:22]
A
I mean, I think when you say it's a race, there's a question of what are you racing to. Are you racing to be the best coding model company or the best coding model plus harness company? I think that's those are different things
- [64:36]
C
or sell in the black.
- [64:37]
B
Neither. So I raced AGI coding for us since day zero of our website has been, and we've said this over and over again, we think focusing on coding and long horizon like software tasks is a path towards AGI because it forces us to solve the hard problems. It forces us to solve the ability to do extremely long horizon complex work that requires lots of reasoning, external tools, data, et cetera. And one of the things I can show you so we'll have a web chat on with this model. And I've loved this model for deep research just using it in my coding harness. It was never trained for it, it was never like looked at it. But it's great at it in my opinion because ultimately the skills transfer, they generalize. Now, where we are not focused on today is to make sure that the world's greatest medical knowledge is encoded in this model or the world's greatest legal knowledge. But it actually did. We won't be publishing this benchmark because we didn't have time to really do proper, but it did really well on legal benchmark and at least on our first runs. And we are very rigorous when we publish evals. We have checked them for every little thing. We have run them many times. We've passed like we've gone and we'll like we try to be extremely honest with this. So if we haven't spent enough time on a benchmark that we use internally that is public, we just say we won't publish it. And so I mean the other way
- [66:02]
A
is just to give it to artificial analysis and let them run it like third party standards.
- [66:06]
B
Oh, 100%. And we are going to be doing this as well. And still it takes time and effort. Right. Because you're working with people to understand like know the infra failures and like the, the tools they're using and like are they set up well. But I agree, you absolutely want to. I'm a big fan of companies like VALS and AA and like others that are doing this stuff.
- [66:21]
A
I thought very nice. You're the first to bring it up.
- [66:23]
B
Yeah, I think they're great. I think they've got like, I love like a lot of the work they've done and put out and so. And there's I think many more. And please create more eval companies. Like create more evals. I think it's so valuable for the industry.
- [66:35]
A
Actual monopoly kind of, I feel like. Oh and duopoly maybe.
- [66:38]
B
You know, I think it can be broken.
- [66:40]
A
Yeah.
- [66:40]
B
Because I think it can actually be broken really easily because creating an eval for many people isn't sexy work. But whoever does it, everyone is happy to get a good eval. If an eval is well constructed, everyone's celebrating it and everyone's willing to pay for it. And everyone's willing like the financial market.
- [66:55]
A
Yeah. Yes. I think creating eval. Yes. But like in terms of being like we are the industry standard ones that will run t bench and make sure that you didn't cheat. And I'll run it the same way that you run it versus your competitor run it.
- [67:06]
B
Yeah, that is very true and we need that. And it's actually nice that that's like a few standard places that we all have to adhere to. It keeps us all honest. I think that's super important to do so. But yeah, no, I think our goal is to build the world's most capable models. And right now we are focused on the coding agent capabilities, long horizon work. But what you see with that is that you get a lot for free. I've always said it's a lot easier for us as we get to sota and frontier on coding to then say, okay, now we're going to obsess in using the model factory to add more data for places that we're not as strong on. Like could be medical or legal or Any other areas. And similarly, I think what we see, and we see this with reasoning models a lot. If you give models access to the right knowledge sources and they have capable ways of reasoning, they're able to go very well into domains that are less known to them or even seamless in their training data. So. But yeah, are we agent, like model plus harness company? No, we're model company, but I think models today cannot be trained without harnesses. It's not possible. So it kind of is just like where before it was just the weights in the container. Well, now there's an agent harness that's attached to it. But I think there's a big difference in being an agent agent harness as a model company than someone who's truly building an agent company. I think they can do far more than we can.
- [68:28]
A
Yeah, understood. Yeah. I think that is my minor pushback. If you are truly identified as a model company, then make the best model for open code. Right. Instead of for pool or whatever. I think that's minor compared to actually, if the goal is AGI actually make the best model for Hermes. Right. Because that is the next stage after coding look.
- [68:49]
B
And we're working actually very closely with them because I do think. And you have to care. You have to invest in it. It's why we do the polishing and we spend time on it. And I think over time, yeah, you're right that you want to kind of balance that out. But ultimately you just want general capabilities that everything works equally in every harness.
- [69:11]
A
Just on the topic, do you guys do much with like, Hermes, openclaw, Nano, claw, whatever. PI. No, pies are different.
- [69:18]
B
It's more coding. I'm a big fan of PI, though, I have to say. I think it's a really. I forgot to mc.
- [69:24]
A
You sound closest to PI, pool and PI in terms of like the minimal surface.
- [69:29]
B
In the minimal surface. Yeah. It's because I don't. I have a. Allow me for one more strong opinion. I've been saying this now for two years. I think MCP and tools are stupid. Oh, let's go now.
- [69:43]
A
You support mcp.
- [69:44]
B
I support MCP and we support tools and everything. They make absolutely no sense to me. And like. And I'll explain a little bit why. And I think I can probably get people to come along on this one. If you are looking for complex tasks, increasingly longer horizon, increasingly complex tasks, doesn't matter if it's coding or something else. You are going to be interacting with data sources, right? And you're going to be interacting with things that are installed on some form of a virtual machine. And what we are doing is that we're putting a layer in between those things. We're putting like MCP in between, we're putting tool calls in between. And this is even more about tool calls than mcp, where the model can just write code and interact with the system. And we're starting to see that, like, Laguna S does this a lot. You'll see this as well in, like frontier models. They're increasingly no longer here. We're going to stuff 50 tools in the, like, system prompt to know, here's a virtual machine with these binaries installed, this code base. You can operate in here, a folder where you can write your memory if you want to. And the model is using code to do complex asks. And when it uses code, it is not one or two tool calls or three things that are chained together. It actually starts, you know, using if statements and for loops and making things conditional. And so I actually think we're moving from. We already are moving from tool calls, you know, to effectively models writing code, little scripts. And you see this a lot when you get the Python, you know, code interpreter. Exactly, like in just the arrow, arrow in, you know, written code and the file. I don't know what you call the eof. Yeah, yeah, exactly. Like, you already see this happening more in models because when you start training them in rl, the models want to be free. They want to be able to do the thing they want to do in the most efficient possible way. And it is not calling one of the 50 tools in their, like, system prompt. And so I'm a very big fan of give the model a minimal harness, as minimal as possible. Give it a container in which it has its own code base, right? The model's code base that has access to the API keys and data sources and little libraries and documentation that it needs, and just let it run free at the task. And I think that is the way we're going. I think we will in 12 months not see a single system prompt that is stuffed with 20 or 30 or 40 tools anymore.
- [72:08]
A
No comment, no pushback there. I think it will be supported for a long time just because a lot of people are trained on that now. But maybe you guys don't have to support it in your models going forward. But yeah, I do think that's writing code is more generalist and it's a means to an end.
- [72:26]
B
And we do support tools. And this is actually the first model we're doing parallel tool calling in which we needed to catch up on. So that's there. So it's there. But it's a personal nitpick. I want the models to have as many degrees of freedom and just be free and do capable things.
- [72:44]
A
Yeah. So that was on the path towards like okay, how do you use poolsides models and Laguna models for my Hermes or my Open Cloud, all those things. And so typically what I look for is computer use or vision. That's a very big one. You guys have a blog post on that. But then also the persistence I think is very strong value as well as long context which you guys have a million token contexts.
- [73:08]
B
Anything else, Luke? So for us, vision understanding is the next thing, right? Like we don't have vision understanding.
- [73:13]
C
So I was going to say is
- [73:14]
B
though we don't have vision understanding in these models yet. And so this is something that we've started efforts on. We think it's super important to have visual understanding.
- [73:23]
C
That's company vision.
- [73:25]
B
And so no, we've got work to do there. And this is actually one of the things I loved about the thinking model. Like from the two minutes I scrolled the blog post.
- [73:33]
A
Multimodal. They're very committed to multimodal. Including audio.
- [73:36]
B
Yeah.
- [73:37]
C
There's state of the art audio as much as it's a trillion parameter state of the art audio. But also all trained from scratch, right? Yeah. No encoder in the sense that's one
- [73:47]
A
of the strongest reasons why you need to trace from scratch is you just have a different tokenizer, you'd have different environment.
- [73:52]
B
Like I'm fully aligned. Like zero disagreement from me here. Like just add the modality and don't put, you know, keep it, keep it simple. I don't think we'll touch audio for a very long time.
- [74:05]
C
It's kind of in the name too, you know. Inklink Inc.
- [74:08]
B
True. Yeah, yeah.
- [74:10]
A
Why? What's so hard about audio?
- [74:12]
B
It's not about what's. Again, it all comes down to focus. That's easy, right? Like saying no to things means that there's a research or in compute that can go to making general progress. And our view is like general progress is going to come from the ability to push these models to far more capable reasoning, far more longer horizon tasks. I don't think audio adds to that. I don't think it pushes us close to AGI. I think it is a necessary modality as you get closer to AGI. I think visual understanding sits in the middle of those things. I think visual understanding can absolutely do so. But it also unlocks capabilities that are just Valuable today. So. But this is the point, right? You want more diversity, you want more different foundation model companies who focus on different things. I think we are just kind of like a horse with blinders on. Just like, yeah, you have your path, we have our path. We want to catch up to the frontier and we don't want to distract ourselves with anything else.
- [75:05]
A
Yeah, I will call out that one of the branches of research is deepseek OCR which is can you just throw away the text tokenizer and just only vision.
- [75:14]
B
I find this. I look geek. The geek in me looks at this stuff and it's like, okay, let's like look at the number of bits in code after. I think it's super cool. Right? But I think this is what we're going to come back down to. Like probably works. It's just is it compute efficient enough? Is it going like. I think so many of these things ultimately will work. It's just like, you know, what's the nice thing about text? And I referenced earlier, Peng, Ming and Nikolai are my two co heads of applied research who are just incredible. Like we wouldn't have gotten here without them and the entire team and Nikolai and I have been debating for years about should reasoning be in latent space, should reasoning be in tokens. But one thing that I think him and I really agree on and all three of us is that language is incredible because it's such an incredibly dense way to encode knowledge and information and intelligence. If you think about what went into a physics paper that then is, you know, 20 or 30 pages, like the amount of intelligence and bot and whatnot to then generate that like in that 20 page document, like those little amount of bits, there's so much encoded and other modalities like video and images are amazing, but they don't have the same density of like knowledge or reasoning or whatever. Like the things that we're trying to push for that are encoded in that modality, they're there. In many cases you can watch an incredible lecture for 50 minutes on YouTube but if you treat that as video in data versus text data, the bits to signal to noise ratio, the compute efficiency of the modality is a lot less. And so we kind of have this view. It's like with language you can go really far, but also when you have limited compute, limited people and they're very much linked to two, I think we can push language. It's a better investment. But I want all the modalities. I find it super cool and I love what Deepseek and Others are trying, I can retweet them all the time, but internally we're just like, let's stay focused.
- [77:17]
C
Which I'll say you can see somewhat works. Looking at anthropic, OpenAI has a lot of vision. Multimodality. Anthropic kind of just didn't. Right. Fable's a big step up in image processing, but they're not known as the multimodal company.
- [77:34]
A
Right.
- [77:34]
C
They're the language model coding company that has multimodal capabilities that's never super flex
- [77:41]
A
and goes pretty far.
- [77:42]
B
Look in this, I think, I think anthropic, they've done many things right, but I think this maniacal focus on just pushing capabilities, scaling up models is. I couldn't agree more. I think that's the first hurdle and once we get that, then we can improve a whole bunch of other things. But at the same time, on the other end of the spectrum, it's really exciting to see people building these spatial models and the world models that are being built for very different use cases. But I think ultimately it all comes together at some point.
- [78:19]
C
Okay, so scaling models, this is Laguna S for small. You have good naming. Extra small, small, medium, large still scaling.
- [78:28]
B
So the new medium started training and it's much bigger than the last medium started training yesterday. So it's a 39 day pre training run.
- [78:37]
A
And how do you know the days and events? Just the computer model factor, but.
- [78:42]
B
Right. And like at this point, like with the model factory, like it's.
- [78:46]
C
I thought it was interesting. So in, in the Laguna Medium and extra small, you even quoted number of GPU hours for how many days and whatever for different size. And I'm like, oh, you can also work backwards to how much that costs, right? What GPUs, how many hours.
- [78:59]
B
And you realize it's not a lot.
- [79:01]
C
No, it's not.
- [79:01]
B
It's not a lot of money. And you know, you started with deep seek of the west and I think that's the Deep Seq moment, right? Was a moment when people realized that you can train incredibly capable models for not a lot of money on the training run. But I think that's the falsehood, right? Like the training run is not the expensive part. The training run is a very anticlimactic event. Right? Like we just had a slack message come up yesterday. Say the new model is training and here are the links, you know, so you can, can follow along the evals and like that's it. All the work that goes into that moment. It's kind of like how people Talk. I know nothing about sports but how like athletes talk about like, you know, it's all the preparation, it's all the going to the gym and then the game is just a game. I think that's a little bit like with model training.
- [79:42]
C
Yeah. People had over indexed on deep seq, was trained for $5 million or whatever. Whatever it was. Right. It's like there's the amount of R and D before that the infrastructure is
- [79:51]
B
built up, all the things, the data. But no. So Laguna M is training and yes, there will be an L and there will be an xl and what you'll see. What you'll see with M. Right. M is much larger than the last M. Right. So these monikers are a little bit our version of the.
- [80:04]
A
Yeah, he loves making fun of people for saying small is 24B or something.
- [80:08]
C
No, small for Mistral now is over 100B.
- [80:13]
A
What?
- [80:13]
C
Yeah, I can pull it up.
- [80:14]
B
I mean our small. Right. Is 118. So I don't want to say anything else like it's.
- [80:18]
C
I mean I think it's also. Okay, you're small enough.
- [80:20]
B
We all know that the single hardest thing for any foundation model company is naming. I don't want to say that we're good at it either. I mean it's. This is Laguna, you know, Laguna S 2.1. It's.
- [80:32]
A
It's.
- [80:33]
C
But at least people understand, you know, medium is bigger than small until you mess that up like exactly.
- [80:38]
B
We have a passion, we try hard.
- [80:41]
A
While we're on the topic of naming, this is going to be at the end, but might as well. Why poolside? Why Laguna?
- [80:47]
B
So when we started the company, it was going to be called Snowball Apps. It was after the snowball effect because we expected this company to become a snowball effect. And it definitely has been a snowball effect for us. Turns out it's an Amazon trademark. I kid you not that. My co founder's next suggestion of a name was let's call it Bedrock. And so at this point it was like, okay, no, you are amazing at naming things if you were at Amazon. And so early on in the company, before we were incorporated, we were at an annual conference of a very big major tech company and we had been discussing with them and you have to realize the company at this point is me, my co founder CEO Margarita. We know the first person is going to join us. We haven't incorporated yet and we were discussing an OpenAI Microsoft style deal with this big tech company. They were going to provide us with a lot of Compute. We would give them perpetual access, a whole bunch of things. And we found out the name was trademarked snowballabs. While we were at that conference and having this discussion that we had no right to have, we were a couple of guys who had nothing yet, but this big company was willing to entertain the fact that we might partner with them. And we were discussing this, and it was in their annual conference in a public setting, and the chief scientist of that company said, people can hear us here. We should move somewhere else. Let's go to the restaurant poolside. And for some reason, me and Jason looked at each other in that moment and said, oh. And then later that night, the name stuck with us, the word stuck with us, and we said, let's call the company poolside. And ever since, we never ended up doing that deal, and we used it as a reminder to never turn down our. Round down our ambitions, because that would've been the easy path. And the hard path was what we did, which is start and try to raise exorbitant amounts of money when you're just a couple of guys who are not even building it in Silicon Valley, who don't come from any of the known navs and things like this. And so everyone assumes poolside because AGI, everyone sits poolside. And it was a playful name and we liked it, and it was a little bit different, but actually, the name is like a reminder for us to never round down our ambitions. And whenever you're faced with those decisions, to just pick the harder path.
- [83:09]
A
Yeah, I mean, that's a great story. I know you've told it before, but I just wanted to, on the record. That's what I did the first time I met you. You told me, you sat me down. We were in the hotel somewhere and you were like, we're raising 500 million. I'm like, and then you gave me the whole vision, and then you actually did it. And I was like, you know, I don't have that much opportunities to ask just, how do you do that kind of raise to that kind of. To those kinds of VCs, what are they looking for? You know, like, yes, vaguely AGI, but, like, what do they want?
- [83:41]
B
When you look, the world's definitely changed, right? When we were raising that $500 million round, the majority of investor conversations were still trying to explain that these models were not just stochastic parrots and that they were going to keep going. I've seen the world go from OpenAI is going to win it all. And there's no one Else who can build company, right? I mean, Anthropic struggled to raise their $500 million round. That's kind of like well reported. They pulled it off gladly. And so I think when we raised that, it was about a year and a half ago at this point, the world was very different than it is today. I think the world today there's been this function where the number of people who believe AGI is real is probably a super linear or definitely some form of an exponential function itself. And I think this is important because if you hold the belief that we had three years ago and a year and a half ago, and we will look for people who share that belief, which is like this technology is going to fundamentally underpin everything that's economically interesting or economically valuable and scientifically interesting for the future, then the value function afterwards is easy to understand, which is okay. If you get there, you are one of the commodity, one of the players who can build this commodity. And over the years building that commodity has become not just about building models, but also about building infrastructure and other things. And so I think today, because the number of people is bigger and the outcomes have been proven, I think the incredible financial success that Anthropic is having right now and the growth that OpenAI has had and others and Google no longer make this a question of is their product market fit. Which really a couple of years ago was part of the question how big can these things mean? You tell people that you'd be at these amount of revenue numbers in our industry right now, people would laugh you out the room. Now I think it's a function of who in the world believes that it's going to be an oligopoly of intelligence and who believes that oligopoly can be broken by other companies. And I think that's what divines investors more than anything else for the ones who believe in AGI. And then you've got a whole layer that is kind of self selecting out foundation model companies because they're looking, look, I can't make the money I put there compared to what I can put in an application company is very different. I think there's incredible application companies and there should be many should be built. But I do think we are still in a world right now where this is the early innings of this can still be the early innings of who is going to be part of the set of people who win. Intelligence is the most, in my view, going to be the world's most demanded commodity. It will more commoditize in margin and Price and the world wants choice and wants options. And so I think treating the world as like, oh, there's only going to be two players, I think is very short sighted from investors. I think that group who thought that was a lot bigger at the beginning of the year than now, I think the last couple of months have woken up a lot of people and going holy shit. Like the world both can use a lot more intelligence, but also the world is far more complex. We should have multiple choices, more options, things that can be turned off, that can be that. The restrictions that people put on models now I think is another area of this, right? The fact that we are entering into a world where model companies are saying, you're not allowed to use me for foundation model company development. They should be allowed to do this. It's capitalism, it's their business, it's their work product. But it is insane. It is wild that we are okay with that.
- [87:31]
A
Do you have more problem with anthropic saying it or the White House saying it? You're picking two, the two different limitations and restrictions there.
- [87:39]
B
Look, I think I'll put it this way. I think we want to, as this technology gets more capable for the better and worse, we do want to yield to democracy to figure this out more and more. I think any single company making unilateral decisions is dangerous. It's a concentration of power in a small number of people with very limited checks and balances. And that has never worked out well in history in any way, shape or form. And this is not a criticism on the existing foundation model companies. This is just more commentary on how I'd like the world to be. I think in a world where the technology gets more capable, government needs to play an active role in determining where is there real risks of misuse. And I do think we need to separate safety between misuse and doomsday scenarios that I think are. No one knows if are going to happen or not. And I think just very practically, I think I'm glad to see there's a lot of conversation now starting to happen again at the government level of trying to figure this out and now what the final decisions are. Maybe I'm happy about them, maybe I don't, maybe I agree, maybe not. But ultimately that's kind of democracy always, right? Like at any given moment I might not be perfectly happy with one or the other, but people chose to vote in someone to make those decisions. And so I think over the long run, over, you know, over a 20 year time span, the world kind of directionally Goes correct. And democracy does work at least. What's the famous quote of like it's the, you know, the worst of all. It's the best of all the worst systems or something.
- [89:26]
A
Like it's the worst form of organization except for all the others that we've tried.
- [89:31]
B
Exactly. That's the one.
- [89:32]
A
You can always call on me for a Churchill quote because I've studied Churchill a lot. I love that.
- [89:37]
B
And so that's where I hope for now. I do think we are in a critical moment of time and so speaking up for anyone is important. I think researchers who are thinking about starting their own foundation model companies start. People who want to share their opinion and be vocal, if that's with their representatives or just out on X, do so. And but concretely to your point, I think we are not at a level of capability right now that we should start restricting open models in any way, shape or form. I think it will hurt innovation if we do so.
- [90:15]
A
Is there a point at which you will change your opinion there?
- [90:17]
B
Yes. I mean, look, and there has to be. You cannot, if you sit with a straight face and say this can be open forever in every way, shape or form. It is just as, I think, egregious as saying the opposite of it all needs to be closed down right now. I think at any ends of extremes of spectrums is where we go wrong in society in any way, shape or form. And so the answer is always more nuanced and the answer is never black and white. And so I think as we encounter like real world scenarios where we have to say, hey, we have to be more careful, we need to reevaluate. If that means training a model differently and opening it up, having different versions, some things that are restricted, I think that's totally okay because I don't think anyone should be irresponsible. What I do want to call out is that people have been calling for the fear of misuse of these models since GPT2.2. Right. And I still remember, like we cannot release GPT2 because the whole world will get misunderstand.
- [91:20]
A
That was Dario.
- [91:22]
B
And so like this is not a commentary on Dario, it's a commentary in general, in the space. And so we have not been very good at this so far and we need to get better at it. And I do think that the work that's happening with like safety institutes and better evals and things like that is probably the right direction.
- [91:38]
A
Yeah, I mean, I would honestly do something in defense of this. It's better to err on the side of safety and then roll it back rather than the other way. Because the other way it's a one way decision.
- [91:51]
B
I think that's true.
- [91:54]
C
The caveat there is also the competition. Right. You don't have global error on the side of safety. Right. You're talking.
- [92:02]
A
Yeah, exactly. So you don't get to do unilateral safety because someone else will just be more unsafe than you. Yeah, exactly.
- [92:08]
C
You can pause innovation here. It doesn't mean it's pausing.
- [92:13]
B
They're complex part. Yes.
- [92:15]
A
Right.
- [92:15]
B
And I think we are much better off talking about certain capabilities that we can commonly agree on and internationally agree on that we want to limit or not have available. Then we should talk about it in black and white of models available. Yes or no. The moment you start getting these big blanket statements, that's when you start getting at the risk of. I always think back about when we banned advertising on cigarettes. Good thing. I'm not saying I'm against back, but it effectively established an oligopoly of cigarette companies because no one else could ever compete. And it was probably the best moment to the tobacco industry that that ever happened. And we don't want to do that right now. If we pull up walls behind innovation. And this is a self serving comment because I'm not at the frontier yet, but it's not just related to me. I think it's related to everyone in the space. You are deciding right now in 2026, based on the current capabilities of models that this is something that only two or three companies can build. And that to me reads like Chapter 14 of the Most dystopian sci fi novel that I could read. Because from there I think you can play out all the scenarios that happen in the world and none of those are the ones that make me excited about the future. And I think that's the thing we should all think about. What's the future we want to be excited about? What do we want to have? And I think that's a future where intelligence is a commodity. Everyone can access it. It becomes cheaper and cheaper. Right. And I think that's important. It can impact more of the world. And it's not one where a single company puts their thumb on their skill of both what it outputs to or turns it on or off.
- [93:56]
A
I think the one entity that has more power than the US government here is Nvidia. Yeah. Because basically whoever gets the allocations gets the compute.
- [94:06]
C
You can take it down to TSMC
- [94:09]
A
or TSMC below that. But I just want to test provocative statements to see if you have any
- [94:17]
B
response, I need to think on that
- [94:19]
C
one which actually I think they are regulated. Right. Like you can see the government can they ship to China.
- [94:26]
A
Okay, but, but you know, they're not China.
- [94:30]
B
Look, I think this industry has existed because of what Nvidia has done. Yeah, I know that people like, it's easy to give them flack. But I also want to say like I remember when we started sourced right in 2015. Post that Karpathi article. It was able for this progress to happen because we were able to put consumer GPUs in servers and they allowed us to do so. And then like and you kept going forward further and so this is something like foundation models are so closely linked to their hardware and their systems. Why do we see these stepwise progress happening? We see them happening because of the next generation of networking and systems that come out.
- [95:06]
A
Right.
- [95:07]
B
The difference of a model you could train on hoppers versus GB3 hundreds is the difference between a trillion parameter model and a 5 or 6 trillion parameter model. And so these things really coexist I think very closely to each other. And I think the more interesting question I think for the future is going to become of what can we unlock in terms of model capabilities as we start co designing these things even more. And we're seeing that with the next generation of systems. And I think the world abhors capitalism, does a really good job at trying to like, you know, push towards things that are, that allow for more competition. Right. And Nvidia allows for competition. It's not, you know, but if a government says no one else can build foundation models effectively through the regulation, that is very different. Is it hard to go build an Nvidia? Absolutely. Is it hard to build a foundation model? I think it's very hard to build a foundation model but we should like make the playing field one that where you know, if someone wakes up tomorrow and wants to do so, they're like allowed to do so and they're allowed to use the tools to do so. And I think there's still a big difference between what we're seeing in the discussions around model companies versus what we're seeing with chip companies.
- [96:26]
C
The gap also seems to be the expertise in who regulates it. Right. Who at the government decides what's too safe, too smart, too dangerous. But while we're throwing spicy questions out there, do you have anything that comes to top of mind that could be changed? So you know, should OpenAI anthropic open source models? Is it open weights? Is it what we do In RL that determines, you know, your safety barriers. Is there anything that should be done there?
- [96:52]
A
Just. Yes.
- [96:57]
B
One of the things that I'm excited about, that I think we're more and more talking about, I don't think anyone is doing yet, is mix and match of hardware during RL training.
- [97:09]
A
Right.
- [97:09]
B
Like the, you think about like the notion and we're seeing this in inference, right? The pre, fill and decode just work better with, you know, a general purpose GPU and a more specialized like chip. Right. Like the GROK chip at Nvidia, the LPU and the GPU combined and there's different versions of that in the industry. And RL is batch size constraint.
- [97:32]
A
Right.
- [97:32]
B
So like you are ultimately in your batch size constraint because you don't have infinite tasks when you've got the entire web. You can be much more flexible in scaling up your batch size because you've got the entire web. But for rl, you have, you know, x millions of tasks that you are going to be training on. And so you cannot blow up your batch size massively, which means that you actually can't scale compute to a certain extent with RL the same way you could skill compute with like, like pre training. And so I'm very excited about anything that improves that. And I think one of the best ways to start improving that is the things that we're already starting to see in inference, which is the separation of the pre fill and decode to different chips to come to reinforcement learning. Right. And I think we'll be there soon and I think more people should be working on this because then all of a sudden we're able to just be way more efficient in how we train RL from a wall clock time. Again, coming back down to the fact that it's a race, right? The race is measured not in how many GPUs, but the race is measured on calendar time. And that's probably one of the biggest impacts we can have right now to speed up our industry. And so that's one like, technically, I love geeking out about and talking to people.
- [98:46]
A
Yeah, I would talk to Etched. I had a tour of their data center. And physically you can see how PD disaggregation is mapped out in the, in the data center and you have to own your own hardware to do that.
- [98:58]
B
Yeah, no, look, I think it's, I think more innovation in the space is just like, is the coolest thing. And so I'm excited because that's frankly, like all of us are like, why don't we Finish, you know, post training, this model, whatever, two weeks before release or not, sorry. Between release, between pre training, then, you know, mid training, SFT and then the time it takes for release. My biggest wall clock bottleneck right now is RL time.
- [99:25]
A
Right.
- [99:26]
B
And it's just because I can't scale it up further because I can't add more GPUs to it because of that badge size constraint. There's a really cool blog post that just came out that was showing RL done in even lower precision than any of us are doing. I thought this was really cool. So this. What date is it today? We're on July 15th. So this came out five days ago and I thought this was very cool. I think, you know, lower precision RL while keeping it stable. We're still doing this in FP8 and so I was excited to see them sharing this work and bringing it out. It's definitely something that I'm excited to be doing once we move to Blackwell GPUs.
- [100:05]
C
But yeah, cool. Part of open research, you know, you take and you give.
- [100:09]
B
Exactly. Yeah.
- [100:10]
A
I'll just quickly mention there was a paper that did a Bayesian on levels of quantization and they roughly concluded that four bit was the sweet spot.
- [100:19]
B
But I don't remember. This was a couple of years ago. Right. I think I remember this one year
- [100:24]
A
came out but I'm like, okay, maybe NV FP4 is it. You can't really. The lowest you can go is ternary. That's it. There's not that many.
- [100:33]
B
I mean there's still quite a difference between NVFP 4 and 4 bit. Right. In terms of what, what's. What's possible. But I think NVFP4 is, you know, underrated in terms of, of what it is. I'm. I'm quite excited that it. When it came out, it's, you know, just getting that extra, like that trade off between. Between range. Yeah. Is very cool.
- [100:53]
A
Couple closing questions.
- [100:54]
C
I have a quick one. Okay, quick question back to technical side. So any big takeaways from XS 2.1 Medium to training the new small? Just, just general in terms of training models, you mentioned a lot in the earlier discussion about. Okay, in pre training, there's a lot you can squeeze out, right. You can learn a lot more from the web. At the same time, you took 30B and scaled it up to 120B. Right. Is there any gating on how small is too small? So I'm just going to ramble for a bit. I'll come to A question at the end, but part of Karpathy's thesis was cognitive core. Right. We've seen vibethinker nanbege3b4bs that reason a lot. And then, you know, the idea is you offload to a different model for the work. These are small reasoning models. So have you found anything interesting in model sizes like 20, 30 Bs on device, 100 BS on single GPU? Can you squeeze out more there?
- [101:56]
B
There's a lot more to squeeze out. Like I think not to make too many forward promises, but I think we can squeeze a lot more out of the excess size as well. And I think we learned a lot during S training that will allow us to improve excess like size even further. And I think already since then we have learned things that could have made S even better. I think there is a lot more still for like our space to squeeze out of models much smaller. I don't think that's an argument against scaling, it's just an and one. By the way, I think this is a nice thing that, you know, it's really, it's not very helpful to have a post training recipe for a smaller model and try to apply it to a bigger model. Yeah, it just in all cases you're gonna have to rethink most of the recipe. But recipe for post training for a bigger model applied to a smaller model is almost always just a really good like improvement and baseline. You can still tweak it more, but I don't think that's necessarily like obvious. And so once you make your bigger models better, you often have a quick lever to quickly improve your smaller models again. But will we be able to squeeze a lot more out of smaller models? Laguna S gave me a lot of confidence that I think we can and I think it's around that that discussion we had earlier about that it's about the behaviors, not necessarily the raw intelligence that you're trying to improve the models for.
- [103:24]
C
And that's on all axes of there's like an axis of how long a model will reason, so how long can it stay gentic? And there's also efficiency.
- [103:32]
A
Right.
- [103:32]
C
You want to ideally push on both. And the thing to clarify you guys aren't doing right now, which we do see at Frontier Labs, is the distillation. Right. You have a big, big model that you don't really ship to users. And what you put out for inferences typically distilled from that, which gets you quite a bit of gains.
- [103:50]
A
Right.
- [103:50]
B
Look, I think it's something we don't do Right now because of why we're also building these models. Right. These models are for us part of our research path. So Laguna medium was much larger than the last two models, this one and last one that we've released. And we've trained even bigger models in the past. So there is the engineering component of like a bigger model and every kind of order of magnitude size. You'll learn new things in pre training about stability, but at smaller model sizes you are able to just iterate a lot quicker like internally right on your research. And so for us, distilling down to a smaller model doesn't actually serve the purpose. These models are kind of, it's not the right term but to us they're dual purpose models. They are progress for us to weigh, to see did we improve in the model factory and something to put out into the world, world. And so that's why we don't do it. We've done distillation experiments and there's like really cool things you can do. And I think if you have lots of user data then you can go even further in that. But I think there's something to be said in having a quick cadence of models trained end to end from scratch so that you as a research organization can learn the lessons and not wait. That was actually one of the big lessons we learned over the years when we used to have a much longer cadence between model trainings, like six months and we would train just like a big model, wait six months, train another bigger model. You would be compounding so many changes of improvements that by the time you're training your next model, it's a bit of a soup and you don't really know what ingredients led to the outcomes. So when you are training far more frequently models, and this holds true for both post training and from pre training from scratch, you are much more able to get an understanding of what led to the improvements. And I think that's important. Like ultimately we are all still. There is no true science yet of, you know, deep learning for large language models. But we are all, I think, you know, trying to gain insights from our experiments because it's those insights that lead to scaling laws that lead to the kind of improvements that allow us to be again more compute efficient and get more capabilities.
- [106:02]
A
Yeah, amazing. I was going to end off with a little bit more history. You spent some time looking at metrics for engineering team productivity. How do you think about engineering team productivity today?
- [106:15]
B
I mean it's wild, right? I mean it's like the golden age. Like it's the fact that you can just take an idea and build something by waiting overnight for an agent to do the work. I don't know to me.
- [106:28]
A
But how do you measure? Because literally in a theory you are doing this.
- [106:32]
B
Look, I think it's a good question. It's one I haven't thought about in a long time.
- [106:36]
A
But you're pretty qualified to do it. No, it's a fair point.
- [106:40]
B
Let me take a second to think about it. Look, ultimately, what is code, what is software, what is engineering is to go from something that is valuable for an end user or sets of end users, like an idea, an extra bug fix, a feature, to like delivering that value. And I think what we're doing with these models becoming more capable is that we are massively like both cutting out middlemen and compressing the time that it takes to deliver that value. And ultimately that iteration cycle for any startup or any company is what allows you to win. Right? If you're able to solve a bug in two hours versus staying in the backwalk for three weeks, if you're able to like be on a customer call and learn, hey, if this feature existed, it would like, you know, they'll be willing to pay more and it's more valuable to them and you ship it in a week instead of in a month. And so I think ultimately maybe the same things that we looked at years ago, prelims still apply and it's just a notion of cycle time, but in this case it's lead time from the moment you have a valuable thing that you are looking to do for someone to the moment that it's actually shipped to them. Every other metric is ultimately a leading indicator for that lagging indicator, right? It doesn't matter if you're looking at amounts of code, PR reviews, all of these kind of things. And so I think in this case we are starting to move so quickly in some of these things that we can just sit back and look at what was traditionally the lagging indicator, which is the lead time from traditionally ticket to like, you know, like an end result. What I would look at in this new world that maybe we didn't think about before is how much can a single person do with that? One of the most, like if you look at AI native companies, they're not designed like the engineering orgs of, you know, pre LLM age. They're actually designed with often just the builder, right. And as close to the kind of customer to the ability to ship. There isn't necessarily a huge team in between that sits there And I think that is, I think, is exciting, like organizations where a single IC can just get much closer to that. So I would look at from where the value sits that's identified to the moment it's shipped and how many people are involved in that. And you want the amount of people involved in that to be less and you want the time end to end to be shorter.
- [109:06]
A
Okay, is there a way to eval that when you're interviewing somebody? Because that is, you know, the most compressed version.
- [109:17]
B
I think the common answer to this is agency. How much agency does a person have? I think in the age of AI, getting more capable, agency becomes probably one of the most important qualities for anyone. And I think agency is something you can look for in, you know, what people have done in the past. Because agency is something that if you have it, you are demonstrating it, right? No one has just agency and is sitting back and not, like, exercising it. The whole definition of it is that it's exercised. And so understanding, like, what were things that people did in their lives and their professional and their personal project showed agency. And your personal backstory shows a ridiculous amount of agency. Right. Like, I think that is ultimately it. It's the, you know, the Silicon Valley, you know, quota of the last, you know, year and a half or so. You can just do things, right? That's, I think, what you're looking for.
- [110:09]
A
I think then aligning high agency people is very hard because they all want to go their own way. That's the whole point, right?
- [110:16]
B
Yeah, but I think the, the notion of, like, I think the notion of a good leader, right, in an organization is to be able to bring people together around, like, a common outcome. And I think what you want to do, anyone who's high agency, I feel very lucky. I've got an organization with incredibly high agency people. Like, I mean, I'm not the one who built the model, right? I cannot stress this enough. Like, it's the team that, like, achieved this, and it's a team that is incredibly high agency. And so if you look at, like, what does it take to bring that together? It's ultimately a common goal and a common set of boundaries. Because if you allow to just go, you can do everything, you become an exploration algorithm. And this is what we see in big tech, right? In research and big tech, everything is an exploration algorithm. Everyone can do anything as long as you, and then becomes political about gathering the resources. So when you say, hey, this is our common goal and these are the boundaries that we've set, right? We're not multimodal. We focus on rl, we do these things and you're upfront with people before they join the company. You get a lot of agency, you can run where you want, but these are the places where we say this is the lanes that make sense. I think it actually gets the best out of people because innovation comes from constraints. We did this with relatively little compute and relatively little money compared to some of the others that are out there. And I've thought back on that quite a bit recently and thought actually it was a good thing because those constraints kind of forced us to become much better in certain other axis that might, others might have not. Right. We, we purchase relatively little external data.
- [112:01]
A
I was going to ask about that.
- [112:02]
B
Exactly right. That was a constraint, but it's a constraint that pushed us to move on other areas to improve and like and there's lots of versions of that. So I think high agency people you want to empower, you want to get them really excited what they're doing, but you also want to say, hey, if you join this mission, this is the outcome I need you to achieve. But these are the places that we don't go. And maybe if you care about those places, go somewhere else.
- [112:26]
A
Yeah, agree us call to action. Who are you hiring?
- [112:31]
B
We are hiring on every possible role in applied research and engineering in the company. So from pre training all the way to evals to post training architecture, we are still in a world where individuals can have massive impact. And I think our pitch to join us, we spoke a lot about the mission, how we think about things but I think we are one of the places where it's the highest ratio to individual to impact. Less than 70 people built this model, less than 115 to engineering and researchers together did this effort. And that's a very broad definition because I put myself in the 115 list. And so being able to do this kind of work on a mission that you're aligned with and you can have that in every individual still has huge impact.
- [113:19]
A
And being able to publish, being able to open source the model.
- [113:21]
B
Yeah, look, all of those things are part of that. But I think ultimately you can today pick between joining a very large foundation model company. But you are one of many and not by any fault of them, but just by definition the denominator has become really big and our denominator is quite small. And so the level of impact you get to have is really high. And I think ultimately all of us, frankly the most incredible high agency people I know, what are they optimizing for? They're optimizing for impact. They're optimizing for impact. And am I aligned with the mission? And if today you heard about the mission and aligned and you're optimizing for impact, I think we're a really good place to try.
- [114:05]
A
Okay. I think we ended there. That's most fantastic statement. You did amazing on four hours of sleep. Thank you, guys. So, you know, Podcast Evil definitely purchased it.
- [114:15]
B
I literally let my eyes are, like, starting to go like this.
- [114:17]
A
I'm like, oh, let you go. Let you go.
- [114:19]
B
It's good to see you.
- [114:20]
A
Thank you for setting this up. We wanted to get this in because we think it's a great model, and I think it's a great story to tell.
- [114:25]
B
Thank you.