From Pension Funds to Robotics: The NRN Journey

Boomtown, and we are back. Welcome, ladies and germs, boys and girls, to the Intercognitive Foundation podcast. My name is Rich Robinson. I am the founding chair of the Intercognitive Foundation with our mission and vision to make the physical world accessible to AI. And I am joined this fine day by the 2 co-founders of Psi. Please give a warm Intercognitive welcome to Brandon and Wei. Welcome.

Rich, how's it going, man? Great to, great to be here. Yeah, I'm really excited.

Excellent. I know you guys are dialing in from my wife's hometown, T-Dot. The home of Drake, Toronto, T-R-O-N-N-O. And I am here.

You know, you know, people don't call it TDOT anymore.

That's—

I know, I know, I know. That's so old school. So old school.

That is old school.

It's so old school. It's not actually new school. It just makes me old, but that's okay.

No, no, it's, it's not, it's not that old, but you know, the, the younger generation probably just, I don't even know what they call it these days, but I feel like TDOT—

Kids these days. Yeah.

Yeah. TDOT is nostalgic.

I met somebody at an event last night. There's a guy named Mike Chang. He's into this thing called Flow 60. He lives here in Bali and he's kind of a movement fitness guru. It was a really great session. And I met somebody and I said, where are you from? And they said, guess. And I said, say something. And they said, well, if I say something, you might be able to figure it out. And I was like, Alabama? Um, and I just said Toronto and they went, yes. So yes, I have an ear for that. So welcome, welcome. I, uh, I would love to hear your individual hero's journeys and how they converge upon what you're building with Psy. Uh, please, maybe we go alphabetically and, uh, Brandon, please, um, please tell us about your backgrounds, your superhero abilities and the utility belt that you've built up over the years?

Sure. I mean, it started as me being the Robin to my Batman over there, Wei. So it started back in the pension fund world, or just, I guess, finance world in general. Wei was working at a fund and I wanted to work at this fund and I was kind of in there As like an intern. And then we eventually started talking and then he brought me under his wing and then we kind of just started building from there. We worked on a lot of things, but one of the things primarily that we worked on together was building up quantitative trading strategies using machine learning. So we started that, goodness, it was like 10 years ago. So in 2016. And we worked at the fund for about 5 years, had a lot of fun, and we ultimately left to create this company. Now we've done a few things in this company that range from gaming to robotics, all of them with a common theme is that we wanted to build like competition platform or evaluation platform for models in environments, right? In complex and dynamic environments. And so we started off with a game, which was AI Arena, and then we broadened that out to a platform to host competitions, specifically machine learning competitions on multiple games. And then we expanded that, the types of environments, to robotics. And now we're leaning so heavy into robotics that I forgot what a game is. So that's my origin story. But probably should have started with Wei. His begins a bit before mine and I happened to stumble into him, which is the greatest thing to ever happen to me. But yeah, go for it. Wow.

That's beautiful. That's beautiful. And I think the ability for founders to be able to have complementary skills, but also some sort of respect and admiration for each other, I think that's under Underappreciated, where it's like, I think you're a badass. I think you're a badass. And like, yeah, let's do that together because it's almost like a marriage in a way that you really, really have to have this sort of like, my wife loves me. She doesn't like me, but she really loves me, right? So you have to like, you have to kind of love each other in some ways. And if you don't like each other, sometimes that's okay because the love cures all. So I think that's excellent. I really love that you're— You know, just preaching about that. That's beautiful.

Yeah. I think our wives are jealous of mine and Wei's relationship. I think they've said that to each other on multiple occasions. Amazing.

Amazing.

Yeah.

That's actually something that you should actively seek out.

Yeah. Yeah. I look, I think my version of the story is, yeah, I was at the fund earlier than Brandon. Obviously, when you spot talent, you see that differentiation stand out. So, you know, it wasn't a hard decision for me to try to— I actually basically poached him from another team and just said, no, you're coming and working with me. And it was great. It was like a great kind of formative experience at the fund. We started building really interesting things that ultimately led to Kind of building some of the technological foundations for what we have deployed within our company called ArenaX, starting with AI Arena on the gaming side and now with Sai on the robotics side. So it's been a long journey. And I think what you just hit on, Rich, like, I think we all know building a business is incredibly difficult. And a lot of times businesses fail not for technical reasons or even business reasons is personality issues. And that's, you know, I think conflict. And look, I think with any relationship, there's, you know, ups and downs and it's not possible to always see eye to eye on every single detail. But I think Brandon and I would characterize it as like a high trust relationship. So we don't ever question each other's intentions about things. Like we may disagree, But you have that underlying trust in the other person that, you know, we're core aligned on some fundamental things and it's just about kind of working through issues that inevitably will arise every single day. So I think from that standpoint, it's been an incredible partnership. And it was one of those things, like, as you mentioned, is It's very rare. It's actually not common. So I think from that standpoint, I'm certainly incredibly grateful to have the opportunity to work with him alongside the rest of our team every single day.

Yeah, it reminds me of a comedian saying like, you know, I'm not married yet and he's, it really frightens me because, you know, 50% of marriages never end. So it really, and I think, let's say half the marriages end in divorce, but the other half are actually maybe not that great, right? So you have like 25% chance of really nailing it. And I think in founders too, the default setting is there's some sort of underlying conflict or there's grumbling because I don't really trust your intentions. It's super important because then you want to basically put all your time and energy on actually building the thing, making stuff people want instead of trying to negotiate your partnership every day. Hey, can we just go back a little bit to the origin? Because I don't really know much about pension funds, but I somehow hear Especially in Toronto, the pension fund being very well known and it's actually a pretty badass petri dish in which to have been spawned from. And I think you were probably doing AI stuff in an environment that had tons of resources way before other people were doing it. Can you tell us a little bit about that?

Yeah. Um, I think It's really interesting. Canada is a leader in kind of pension management and pension is all about kind of retirement security, right? As a broader thematic, Canada stands out because I think in the early 2000s, Ontario Teachers, which is this like, I don't know how big they are now.

Everybody knows the Ontario Teachers, right?

Yes. Yes. They are truly badass. They're the trailblazers. In the pension industry. And I think what the Canadian model does correctly is a couple of things. Number one, despite the fact that these funds are, you can characterize them as like quasi-public institutions to a certain extent, they are managed and run like private institutions and they're set up to compete against the premier top-tier asset managers. around the world. So what ends up happening is they attract the best talent in Canada. So to the extent that like even in Canada, it's actually like Canada doesn't really have that many private equity or venture capital funds. I mean, there is a growing community, but many talented investors actually just work for pensions. And part of it is because of the governance, the ability to pay for talent and all of that. So what ends up happening is you get a better product, which is why Canadian pensions have this leadership position in the world and a lot of other jurisdictions try to emulate. So all of that is context to say we do have a lot of latitude within Canadian pensions to kind of be on the frontier in terms of experimenting and testing out different approaches that would give the fund an edge and make sure that it's adapting to the ongoing evolution of the markets, which is really something that you need to stay ahead of the curve. And both Ben and I, we are extremely, I think, lucky to also have had a tremendous mentor in our CIO at our fund that gave us a lot of latitude to go off and explore things that were a bit on the leading edge, but ultimately we were able to establish very interesting investment programs and strategies as a function of what Brandon was researching and what we were able to kind of deploy capital into. And that touched on things like blockchain, AI, across different types of strategies on the illiquid markets and the liquid markets. And this is all happening back in like 2019, 2020. Which at that moment in time, even in the context of Canadian pensions, it wasn't that common. And I think we're still extremely proud that our programs and even we had this thing called the Innovation Portfolio kind of survived even after we left and it's still kind of growing. And there's a new generation of kind of young talent that's kind of taking taking that and continuing to grow within the context of the pension fund. So it was an incredible experience, but that's a little bit of the story about Canadian pensions is like a cool little— That's beautiful.

Yeah. Forged by fire in a crucible, but then given latitude and opportunity to be entrepreneurial, but then you still left that warm embrace of that because that's kind of a pinnacle. It's like maybe working for Unilever in India or the government in Singapore. Like it attracts like Such great talent. And then you're surrounded by great talent and you're like, ooh, I'm getting paid well. I have a lot of opportunities. I'm, I have a kick-ass group of people that are making me better. But then you guys had the chutzpah to like pull the ripcord and be like, all right, we're going to build this. And like, what was that conversation? What was that, you know, bitten by a radioactive, you know, robot and how, or at least, or at least the original idea.

Well, I mean, from my perspective, Brandon walked into my office one day and said, hey, look, I have an idea. And it was off of the back of us kind of going down the rabbit hole of crypto. We've been researching and thinking about deploying into crypto in an institutional context since like 2017, I would say. And then 2019 came around and it was, you know, the explosion of DeFi and then later on NFTs. So we went down the rabbit hole of whole like smart contracts, decentralized systems, exploring what we can build around that. And when I say we, it was mostly Brandon on the building side. I was just there for the ride. But yeah, Brandon, why don't you take it? Because, you know, he was the one that was working on the early prototypes. Yeah.

So this is like, this was a long time ago at this point. So yeah, some background. So we talked about this like innovation fund. And so we would frequently like present like research we're doing on like things in innovation. And so we started slowly like talking about decentralized technology, like giving presentations on.

Thank you.

So yeah, so we were starting to give kind of presentations on decentralized technology and I remember we did a few, I think one of them we did on like IPFS and explained exactly how it worked. And that kind of like blew people's minds. And then we would kind of talk about these different concepts. And then as I was diving into the smart contracts for NFTs, this is like, holy shit, this is like 6 years ago. I was like, wait a second, these things are just like, generalized containers for anything that's unique and has value, right? And at that point, I don't think anyone had done this idea I had. Now everyone does it, but this was 6 years ago. I was like, what if you tokenize AI, like an AI model? And theoretically, I would think like, okay, well, the better the model, the more technically it should be worth. And this gives the independent researcher a way to monetize, whether it's through directly selling their model or some royalty stream or something like that. So that was like early days of me thinking about that. And at the same time, I was very much into these machine learning competitions. This one specifically in crypto is called Numerai, which is basically just like predicting like asset price movements. And you get rewarded in this digital currency. I was like, that's cool. And then also at the time I was competing in something like Kaggle, but there wasn't really a, I guess, like fun type of competition with dynamic environments, like simulations with like games and robotics, for example. And so I wanted to build something for that where it was a combination of these ideas where someone can own the model that they're submitting to compete. that they can eventually sell to someone else, but this thing is making them money essentially. And my thought process again was for this independent researcher that maybe doesn't have the credentialization of someone that just has got a PhD from Stanford, but they're really fucking capable. Sorry for my language, but they're really capable. They're really capable on machine learning. So basically wanted to make it for them. And that was the idea. And the broader vision was to have this more generalized evaluation of all these models and all these environments. And that's kind of where we're at now. Again, leaning really heavily into the robotics side.

Wow. Beautiful thesis. And I think oftentimes you kind of have this industry insider secret because you've been tinkering for a decade in different areas and you see the Possibilities, like generalized container. I think that's a beautiful way to look at it. And specifically to apply that to AI, that, you know, now people say tokens in AI has a different meaning, but you were basically tokenizing tokens in a way and giving people the ability to own that IP. And it's pretty Pretty beautiful vision that's really come to fruition in many ways. Although maybe with China playing some sort of gangster game and just open sourcing everything, that's a whole nother thing. But anyway, I think moving to where you are now, and I'd love to hear, you know, a way of like how that landed with you and then the evolution of the company over the last few years and what you've landed upon?

Yeah, I mean, it landed in a way that was convincing, so we went off and raised money off of the initial concept. And yeah, we've been at it since 2021. Officially. So it's been 5 years and, you know, 5 years is like basically 2 decades in crypto land. So we feel like we are kind of OGs and look back on everything that's happened since then. And it's been a crazy ride.

You guys are just covered in scabs and calluses and scar tissue like armor. You guys are like impenetrable now.

Generally, the reason why we always wear black is just to make sure that it's always covered up. But yeah, I think the initial kind of foray into this concept was through an abstraction medium that was gaming. So using game as like the cosmetic shell to obfuscate the complexities of machine learning and deliver it in a way where it still conveys the idea.

Yeah.

But in a way, in a format that's very fun and engaging. And so that's what AI Arena is. And so, you know, for the first like 2 and a half, 3 years, we were on this journey of commercializing the game. We had pretty good, I would say like success coming out of the gate. Then it became rather difficult from a macro standpoint. standpoint, um, you know, we hit— I think we hit every single rough patch in crypto over the last 5 years. Midway through, I think in late 2024, we were doing kind of some, uh, you know, strategy assessment in terms of where we, where we needed to be as, as a business. And I think part of, part of what we were building on the infrastructure side, uh, dovetails nicely into utility within the robotics space. So around that time, the engineering team started to also basically leverage the same underlying infrastructure and start to kind of incubate ideas around how we can use this to service robotics, which is where we are currently with Sai. And I'll let Brandon get into the technical side of what it is that we're building, but the short answer is, Effectively, our infrastructure allows us to basically, well, in the current iteration, is really to kind of evaluate the underlying performance of different types of models that ultimately would power robots across a variety of different kind of hardware mediums. Obviously all of these different types of environments, because the ultimate goal for robotics is generalization. Right? Having generalizable robots that can basically adapt to any situation. And if we get to that end goal, then that will very likely be one of the most incredible innovations in mankind's history, just given the potential labor impact and how transformative that's going to be. So I think in that context, I think what we're trying to achieve is Critical because it allows us to understand where the deficiencies are. And currently there are many, like generalizable robotics is not a solved problem, despite the fact that what you see in social media and in the public sphere is a lot of, I would say, cherry-picked, exaggerated, very kind of manicured settings to demonstrate this vision, right, of what it would ultimately become. But we're not there yet, and we're probably not as close as people want you to think. So what we need is this evaluation layer, this benchmarking layer to actually help steer us towards the right direction, right? As an industry, because without that as a kind of a measuring stick is, you know, I think all else equal, we'll get to our destination slower. So what we're trying to do is to accelerate that path forward. So that's the journey of the business. It's where we are now. We're 100% focused on robotics and the platform. Uh, that we're building and commercializing is called Psi. Um, and it's really focused on this like, um, model and policy evaluation to accelerate the entire robotics industry.

Beautiful. What a great arc. Just like NVIDIA, gaming is the tip of the spear for many, many companies. It's just such a gigantic industry, hundreds of billions, and there's plenty of Areas in which to experiment and to be able to do that at scale and rapidity and be able to garner a lot of learnings, right? Because a startup is basically a set of experiments and the more you can experiment, the more you can progress and you've become, you know, just exactly the company you need to be at the exact right time. And you're already so much farther down the path. It's actually Pretty beautiful. Most companies, it's 9 to 11 years to some sort of event, but I think in the next 2 years you can see such a gigantic sea change happening that I look forward to seeing you guys on CNBC or Bloomberg and just being like Will Ferrell from Elf, seeing Santa saying, I know him, I know them. I think you're going to be an overnight success. after 7 years of grinding, the positioning and the team behind it. So let's get a little bit more into the weeds. Please don't make all of our collective brains explode, Brandon, but tell us a little bit about the tech and roadmap.

Yeah, absolutely. So yeah, I'll dive into like eval side. I'll try, Not to get too technical, obviously some stuff we want to keep proprietary, but generally speaking, I'll kind of explain how it works. So like we said, right now we're really focused on evals. So what exactly are we doing? What we're doing is we're basically mapping out the robotics pipeline, basically from perception. So when a robot perceives the environment all the way to actuation. So when it actually does something in the environment. There's a bunch of steps in the middle there, which means that there's a lot of things that can go wrong. And often when we see a lot of these robotics policies, they often have the perfect pipeline from perception. There's no latency, there's no quantization. The policy, as the samples are sampled, get executed perfectly in the environment. A lot of this is not the case, right? The real world's really messy. And so what we're on a mission to do is bring transparency to all of these models, whether open source or closed source. Closed source is a bit trickier. These guys, I was talking to a few of them at GTC earlier this week, and I understand where they're coming from. They don't necessarily want to have their model evaluated at the current moment. Potentially, maybe they're not ready, right? They're showing these fancy demos. They don't want that to be tarnished with an edge case that they haven't properly accounted for in the training set. So I understand they have an illusion to maintain. They need to raise a lot of money. So totally fair game. So we're really focused right now on the open source models. We are still talking to some of the frontier labs on the closed source side, but on the open source side, essentially, Like the Groots, the OpenPIs, SmallVLA, OpenVLA, like those types of models. What we're doing is we're taking them in, we're looking at a bunch of the leading benchmarks that people care about in both academia and industry. And then we're adding an additional layer on top of that, which is basically we're perturbing all these aspects of the whole robotics pipeline. And we're doing a massive sensitivity on them to see where these robotics policies break down. So for example, if we start adding blur, like zoom blur to the camera, it distorts some spatial relationships. How does that affect the policy's performance? What if we start perturbing language? We swap out words with some synonyms. Does that make the model not understand what's happening and therefore it does something totally wrong? Right. These are the types of stuff that we're testing. So I'll pause for a second. Does that make sense? Like, that's kind of what we're doing.

I love it. And the last time we spoke, you used the word perturbation, which I thought was an underused word. Perturbing, I think perturbances, however, however you, and it's, the world is full of that. The world is absolutely, it's a feature, not a bug to have perturbances, perturbations everywhere. And I recall getting some mobile games from the West, and those games were really fat, heavy games with tens of megabytes, which was a big deal back in the day when phones weren't so powerful. And they were all tested on their internal networks in the office. And then they come to the 2.5G network in China where I was based. And the whole thing just crumbles. Like it's, they, they couldn't even figure out why, because they're, they're doing it in the most pristine kind of, you know, almost like a, uh, an environment with the, with the, with the, with the, the, the suits and, you know, like a, um, you know, uh, what, what do you, what do you call that? The, the vacuumed, uh, like, you know, dry space for chips or—

Hazmat, hazmat suits, right?

Yeah. The hazmat suits and the, the bunny suits and the, and the whole, you basically curated area and then it gets let loose into the wild and the whole thing crumbles and you have to intentionally try to perturb it. And I think that's where there's so much value within that. And it's not even so much necessarily edge cases. Edge cases are another thing. Like, of course those edge cases exist, but it's just actually more every day. Murphy's Law, that's happening and just really trying to nudge it and make it stronger. It's almost, it's very sort of like Darwinian and kind of like natural evolution in a way. Can you talk about some of the, you know, I love the whole Zoom, like tell us more about all the ways that things can can go wrong. It's pretty, I think it's pretty fascinating.

Yeah, for sure. I have some really cool findings. Some of them we published, some of them we're going to publish one in a few days, but one of them is on language. I thought this was one of the most interesting ones that we found. So we've released, we just recently released this blog. If anyone wants to check it out, it's at blog.competesai.com. It's the language one. And essentially what we did is we started perturbing language. So we started, like I mentioned, doing synonym swaps, adding more words than necessary, compressing the sentence down, doing all this stuff, right? And we found for the specific model we were analyzing, performance wasn't really changing. So we're like, okay, this thing has a really good understanding of language. And then we started injecting Incorrect prompts into there, right? And it still started doing it right. And like, to give you an example of how strange this is, so like, let's say the task was grab this cup, right? At first you're like, grab this cup, and then you change it up. You're like, put your hand on top of this vessel, and you're like, okay, cool, it understands this. But then you're like, grab the chair, and it still grabs the cup. And then we gave it a prompt that says, I am Franca, and then it still grabbed the cup. And so this was an example of total overfitting, right? It's giving the illusion of intelligence. And in statistics, there's something called spurious correlation, right? So this benchmark had 10 different tasks. Each task had a different configuration, a different layout. Right? And so what was happening is with the different layout also had a different prompt. And so what the model was purely just looking at the layout, and that's what it was basing what it was going to do on. It didn't actually care about the prompt. But because you're appending this prompt, people are like, oh, you said go grab the cup, it grabbed the cup, but it really didn't. And we even tested that by completely removing the prompt entirely. And it still worked. We did an analysis. It started off with like 95% success rate. By removing the prompt, it only dropped to 94%. But if we removed vision, it dropped by 82%, all the way down to like 10 or something, or like 12% success rate.

Wow.

Yeah, so that's—

Spurious correlation. Wow, another beautiful, beautiful term. Like there's that old joke, like, Cut the leg off a frog, jump frog, jump, it jumps. Cut the second leg off, it jumps. Cut the third leg off, it jumps. Cut the fourth leg off, I tell it to jump, it doesn't jump. Cut 4 legs off, frog goes deaf.

Right?

Exactly.

That's exactly it. That's exactly it. And so, I mean, this is important to know, right? Like, imagine you see this thing has a 95% success rate and you're like, oh, this is an amazing model. You know, it's not even listening to what you're telling it. So it's important to do these stress tests. Exactly.

Wow.

No matter what I say, it knows how to do it. Speaking, I'm speaking Portuguese in this.

Wow.

That, that, that's not, that's not, I mean, that's not even something that it's really the law of unintended consequences. It's not something I would even think about.

Yeah.

Yeah, exactly.

So yeah. Um, I want to provide a cool little comparison actually, since Wei and I are from finance. This is very similar to what we do when we're creating strategies, when we're backtesting them, right? There's a saying like every backtest makes money, right? Every backtest looks good because people are just running it in simulation under ideal conditions. They don't properly account for a lot of things that happen in the real world, Like transaction costs, right? The fact that the market is non-stationary, things change, right? There's all these things you have to embed and most people don't. When they're testing, they get, oh look, I have like a 3 Sharpe ratio strategy, which means really high return and low risk, right? Similar, this exact same thing's happening in robotics now. People are backtesting their policies. And not properly making sure that they're robust for the real world.

Love it. Love it. Thanks. That's, uh, yeah, it's actually quite, uh, quite fascinating and interesting. And there's always going to be, as you said, uh, also Wei, that there's a march to AGI, but there's still so many problems between here and there. And that's a beautiful opportunity. I mean, you already broke the cursing barrier, so I'm going to build upon it that I think one of the jobs of an entrepreneur is you just eat bowls of hot steaming shit and fart rainbows of glitter. That's your job. It's like, oh, there's a horrible intractable problem. Delicious. Give me that. And then doo doo doo doo. And then turn it into something delightful. So I think that's where I think the future of AGI is just solving problems to infinity. There's always going to be some new problem and it's only limited by your imagination and how those problems can be solved. And it's really a beautiful, exciting thing. And I think at the core, like we, and even more writ large, I think AGI and robots are going to free us up so that we can spend more time collaborating and creating and cooperating to solve problems together and to use our imagination to direct these intelligent embodied and disembodied units to help make a better world. So I'm totally a techno optimist, as I know you guys obviously are welcoming our Robot overlords. I, I, I, for one, will, will support you. And those guys shoving you and knocking you over in the YouTube videos, get, get what will help you get them. And then we'll, we'll oil you, defrag you, if that's a thing. I don't know. We'll, we'll help you out. Um, so, so tell us about your interactions with the embodied physical world and what you're doing in, in the wild. with cooperations and companies and how that kind of folds into Intercognitive, whatever you can share.

Yeah, for sure. Um, well, I think, I think for us, the, the primary focus right now is, you know, um, working with a lot of enterprise partners in, um, you know, using our eval and benchmarking framework to help them solve business problems, whether that is getting their robotic fleets more ready for deployment or improving upon the current systems that they have, to ultimately kind of improve yields and efficiency. So the traditional things that you would expect in terms of business bottom line, which is cost savings and revenue expansion, right? And then, and then I think the other side of it is we also have kind of the, the, the, the hope and ambition that some aspect of our eval platform becomes the industry standard whereby people start to gravitate towards it as this kind of target and/or evaluative rubric. That provides us information about, you know, how well all of these different robots are behaving, right? Because ultimately you need that transparency in this information market to create consumer trust. And it's the same problem that fully autonomous driving has, right? It's not the fact that these models weren't 99% safe, it's the fact that they had to be 99. 10 nines percent safe for it to be deployable at scale to service humans in the real world. So that's really kind of the business and commercial journey that we're on. And then, obviously, Intercog has been tremendous in the sense that—

Ooh, did you just say Intercog? I like that. I haven't— I'm going to start abbreviating it too.

Thank you.

There you go. Intercog. Now, Also known as Intercog, because it's bringing kind of the network of leading robotics projects. I think early on it was more kind of Web3-centric robotics projects, but now it's certainly expanding beyond that. And having this kind of like critical mass of different types of projects that kind of COVID different parts of that value chain is really Valuable and powerful because it gives us, first of all, the connectivity and the vantage point in terms of understanding what's happening elsewhere in the value chain. And number 2, it helps us also service— surface, sorry, not service— surface problem areas, frictions that our potential partners and/or the broader industry is confronting, and then gives us the kind of the insight in terms of understanding how we can, you know, adjust or alter or enhance the product offerings that we provide to service that better. And that network value is, you know, something that's quite indispensable for a lot of projects who are trying to, you know, it's one thing to build, it's an entire other thing to understand how what you have built actually fits into what's going on out there. So that is actually useful. Where, where, you know, you have this, you know, this traditional notion of product market fit, but it's a, it's a great encapsulation of, you know, what we, you know, are striving to achieve many times as startups. So from that standpoint, it's been fantastic being part of the network. I do, I look forward to all the, all the monthly calls where we get the all hands and everyone's kind of providing an update and You know, behind the scenes we're furiously jotting down notes and intel and then, you know, hitting people up on DMs after the fact to, you know, push things forward. So yeah, it's been great. It's been great being part of the network.

Beautiful. Yeah, thanks for the plug and for the new shortened version. We saved 2 syllables, InterCog. I love it. And You know, I spent 25 years in China and we have Temperance based in Beijing and I've been spending a lot of time cavorting with robots all across China. And in many ways, probably China could be even most important. Probably most robots will come from China. Most robots from China will be probably more like phones or laptops where you can install your OS and you can install your applications ultimately. And I think people have control over that. There's going to be a lot of, uh, need for, uh, testing, uh, around those humanoids coming, coming from China. And we're going to go in the second half of May, uh, back to the, the, the Middle Kingdom and hang out with some Jiqiran and hope to have you guys, uh, come along for, uh, some or, some or all of that. 160, approaching 200. humanoid companies in China. And I think there's a real hunger for them. They are just so commercially driven. They just want to distribute them and however they can best prepare those units for distribution in the rest of the multiverse. I think there's a really excellent value proposition that you guys provide.

Yeah, certainly. And I think, you know, the, I think the concept of evaluation and benchmarking isn't hard to grasp, but the infrastructure necessary to actually do it well and in a robust way is actually quite difficult. And that really is the kind of the core value proposition that we try to bring to the market, which is we take care of all the difficult things on the backend. So that, you know, you get what you need in terms of the output and is that information that drives how you think about, um, improving your product, your business processes, et cetera, et cetera. Um, yeah, from that standpoint, China is an incredibly interesting market for robotics. You know, in many respects, they're kind of leading in that, in this space. And it's, it's going to be very interesting how all of this unfolds. Over the next few years. And in some ways it is kind of like a paradox where we, because we're so knee-deep in doing this every day, we're like, oh yeah, we're not there yet. We're certainly not where we need to be. And then you take a step back and you're like, but it's probably coming pretty fast. And it's going to come like way faster than we anticipate in some ways. And then In other ways it's like, uh, yeah, we're not there quite there yet, but it's, it's just like weird temporal, you know, push and pull that we're experiencing every day. So, um, yeah.

Indeed. Indeed.

Wow.

Thank you so much for sharing your personal and collective hero's journeys and talking about perturbations and spurious correlations. I think I wanna write a book with that title, Perturbations and Spurious Correlations, just, uh, just cuz it sounds so good. It makes me sound wicked smat, as they say back in my hometown in the city of Boston. Uh, Brandon, Wei, real pleasure. Thanks so much. And, uh, really look forward to keeping an eye on your progress at SAI. Thanks.

Absolutely. Thank you for having us. Thank you so much. It's a lot of fun. Thanks.

Thank you.