00:31:32 · Bernt Øivind Børnich
[00:31:32] S1: How do you, like, do that? I think it's gonna be essentially exactly the same as you see in large models today. So there will be a few dominant models just because they have the best data and evolves. There's, like, this not so well kept secret in AI that there are no secrets in AI. It's all just data and evals. Like, why are the best models the best models? It's because these companies have the best data. They were early, had huge user bases. They got enormous amount and still get, like, an enormous amount of data. And since they have so much user data, they are able to train better models. And when you have better models, you get better user data because you get more users, it's kinda like the self enforcing flywheel. There are other models where you can kinda, like, specialize. You can download an open source model. You could try to, like, do some fine tuning on it to, like, get it better on your specific task. But it has to be a very hard, very niche task before this makes more sense than just using their general model. And I don't think this is gonna be different. We we see the same trends, the same scaling laws, the same patterns. It's just robotics has until now been, like, kinda like a toy problem. Everyone's been working on extremely small sets of robotics data where it's like, oh, I gathered four thousand hours of, like, these specific behaviors with my robot, and, like, this is a very big dataset. No. It's not. Like, four thousand hours is nothing. Even, like, few hundred thousand hours is nothing. And if you see the models that people train now of, like, equipping people with sensors, right, and then, like, having that happen at scale, and then they're saying, like, oh, we train this foundation model at two hundred thousand hours of data, which is huge compared to what it used to be in robotics. It's actually still tiny. Like, general intelligence emerges at, like, hundreds of millions of hours of data. So you're not gonna go gather that data. That's why I deeply believe that you need to use these machines as kinda like the bridge between human data and machines. And I do hope that through doing this, the models that we train can operate a lot of robots, not just NEO. And you can generally go in that direction. Right?
00:33:33 · Bernt Øivind Børnich
You can take a large, complicated, like, very capable model that's trained on this enormous diversity of data, and you can distill it down on a smaller problem. And then that model will be very good at the smaller problem. You can't go the all the way around. You can't take this model from, like, this smaller problem that only solve that data and distill it up to solve a general problem. That doesn't work. It's a one way thing. So you kinda have to have the most general system that has the most diverse data and, like, has experienced the most things. And then you could take that intelligence and distill it down to, like, more specific applications. So the question is when that happens. Right? Because we're not there yet. And in the meantime, there will be very good applications for a lot more kinda, like, specialized models, and that's what we're seeing today. Right? If you wanna fold laundry, then actually, if you go and just gather a lot of data on folding laundry with our robot and train the VLA, that will work better than our role model, but it won't be good at doing anything else. It it doesn't have a general understanding of how the world works or, like, it's essentially intelligent. It won't it won't exhibit kinda, like, social behavior and, like, understanding how to interact with people. It'll just fold the laundry. And that's kinda like the regime we are in today, but it's quite quickly evolving. So I think 2027 will look very different. Maybe not 2026. I think 2026, the specialized models will still be better. But I think 2027, you're gonna see a shift towards more like general intelligence.
00:34:59 · Ty Morse, Bernt Øivind Børnich
[00:34:59] S0: So initially, you kind of have to use, like, teleoperation in some form or another to, like, I don't know, bridge edge cases and stuff like that? [00:35:08] S1: You need all the data. That's the simple answer. So, like, you you you you use the web data, which is gonna be 99 of the data. You use simulation, synthetic data. You use sensors that you equip people with to get some of that data. You use ecocentric video data. You're gonna use robot data with teleoperation, and you're gonna use robot data where the robot is just learning by doing. You use all of it. I think the specific data of, like, oh, I have an intervention in in the sense of, like, the robot failed and, like, now I correct the robot, not that important at scale. It's extremely important at small scale. Like, if you wanna, like, solve a specific problem with a very small amount of data, this data is essential. And the technical term here would be that your distribution is very small. So, like, you know how to be within this kind of realm here, then the robot knows what to do. If it gets outside, you need something that tells it how to get back on distribution so it knows what to do. But if you have a general model that is very capable and diverse, then, actually, it's very hard to get out of distribution, and then you don't need this. You can always kinda, like, find your way back to where you want because you're still in distribution. It has to do with the robustness of the model. First, you try
00:36:24 · Ty Morse, Bernt Øivind Børnich
[00:36:24] S0: on the Internet data, then you get a little bit of your your own data. But eventually, in order to really get the data that you need to train a super genius robot, you just have to have a whole bunch of these Mhmm. In operation. Like, if you think about if you have 50,000 next year, suddenly, and they're operating, let's say, sixteen hours a day, you're getting 16 times 50,000 every single day of new data, I assume. When you're thinking about, like, your bottlenecks for scaling your data, is, like, one of the biggest bottlenecks just we need more robots in the field? [00:36:55] S1: Yeah. And they need to be in very diverse environments and very diverse applications. In reality, you're almost never data bound. You're diversity bound. So when people say, like, oh, you need very large amounts of data. That's true, but you don't actually need just large amounts of data. You need an extreme diversity of different experiences and tasks and things that you haven't seen before. And if you have enough of that, that becomes a lot of data because you have so many different things. But lots of data of the same thing doesn't really help you. Right? [00:37:29] S0: Let's say you ship 50,000 robots next year. What is the breakdown of where those different robots are going to go? Because, like, maybe, you know, initially, I was thinking, like, 50,000 to homes. But maybe you don't want that. Maybe you want, like, 10,000 to homes and, you know, 2,000 that'll just go into the woods and build a cabin or something. I don't know. I think you just
00:37:48 · Bernt Øivind Børnich
[00:37:48] S1: just cross the stack. Like a couple years more out. But I... One of my dreams is to be able to leave my house and say, like, hey, NEO. We're having a part... Party this weekend, and I would like my garden to be Japanese. And then just, like... It'll, like, go and reconfigure the entire garden and, like, build a nice pavilion and, like, you come back, like, a week later and it's like, oh, it's ready. That's nice. But that's gonna be a couple of years out. So so so this year, there will be a couple of enterprise applications where we have very large customers so we can deploy a lot of robots into specific verticals. And then there's gonna be some to the home, and it's gonna be a lot to the platform. It's gonna be a good good mix of all of them. I think it's important to get, like, information from all of them and, like, get the lessons learned from all of them. But I also think it's important to have some applications where solving the task to actually be truly useful and, like, a return on investment for your customer point of view is very important because that's how you drive adoption. Right? And you need scale to get the cost down and to get the reliability up and everything else. So, a good good mix of these. We're not quite ready to announce yet what the specific enterprise applications and customers are, but that's in the works.
00:39:07 · Ty Morse, Bernt Øivind Børnich
[00:39:07] S0: Do you kind of think about it also in the same way that maybe you need a bunch of, like, folding data? And so one way to get that in an enterprise setting might be like, hey, old Navy. You know, here's 5,000 robots, and they just spend all day folding clothes, and then you get, you know, 5,000 times sixteen hours a day of [00:39:25] S1: folding. Application. We're not gonna do folding as our main application, so I can I can debunk that myth? Everyone else is doing folding, so that there's already a lot of folding data. [00:39:34] S0: Okay. So everyone else is doing folding. What is the one thing that no one is doing that seems fucking obvious to you? Is that... Could you say? [00:39:40] S1: No. Fuck. Someday, I'll reach out to you again. Someday being just a few months, and tell you, like, here's what I thought about. [00:39:46] S0: Okay. Next year, we can [00:39:47] S1: we can know. Okay. We we think we've figured this out. We we we think we have a couple of things we're gonna do that. It's gonna be, very material. [00:39:56] S0: I remember this one situation where one, like, oil company bought a fertilizer company, and then suddenly, like, all these other oil companies started buying fertilizer companies. Like, the first guy was just like, I just bought it, you know, for shits and gigs, and then other... A bunch of other people started doing it for no reason. And is it like full... Some guys started unfolding, and then everyone else is like, well, they're working on folding. Maybe we should too.