#292 Brett Adcock - Shawn Ryan Meets a Humanoid Robot

Found cover, an AI security company using NASA jet propulsion laboratory technology to t to detect concealed weapons in K through twelve schools. That's amazing. In late twenty twenty five, you launched HARK, a new AI lab self-funded with a hundred million to build what you call human centric AI. You've raised billions in venture capital and time named you one of the one hundred most influential people in AI in twenty twenty four. Married

And a father of three children. And before we get too far into it, we always start off with a gift.

So they get the opportunity to ask every single guest a question. This is from Stephen Casey. In today's marketplace, we find that AI platforms can sometimes invent answers rather than admitting to l to a lack of information. Combining this in the physical realm of robotic action seems to multiply the downside effects exponentially.

What safeguards are in place that we can put our trust in to prevent the potential for downstream harm to humans as a result of bad programming or computing errors? Yeah.

What you really need is like a cheaper electric uh humanoid that like uh you basically can like use use like neural nets, like use like basically an AI first strategy with. There was there's just none of that existed. I think we're like we're thankful now.

Like looking back, like that we like uh feels like we somehow pulled like ten years of the future forward. We have like electric humanoids at like a reasonably priced that can do like uh useful human work with neural nets. Uh and it's just like uh I think we're s it's it's just an incredible

It's an incredible place to be in, getting those questions, which is like how do we make this work now at scale in a safe way?'Cause um, you know, that's this that's the spot we want to be in, not like uh trying to make this work for twenty years. Yeah. Um

So I think I think the it's a very it's I mean, this is a very, very tough problem. We have to get the product cheap enough. We have to make enough of them. We have to make it like but the performance work in like very complicated things like walk around a house. and like do dishes, like laundry, like

very complex things like small kids can't do this. Like it takes adults to kind of do like this level of work. And we need that all done in a mechanical system that doesn't have any humans around for maybe most of this that does it autonomously and not makes any mistakes.

And then like like like your fan mentioned, like we have to do it safely over time. It's just, man, it's just an incredibly complex problem. Um, I think for us, like we have a safety strategy, both uh intrinsically, we want the robot hardware and the robots around humans to just be safe all times.

And separately we have a there's a bunch of like semantic safety and other things that we need, uh like that we're we have either put in place or putting in place now to make the robot just like work safe in the environment. Like uh You have a candle at home, you don't want the robot to accidentally knock it over. Uh that's like a s that's like an intelligence thing, uh, in a lot of ways. Um, or there's a boiling pot of water, like making sure we're like very safe around it.

Um, and then there's like the intrinsic safety of like making sure like this this mechanical thing in your house is like safe around all everybody around it. Uh, I think the r the direct answers is like still a lot of wood chop, uh of getting this play the getting this thing to a point where it's like uh Uh we uh trust it to be autonomous next to my kids all day long in my house.

We've had we've had many robots now throughout my house and testing for the last like a year or so. And um I've had them like, you know, kinda near my kids in some aspects, but we're always like we're always like monitoring it and

But like we we need like millions of robots to make an impact. That's just gonna take some time and it's gonna be crazy cool. Um so we're just we're at the start line of that happening, which is like how do we get AI out into the physical world at scale? That that that'll for sure work and it'll go really far in our lifetimes. And then separately, um We have AI now that can use computers like humans and can think.

these little mini humans that can do human like work and they can think and use computers and use machines. And um I mean, that's gonna lead to such a productivity. Um like we measure like GDP per capita, like per human. But if you're able to make like as many synthetic humans, like you know, millions, billions, tens of billions of synthetic humans, in the case of the digital world, maybe trillions, um, that'll lead to the I mean, I think the greatest

uh increase in productivity we've ever seen in our lifetime and ultimately re like r reduce goods and service prices to unprecedented levels. Like a like a true age of abundance.

Trying to call, you know, call the car service and then trying to get on my flight and, you know, coming here and it's like or ordering lunch, like all this stuff I'm doing all day. And I don't want to do any of that. I want to be like fully free.

in our lab at Hark that can use computers like a human can. It can talk to you. It can like uh I just I made a phone call to ours right before we started and talked about my schedule and how to ask for things and I ask it for things and have it do things.

You'll have it in the commercial workforce and the billions, like manufacturing and healthcare and construction. And then in every human at some point will have a humanoid just to do all that busy work for you. And not only that, but like something to come home to that you can talk to that will like will know you.

I'm excited to introduce to you the newest member of my family. We call him Stanley. We got Stanley this past Christmas, and pretty quickly my focus became making sure he was safe. While still giving him the freedom to actually be a dog. I went looking for the top rated GPS fence, the number one, and that's how I found spot on. Stanley wears their Nova collar.

With spot on, I set up a GPS fence for Stanley right on my phone. No physical fence, no leash. I just walk my property line or draw it on the map, and that's the boundary the collar recognizes. I can create multiple fences, save them, and adjust them whenever I need to. So whether we're at home or traveling, Stanley always knows where his boundaries are.

Another thing that stood out to me is these collars are designed right here in the USA and assembled in New Hampshire by a team that's been working with high precision GPS technology for years. And you can tell, a lot of attention went into making this thing reliable. The NovaColler uses a dual-band GPS system connected to more than 150 satellites.

Along with an antenna that's over five times larger than typical GPS fence collars. That keeps the boundary accurate, even around trees, terrain, and changing conditions. Spot on's true location technology has been independently tested and delivers ninety-nine point three percent containment. which matters when you're trusting something with your dog's safety. It's incredibly durable, and on top of that I can check his location in real time, send voice commands directly to the collar, and

Track his activity through the day. I use spot on so Stanley gets the freedom to run and explore, and I get the peace of mind knowing he's safe. Let your dog roam with spot on. Go to spotonfense.com slash SRS and use code SRS for$50 off the Nova collar. That's spotonfence.com slash SRS and use code SRS for fifty dollars off.

Well, I would like to do a little bit of a live story on you if that's does that sound good to you? Yeah, let's do it. Where'd you grow up?

But yeah, no, I got I got really interested in computers like really young. Um Started a bunch of like startups in like in like you know, in high school and college. Um, you know, at first just like uh mostly things on the web, like selling things, uh um, did a bunch of like uh like different types of um like

products I was selling on the internet for like throughout like high school and college. Um small like drop shipping, retail electronics, like all kinds of all kinds of things. Uh Legion marketing and just fun stuff. It was like not nothing serious, you know, just like uh playing around the internet, trying to make some money. Just I didn't grow up with money. So it was like internet was a way to like uh like, you know, maybe make some money. Like it was really fun. Um

You know, I loved like the ability to go out and create things and uh kind of control my destiny. So it was just something I I I attached to really early on. Right.

And then uh he just moved out to California. We live literally a block away. Uh I see him almost every weekend. He has a startup like basically 10 minutes away from uh from mine of you know, where I'm at now. And uh he's he's doing great.

you know, like do what you really want in life. You need to like run your own business. Mm-hmm. And that was like beat into our heads like growing up. Like uh, you know, at some point you need to, you know, you need to get probably get out of here, get out of farming, it's not doing well. And uh you need to start some on your own. And uh so it just Like kind of just like by default I was like, Okay, this is what I'm gonna I'm gonna go do since I was a kid. Yeah.

I just like always sat there every day like wishing I was working on hardware. stuff I could like touch with my hands. Uh stuff like when when growing up was like, you know, build I was like rebuilding computers or just like on the farm and building stuff. Uh I always like envied things that you can go touch and build.

I I got really um kind of going through college, it's like you gotta look for a job, you gotta go find something full time and um got caught up in like the whole interviewing process of like looking for jobs. I just thought it was so broken. like applying for jobs and like never hearing back and like you have to go through headhunters and then it basically became like a

Can we make matches at scale? Like without like any humans involved? And uh like the head hunting industry is like hundreds of billions of dollars a year.

At the time we were like, let's use AI, but it was basically like uh how do we use a lot of algorithms to figure out like what people want and then make matches. So you can just like push up a button, connect the right folks. and then uh make placements and then we ended up charging most of our revenue came from subscriptions from big companies like big banks or startups or tech companies basically looking for talent.

Um we started just in tech in the U in the US. So um at one point we had about I think about a little under 20 or so cities globally that we were operating in. Wow. But most of it was tech, uh talent. The tech space is, you know, at that point and still. Started in twenty twelve and then ended up selling the business in twenty seventeen or twenty eighteen. Right. So about five years, six years. Right on. Yeah.

And at the time we were like I was like completely dead broke and put everything out of the business. It was like I think it was at the point almost seven years in. And um And uh, you know, they they c you know, we were uh excited about an acquisition a year before that at ten million dollars from one of the big tech co companies. And um they came in at a hundred and ten million and um

And it was a it was a good time for me. I felt like uh the business was doing well. I learned a lot and I was kinda ready for my next chapter. So I ended up selling that business, uh, to the Deco Group. It's like the world's largest recruitering company. Um and uh

And there was like there's something better to like a human jam and you and roles, right? Like it's just like you need like um and then to extent you can get, you know, all the world's uh like talent there and all the world's uh companies looking, you can really create an amazing environment where you can get people to like the right job. And right now it's not like that. It's like a really black box, like trying to both finding talent and looking for a job. It's just a terrible experience.

Um so we kind of that clicked. Uh the yeah, the world's largest recruiting company came in and said, We we gotta buy this thing. And um

It's a it's a long process you have to go through, like tons of docs and then you announce the announce the deal, then you actually close the deal and then it it went into escrow, then it finally hit my account. It's kind of like one of those processes.

And um I want to go work on something really important, hard and uh a couple of industries that I've I've been interested in robotics and aviation and some areas of security for like basically since college. And um I basically spent a lot of time trying to figure out if I was either going to work on at the time sch school shootings like basically 10x.

And I was like, man, there's got to be something to do here and we can, you know, and uh and then secondly, um, I really wanted to work on like flying cars. Uh like having a watch lots less sci-fi as a kid is like, man, like um I really want to go. I there's a there's a there's a a near term problem of like we gotta go help. with security in schools, K through 12, mostly in the US. And then um and then how do we I want to work on flying cars.

Um and I ended up making the decision to work on flying cars at the time. Uh so in twenty eighteen, uh shortly after the sale of Vetery, I started Archer Aviation. And um basically like the story here is um You can build like an electric

the aircraft inside of cities. Half the world lives in cities today. It'll it's it's um, you know, by like in middle of middle of the century be like seventy percent of the world. And you just like can't get around. Like it's just gridlocked everywhere in major cities. It just like sucks to go like twenty, thirty miles. It takes like an hour in most cases.

So basically you can build a design an aircraft that can take off vertically and then fly like an airplane. So you can get like a lot of distance. And you can basically then re-architecture the whole aircraft to be fully electric. The reason you want to do that is is for cost and safety. You basically can make it like like a lot like less expensive. You can put a lot less parts in the aircraft that are also good for safety.

So basically you can build like an electric flying car that you can move around. So instead of like calling an Uber or driving that might take you an hour in LA or SF or New York, uh you basically can fly there in 10 minutes. And if we if we can pool everybody together like in a kind of like an Uber pool style um business model, you can do it for as cheap as an Uber. But the time the problem was like I didn't know I didn't know anything about how to build electric aircraft.

I try to learn as much as possible about three subject areas. First was like electrification, which lot like you know, at the time like electric vehicles were like really doing well. um and and even drones. Uh vertical takeoff and landing, like vertical lift, which is like traditional like rotorcraft or helicopter. And the third is like winged aircraft, like airplanes. You really need wings. Like so you be okay. So you basically have to learn about those three subjects.

So I started I ba I basically bought my my my basement downstairs at home where it's like every possible book on these subjects you could imagine and started reading as much as I possibly could. And this was during the year transition as I was transitioning like out of out of veteran into archer. I was reading every possible thing. And then I found a small community of folks that were like hosting uh on-site uh like either half week or week-long courses for this.

Um, and so I would go to these, sometimes they're sponsored by NASA or by colleges or whatever it would be on like basically rotorcraft or electric propulsion um or winged aircraft, aerodynamics, and I would basically like try to like like a learn as much as possible.

It got to the point where I was like completely obsessed with this algorithm I was building on electric aircraft sizing. Like how how do you actually like how would you actually build an electric aircraft? So electric aircraft, what's interesting is like in rotorcraft, um, like

You basically want to make to create the most efficient lifting device possible. You need as much of uh like as the rotor disc area, like like in terms of surface area as you possibly can. That's why the helicopter rotors are so large. You really want that to be really large. That'll reduce power and get you up off the ground. In electric aircraft, the problem you're starting with is you have like one thirtieth of the energy is as you're doing kerosene in a battery pack.

Wow. So you're just like you're off the bat of one thirtieth less range or less one thirtieth less energy. And so like power becomes like the dominating factor of like uh of how to basically build electric aircraft. Like how do you get power down as much as possible? You really want a lot of disc area.

Um, a lot of disc area uh is w well one, it could could be good for power, but it's also bad'cause you have like no like no redundancy in the system. You have like one rotor blade that if it didn't go doesn't go well, you like you go down. With electrification, you can basically build uh much smaller um uh uh like basically rotors and be able to fully electric. And the reason you can't do that with a like traditional um kind of like turbofans or engines is it gets too inefficient at these sizes.

You can't build twelve. like uh propellers on a helicopter. Gotcha. The efficiency just drops like to nothing. Um so with electrification you can. You can size down electric motors to small sizes and they're still ninety percent efficient. So like small electric motor on the table or a big one the size of your chair. Same same efficiency. Uh and when you do that, you create a lot of redundancy across the system. So you can build like an aircraft with twelve electric motors.

So it was actually at a I was at a Hyatt Regency hotel in Atlanta in 2018. Uh I was on a it was an electric propulsion week-long design course and like an aerodynamics course for winged aircraft. And I met a guy there that um that was basically in the engineering department at University of Florida. He was doing his PhD in aerospace and asked him what he was doing there and

I was like, oh, I went to school there as well. And um he's like, uh, I'm like, what are you doing here? He's like, oh, I want to go like do a career in EVTEL aircraft. And um it's this is called electric vertical takeoff and landing. Um so you know, a a helicopter is a VTAL and a you put a little E in front, it's over electric. And I was he's like asked me what I'm doing here. I was like, oh I'm sorry I'm starting a company to do this and I need to figure out how to go build these things.

And um he's like, Well, listen, my professor runs a small drone lab. He's got a full building, he's got twelve PhDs. Why don't you come down and like meet him and see if you can start building aircraft with him? So I flew down that weekend to go meet um his professor that runs all of basically mechanical engineering and aerospace.

And uh long story short is I ended up taking over his like his lab. And um and me and him and his team started building aircraft in 2018 and 2019 down at the University of Florida. And I temporarily moved down there with my daughter at the time and my wife.

living in Gainesville, Florida. Like, um, and um and it was great. Like we ended up building I ended up funding a lab uh right off of Archer Road, a new lab'cause we needed more space. I mean calling business Archer Aviation. It was the main road down of University of Florida. And I spent the next like year, year and a half like um modeling and building electric EV tell air.

About basically a few years into the business. Um, and then, you know, things took off from there. We'd like built bigger aircraft. I took the company public within three years of starting it. We're like a six billion dollar publicly traded company today. And um yeah, designed basically like now four or five generations of aircraft at Archer and um

It was hard. Uh, you know, um, it really set me up well for like, you know, doing figure and cover and the rest of stuff. We we can talk about later. But like uh it was um it was it was hard. Even going public was like probably like one of the hardest experiences.

electric vertical takeoff and landing like companies. Like you know what I mean? The big V S venture capital groups were not funding SpaceX or Tesla or Rivian. Like none of these were getting funded by traditional investors. They weren't raising money from the named investors we all know about now.

And uh and so like I end up I end up going all so I, you know, like um, you know, made just made$110 million and or just sold the company for$110 million. I made a lot of money, like personally, and ended up going all in on Archer. through the IPO, like uh through going public, I I put like all all the money I basically ha I bought a house and the rest of the money went all into it. Right up and um

And it was a stressful period. So we went went public through this back. And the reason it was tough is um we ended up getting to a point where we just couldn't raise enough money privately. Like it was um it was either like raise, you know, a hundred million dollars privately at like, you know, some valuation, three, four, or five hundred million dollars, or is or it was like go public and raise like a billion dollars. Wow. And uh we end up going public and raising a billion dollars. Wow.

like within the first two years. Wow. And uh they retaliated by just like trying to harass us while we were going public. And um so yeah, it was just a crazy story, uh getting public. Uh ended up getting public, you know, billion dollars on the balance sheet and we just started building like aircraft and um started building the service, like think about the app and how you're gonna check in and how you're gonna build uh places like real estate to fly into.

And um yeah, and then like the engineering work we had to do around there of designing, you know, it's basically a it's basically a flying robot. You have like uh battery systems, electric motors, sensors, embedded software, and control systems. And um basically like the like the robot you saw this morning. Uh like very like you know, that's like it's a flying, it's a 6,000-pound four-passenger piloted robot. And it has 24 degrees of freedom on the system, like wing, wing flaps.

Uh we like we have uh we we tilt the front, like the leading edge, six uh motors, 90 degrees for basically takeoff vertically and then go into forward flight. And then all the propellers, uh fan blades on uh have variable pitch propellers. So it's like a highly overactuated system that needs like a really good software. Like no human can like fly it basically without uh really good uh control software.

I know there's a lot of pushback about that, you know, for for for from an overreach standpoint, but if you th if you just think about the traffic in the cities and if you have the AI, you know, processing all this. that even even without air vehicles, I feel like a lot of that would go away because the the the the AI will route you the quickest. Yeah. It take all the traffic patterns into account and it would just flow.

retrofit all the cars in the world right now. Like it's a it's a hard problem. Uh well.

uh for the technology roadmap here where like we're living through it and it's like a little messy and it's not quite like straightforward and uh we don't quite know where it's headed next. But where it's headed is at some point in like five whatever years.

Where, you know, you know when our kids grow up, like they're never gonna have to think about this. It's just gonna be autonomous from the start. It's gonna be like, you know, by default native. Yeah. And uh it'll be trusted and easy and safe. I mean, we're just like living through this period right now, which is like a weird thing. But like If we close our eyes long enough, you'll have this autonomy and electrification everywhere.

Car smells or it's dirty or whatever else. And it's just like, you know, this is it's just easy. Um, it's really cool. So like the technology is like uh in the like the early chapters, but it's all here. Like we're we're gonna have autonomy as scale, like everywhere. It's just gonna take some time to roll that out. It's the it's the time it takes to

Um, get the technology mature enough where they can run enough cities, enough places. And then it's the time it takes to get the install base of autonomous hardware and software in all these places. That's gonna take some time too. We just can't snap our fingers. We just don't have enough. install base of autonomous vehicles in the world. I think like Tesla's got like what, ten million cars on the road and like um maybe there's thousands or so of like Waymo's. But you have like a

over a billion cars on the planet. So you need to like like a like a you know, make m a large fraction of that all autonomous. Um so you're looking at like a this isn't gonna happen in a year or two. It's gonna take some time.

Uh, that process moves at the speed of like the post office. And the FAA is not incentivized to put anything in the air unless they know for sure it's gonna be really safe. The safety standard for us that we wanna certify to is one times ten to the minus nine. In terms of hours of reliability before a a catastrophic event. So that's um one in a billion hours. We can have a billion billion hours.

You can't I to be able to s uh that that that is like uh that's the sta but that's the standards when we fly, it's like one of those it's the safest form of transportation we take. And it's because of those standards like uh uh governed by the FAA.

Um, which is great. I'm like, you know, we're like that's that's w that's the bar you need to be at and that's the bar you need to hit, especially taking passengers over cities with aircraft overhead. Uh, you need to be at those levels. So uh that's like the

That's the long pole in the tent for us. And that's wherever you go. If you go to, you know, Europe, it's it's Yassa uh or you know, CAA in China, where where wherever you're gonna go, there's like there's federal mandates to get to get basically an air aircraft to take passengers. So we're we're in the middle of FA certification now. Uh we hope to be certified in the coming, you know, as soon as possible, basically. Um

But it's like it's not something you can like there's not like a date on the calendar of like you'll be certified here. You have to work through a very like uh very long and slow process with the FAA to get through this. And then we're uh also uh dual tracking that against a couple of different uh entities globally. Uh to make sure we can get certified and get in there.

Um, but it'll it'll happen, man. The aircraft, um it just again, if we're in like this chapter, we're like flying cars, electric, you know, aircraft are just like it's early. It's earlier than like uh AV, like AVs or EVs, uh autonomous vehicles, electric vehicles. Uh but it'll happen. It'll happen in our lifetime. We'll be taking these things around.

So you uh you have left with no choice if you and then humanity are moving to cities. We have this like secular trend where we all want to live in cities right now. It's like half the world lives in cities, it'll be like seventy percent by twenty fifty. Um So we're like all moving to cities, the roads can't grow anymore and we want like we're like we're constantly moving around, going to work or going to restaurants and just like and it just it's like it's like um it's like this uh

It's it's it's getting it's basically getting worse. It's like this the arteries are hardening here around the around this. It's getting worse and worse. And it's just like it's some of the worst time to spend on a road in traffic. It's like so soul sucking. Yeah. It's just like the worst like like time to lose. Um so the good news about the air is it's three dimensional. You can stack like basically an infinite amount of like say roads uh in the air.

Yeah, exactly. Like just like dig tunnels and it's great. The problem with t the only problem with tunnels, um With the sc with with the with the node system uh uh on on um on the ground, or sorry, in the uh on the uh uh on the ground with uh with with like say like we call them vertiports, but basically uh like real estate for flying cars.

is you can let's say you had like, you know, 10 different um or even like, you know, five like you say 10 different verte ports inside of a city, like places to like take off and land from, you can travel between any one of those routes. Uh so opens up like, you know, basically exponentially more places to go to. I can go to like any node on the system at any time.

And then uh and then you'll get on these and they will go to any node on the network. You can't do that with tunnels. Tunnels have to go to A to B. You can't like you go from like you can't jump to another tunnel uh downstream. Like, you know, I want to jump to another tunnel like a hundred meters down. Like it doesn't happen in tunnels. You can do that with a sky. You can basically jump to any node on the network. It's exponentially more routes you can basically do with like less real estate.

And then you can basically stack as like you know orders of magnitude, more traffic and humans in the sky. So my envision is that like you're gonna be for most most trips that you know that you're traveling over 20 minutes, all that will move to the sky. And not only that, but you will you will have us you'll have like cities being re like um being transitioned to a point where you can live well outside of cities and get to cities really fast.

The reason we live in cities is because we're like we're working there and we have friends there and we have like it's yeah, it's like I want to be like I want to go to dinner with somebody, I want to see my buddies over here and uh we wanna go to work over here or like go to the mall over here. It's like everything is there.

And that's what we want to be or social creatures. We want to be there next to other humans, or some of us are. And um so uh yeah, so anyways, uh, but like, you know, now that you can fly this 150 miles an hour in the air with no traffic, point to point, like no stop signs. No construction, no things jumping out in front of you. You don't have to like travel different distances. You're going straight from A to B in most cases.

So you're like you're removing 10 or 20% of the top like of basically the distance just by going point to point. And then you have nothing stopping you going 150 miles an hour most of the way there. You can live like far outside of cities and get down to city center in under 30 minutes.

It's whatever, it's forty bucks and I'll be there in under thirty minutes and you'll you'll you'll say, Great, I wanna be there at that time. You'll hit a button, it'll be on demand, you'll ride your bike over or walk, you'll get in one, it'll leave in seven minutes and then you're basically flying right down to town.

And um, and so like uh it just like from a safety, from a part count, from a cost, uh, from like acoustic signature, it's not gonna like helicopters are loud and very noisy. And um like it's it it's just a much better technology for this.

No wanna like go do this and then you find out it's like, Oh, it's like a twenty forty event and you can't get it done. It's just like a it's just like a waste of time. And um So the good news, you know, for Ultra is we like we're we're in a sweet spot here where uh this is gonna happen. Aircraft now work. Uh we're certifying now with the, you know, government bodies like the FAA to make it happen. We have a good balance sheet with it with cash.

Um the team's great. And so it's just like uh, you know, get certified and get this thing going.

And I have this big belief that, you know, we like we have like we're it's like weird biological species. Like we look we're like, you know, we have these weird hands and arms and legs and certain height and um sensors. And then we ended up building this world around us so we can interact with it.

Um, I mean if we get dropped into Mars today, we're gonna build like coffee cups that we can hold and stairs and doors and we're gonna build this stuff again. And it's like the it's like the human operating system. We're building things we can like use and operate in um that makes it like easy for our lives. And we built it around the form factor.

that we are. Meaning if we look differently, the world would look different. Our espresso machine would be look different looking. Uh we might not even like espresso or call it caffeine in this case. Um so we built this whole world around us. The holy grail for robotics is can you basically build a general purpose machine that can do what humans can, which for me is like a humanoid robot. And a humanoid robot's just a a robot that has like a human form.

So has legs so it can walk upstairs and walk over like, you know, uneven terrain or say things on the ground and um bend down, which are important legs are important for, um, or reach up. uh has like arms and hands so we can manipulate objects and do things like, you know, grab the stuff, open open this this these gummies and um

you know, fold laundry and do do real work. Um, and then we have the right sensors so we can like see the world and understand what to go do and use a you know, our biological neural net to kind of figure out how to reason from. And um You know, having worked on like you know. kind of like aircraft for, you know, now five or six years, I I thought it was like pretty possible to go build um an electric humanoid robot. And electric's important for cost.

And it's important for safety and it's important uh because the performance will be much greater. And at the time, I mean one of the best humanoid robots at then was probably like the Boston Dynamics Atlas. It had like a hydraulic system. It was like really heavy and big and high torque and very leaky, like the oils everywhere. And uh also didn't run very like maybe ran for 20 minutes on a single charge.

So you need to kind of radically transform the hardware. And then you needed to figure out a way to build like um an AI brain. The humanoid is so complex, it has um has like let's let's call it like uh like 40 degrees of freedom and degrees of freedom is like a joint so like an elbow is a degrees of freedom and you know shoulders got three a ball and socket has three like a pitch yawn roll and our robot has about let's call like 40 degrees of freedom in it

Each degree of freedom is a motor that can spin three hundred sixty degrees. So if you want to look at like how many positions the body could be in at any given time, like this is a position, this is a position, and keep moving, the amount of states, it's uh the mathematically it's 360 degrees to the power of 40 actuators. So there are more states in the robot than atoms in the universe. There's more positions the body can be in.

And I'm moving my fingertips and my hand like my you know, my hand my my wrist and hand uh get in position to grab this. Like it's uh It's an intractable problem for code. So um

And we just fall over just like uh yeah, we have to fully balance, you know. We're it's dynamic. So it's uh if you if you if uh generally if you power it off mid run, it's gonna just fall down. It's not like a four legged dog or quadruped robot where like at any given point it's usually like uh statically stable. So it makes it very difficult because you have to be able to even move your hand. I'm moving my pelvis.

And my whole body, my torso's moving, my head's moving, like uh all of it becomes very complicated now. It's not just like move my hand, it's like move my whole body to get my hand in the right spot. So every joint, all those 40 joints have basically position encoders. So we know exactly what position the motor is at. or even the case of the knee or this. And we have force force sensing, torque sensing on board. We have uh the ability to detect all the forces that that knee is seeing.

It could be really high when it's walking or it could be like, you know, it could be like it could be powered off and have no forces on the leg. All of that feedback and is is being sent in the main computer. And then we're telling uh all the joints what to do over 200 times a second.

Uh some of the other feedback is happening at like s five or six kilohertz. So the force feedback's happening six five six thousand times a second to the motor control uh on board. And we do the motor control, the brain for all the motors has done it locally at the motor level because it needs to happen so fast. That's being fed back to a main computer that runs a control software that tells the rest of the whole body what to go do at every timestamp to keep balance.

Shopify is the commerce platform behind millions of businesses around the world in 10% of all e-commerce in the U.S. From household names like figs. To brands that are just getting started today. One of the biggest hurdles is efficiency, and Shopify helps you accelerate that. Whether you're uploading new products or trying to improve existing ones, the platform is packed with helpful AI tools that can write your product descriptions.

page headlines, and even enhance your product photography for you. They also allow you to tackle all your important tasks in one place, from inventory to payments to analytics. There's no need to juggle multiple websites. Everything is in one spot to make your operations. Smoother. And you've definitely seen that iconic purple Shop Pay button is used by millions because Shopify has the best converting checkout on the pl.

It helps boost your conversions, which means fewer abandoned carts and more sales for your business. It's time to turn those what-ifs into with Shopify today. Sign up for your one dollar per month trial at Shopify dot com slash SRS. Go to Shopify dot com slash SRS. That's Shopify dot com slash SRS.

Humanoid robots in AI software to try to see if you could make this work. At the time, there was no there was no good precedent. There was no precedent for showing that. There was no AI that ever worked on a humanoid robot in history. And there was uh no humano electric humanoid hardware that. was e uh remotely okay to show it would work. And there was no hands, there was none of this stuff.

So I actually had I actually had a lot of trouble even ra I so I th I had a lot of trouble early on even like getting people excited about this because they were like, What the hell are you doing?

uh, you know, control software, embedded systems and sensors. And then even within like electric motors, we build actuators. They have like a rotor and stator and a gearbox and sensors, uh, electronics and wiring and connectors and uh multiple sensors inside of there, uh, and then firmware it lives on the the like say say the microcontroller lives inside the on the motor control side. Uh and then we have like thermal characteristics it's hot.

Um, and then you gotta all make that work at very like high speeds and high torques, meaning motors don't like working when they're not moving. Motors hate not moving. Motors want to run on h highway speeds.

Okay. When motors are stuck and not moving, but holding power and holding forces, they're uh they're all it's a really bad point of the torque speed curve. So um they're they're not they're not built well for this. So humanoids use all the time.

Like we're like when we're standing, we're like not moving but holding forces. When we're like holding something out and like hold like giving me the gummies, like it's not moving now, but holding forces. It's just a really hard engineering problem on just that one little aspect, which is like hardware umbrella, just like just just motors.

And so we have whole teams in those areas, just in that one little area here, doing like rotor design, um, electromagnetics design, SATA design, gearbox design. Sensor design, electric like motor control design, uh like all all of this inside of teams. It's it was an enormous lift to just get the team members there to do it. So we spun up a team to go uh operate that really quick. And then um, you know, now like looking back, I think we raised, you know

You know, well, two billion or so like now it's a much different story. We have you know We we've we we ended up building figure one, which is our first generation robot, and had that walking in under twelve months.

You know, and and then from there continue to build the capabilities. We built built generation two, figure two, which is that guy right there is our second generation robot. And um and I think w one thing we did before we kind of moved even to Gen while designing Gen two is um I think it's probably like twenty twenty three at the time we um we did a demonstration where

we basically wanted to put this K Cop K cup inside this curk and run it. It was just on a very pretty simple, like like not nothing crazy, but we had a you know curk machine, coffee cup. in a K cup and we had to go grab the K cup, open, you know, open the curic, put it in, close it, run it, and then, you know, make coffee. And um, you know, it sounds simple, but like for a humanoid robot to do that, Uh is extremely hard and then we wanted to do all of that with just neural network.

We were like, uh internally we were like, you know, how do we get neural nets to run on a humanoid? And I I don't I don't know. Ma I I I think it's probably one of the first examples in the world to ever have shown that. And In tur I you know, this was like game on. This is like we have we let's go build really good humanoid hardware. Let's make it cheap and really reliable. Let's make sure it can do what humans can from a hardware perspective.

Meaning you want to look at like a phone, like where you can just add new apps to it, like the uh do laundry app. And the hardware doesn't need to change. in the same exact hardware. Like humans don't need like new I don't need like new hardware uh to be able to go off and like learn how to do new skill now in the physical world.

So you wanna build the humanoid hardware so it's like uh can do everything basically a human can or as most of possible. And then you wanna go all in on neural networks because you just can't code your way out of this problem. And that was the first moment in 2023 where we're like, hot damn, this is gonna this is gonna really work. This is gonna be humanoid robots.

Hardware gets good and then you're basically gonna be this is gonna be a data play to train neural networks to run on humanoid hardware and do what humans do. And um and then we launched, you know, uh

We launched Figure Two. We did a lot basically more work. We started unveiling Helix, which is our neural network stack internally that we do um here. And uh now we've designed figure three, which is our third generation robot you have here, which is like Oh, it's the best human art hardware in the world by far. And we're now running robots that do like, I watched it, you know, the other day, uh unload dishes and fold laundry. We had figure twos at BMW last year that worked six months.

Every single day. Every s uh

But the real like uh level boss is like how do we get robots out and do clients fire us? Do they love it? Does it work? And the goals we have for clients is hard because we have to do human work. So we get like human KPIs in terms of speed and performance. Like humans like, You know, in the case of manufacturing They don't like you might mess up every once in a while, but you like refix it. So you're not, you're not messing up every single time. You're pretty fast.

In most cases the humans are there, not like, you know, quitting or not showing up to work, but sometimes that does happen. Um, so it's like it's a hard. It's a hard bar to go hit. And we have to wake up every day and be able to do that. And so we had robots on the manufacturing line. Uh they have a they they basically have a basic body shop that basically builds like X3 and X5s. And January of 2025. What was it? Twenty twenty five, we started building our first uh

B and W X threes on the line. And I bought it. I bought the first four. that did that. They're at the on my campus now. One's at my house. And um and they uh they didn't build obviously the whole car it's like there's like a ton of parts, but like we built uh we did we did part of the whole process. And um

The cars are so big and so heavy, you can't human can't hold it and pass it around. So you basically have machines that are building the car and then moving the car. So the whole body shop line is only automated end to end. And it's like uh not end to end, but like there's humans involved, but like it the car is being like built by machines.

No, it's just and then there's machines everywhere. There's special indefectors on every machine. They're switching this things out basically in real time, like grabbing a big ma like big uh end effector. The indefector is something that is grabbing a part. These ineffectors are size of my soap like my chair. They're switching them out in seconds. Wow. One or two seconds. They're doing it really fast. And then they're basically building a car with this. There's robots everywhere. And um

So like B and W, like they're like it's it's been a privilege to see how like much automation has gone into automotive. It's unbelievable. It's these things these machines like kind of make what we're doing sometimes look like

And we were in the body shop line helping basically attach the rear header, like basically putting the rear header on this fixture. So we so today um we basically or last year when we're there, uh we basically take a piece of sheet metal and we basically put on this fixture and we do that over and over again.

And they do that, you know, t ten hours full shift. And uh there's three different parts on. Those three parts go on, this thing rotates and this big giant kookah machine, like this robot arm, goes and spot welds it. And switches it, switches out to another effector and then grabs it and puts it down the line. So like these these uh these facilities are being like fed this these parts into the machine. And we were like we were a piece of that. And um

The goal was just like, can we run robots every day? You know, I mean are we gonna get our like ass handed to us? You know, is it gonna be easy? Is it gonna be hard? And uh I think it was in the middle. I think we like um we g we got the robot to a a great spot where it was brand every day was great. I think the biggest learning lesson we took away is we had

We really ca I really cared about if can we do that and can we like clone it times a thousand, times ten thousand? Would we have any issues scaling? That was the part for me. Like is it just like, you know, does it completely shit the bed and means like we need to rework our plan and go back to the office? Does it do it incredibly well and you can just copy paste this thing to everywhere in the world? Like how how does it how did it work?

And the biggest learning lesson we got was that the robots the robot that started the first day at the start of six months and the robot that ended the shift that day was the same like even though we had multiple robots in operation every day, we had like the same robot. I did the start and the finish.

I think from a software perspective, we did like a I would say a B B job, B plus. We and that's mostly from like my perspective, like the architectural decisions I made to scale. We about half the stack we had. uh like traditional like uh code and heuristics in. So like the controller to walk was done by a C plus plus controller. It was done by code. The the walking you saw today we had back then was done in code. Okay. The rest of this we had a bunch of other stuff in there done by neural nets.

and like uh some of the perception stacks, some like, you know, s some of the how to move parts around and everything else. And, you know, this was a year and a half or so ago when we were first launching and um And I was like, man, this the the the biggest problems we're having is the the coding parts get stuck.

The robot like either doesn't see like the right uh like doesn't like see something right on the part and misses the the object detector, doesn't really understand what's going on, the controller When it gets out of bounds of like what it's ever seen before. Like you have carpet in here now and it's like really squishy. The robot's doing fine, which is great. But I think our old controller would not do well. It's like you have like really shaggy carpet.

And it's like, yeah, and so like uh and it's like not very it's like it's like hard, you know, it's harder for a robot to walk around. And um so that that that was like we had a really difficult time uh seeing that, even though it did well every day, seeing that scale to like lots of robots. So we went back to the office, this is about a year and a half ago, and said we need to basically refactor everything into a neural net.

And one of the big like in and I think we just announced Helix two two or three months ago now, I forget, like end of last year. And it's basically entirely down the stack, including the controllers and turn and neural network.

It's uh there's there's like no code left really on the robot. Um some code in certain pieces, but mostly just uh almost all of the thing is like a neural net at this point. Uh yeah. We removed the need for like almost like over almost over a hundred thousand lines of code at the when we launched Helix two. And so what you saw today was just like a like a a robot that can, you know, that we can we can put now back in the say the factory in these places that will run all in real.

And I think I think we're running these robots right now. We're getting ready for deployment to customers, and they're running incredibly well. Um we've we have robots running like basically now in 24-7 shifts. Without stopping, without any faults for uh like days and days.

Uh we just w we just went like over yeah, we just we just had like record time this past week on the robot running until we saw like a fault, like almost yeah, basically a whole week. And they basically like they they can run. uh like f four hours or so, five hours, and when you charge, another robot knows that.

Steps in, steps behind the robot, say, and gets ready for work. The robot then backs off. Another robot swaps in spot and like in the next like ten seconds it's doing work again. So we can run that now in twenty four seven shifts where they're talking to each other, all autonomous. No humans or you can go to bed, whatever.

And um they're running like shifts all day and all night. And we do we do it across multiple use cases now at the office and 247 shifts. And it's just like we're running them hard.

robot gets to some spot where it feels unsafe, doesn't know what to do, and he'll stop for a little bit. And then if the you know the robot's not on the line for a couple of minutes, we we call out a failure. And we're not happy with it. Um, we have robots that are greeters and visitor bots that walk around the office all day, 24-7. So you're over the office, you're getting lunch, or you're walking around, you're interviewing with us, you see robots everywhere.

And those run in twenty four seven shifts all day, all night, weekends, Christmas Day, whatever whatever we run them.

And whole experience. Um yeah. Right now they but they're walking ever I mean, right now they're walking in the office like at night time, they're walking the office everywhere. And it's a good it's a good stress test for us because these are neural nets that are running for navigation or uh planning or manipulation or whatever it would look like. Um

And it's it's hard. And this is a news thing. It's not like these things have been around for decades and we like understand that they're really mature. They're not. So we really stress test them like crazy by running them all the time.

Hey man, go make me like go make me like uh go make me food or like do when I'm gone today, do the laundry after you like after you like unload the dishwasher, like you know, uh do do laundry in like my kids' rooms today or something like that, or or text it. Um so like language is like super important UI. So we're like spending a lot of time on speech.

You know, if our if our like if our systems want to tell the robot what to do or command everybody to do something, we do it through 5G. And um and so yeah, you can like you can text it.

You really want to treat the hardware like this, where you basically similar to a phone, where you um you don't have to change the hardware for new capabilities. Uh you wanted to learn how to do uh like, you know, complex. towel folding or um like a lo like unloading the dishwasher, um making coffee on a curic, like all this like walking the dog. Like these are like uh like almost like the matrix.

Where you get like plugged into a system that um reuploads like weights into the like neural net weights into the robot where it can like learn new things. So that's what we do now. Like if the robot, like, if we can't do package logistics well, we get data for package logistics. We train our Helix Neural Net.

for a week and then we load it to the robot and it can like then the same robot that was like folding towels like the week before can now just sit there for 24 seven and do logistics work and package work.

Nothing changes.

You know, if we go to your house and my house, we probably have like different appliances or different toasters and different microwave, it's all a little different everywhere we go. So the home is just like this like um tons of entropy, like uh tons of veritability, a wide distribution of tasks. It's like the It's like the ultimate like challenge for robotics in in the home. It's like the hardest

most variable thing we got we got going. And um in the workforce it's like you have this like work cell that you're doing. So like in if you're doing like manufacturing logistics or you know a lot of tasks you have like this area you're doing work in and you can basically kind of write down on a piece of paper like how to do every step.

In the home you can't do that. Can't write down a piece of paper. I can't write down a piece of paper like how I can interrupt your house. I don't even have seen it.

Or like me and you feel safe like being like having this here with our kids. And that is um that that's gonna be the hardest challenge by far. And um that's gonna take some time. It's a very it's there's some trust that needs to build, is a track record that needs to be built. There's like system safety engineering that needs to be done extremely well. Um so that just and then the home, like you can charge like

10x you can charge like 10x more in the commercial market than you can the home. Home needs to be like 500 bucks a month. Yeah, Carly's.

Wow. So like you're talking about the largest market in the world is sitting in the commercial workforce. Wow. So you have like you have like that plus the variability is lower, plus you can charge ten times more. It's like the

Like for investors or like, dude, why would you ever work why would you ever do homework? Yeah. You know what I mean? Like why would you like spend time over here when you can just go over here and build like a twenty trillion dollar company? Um And I my answer for that is just like I I just want to I want robots in the home. Don't really care. You know, like we gotta make that work. Yeah.

I think will be like become like the largest company in the world. And you can do that with maybe a hundred rope.

And but there's a lot more spots to go fill for like long horizon work to just that. And we have to be like extremely robust. um, to maybe different types of clothes or like different types of like I I I don't don't don't wash my jeans, like that type of thing and all these different like variability that you might look see. Um And we haven't been able to solve we we as of today, that's like that's the hill we gotta go solve. That hill looks really hard.

When you're doing when you're doing the laundry, I want these ones washed cold, I want these ones washed hot. This is where they go. This is where the the jeans drawer is, is where I hang my shirt. Yep. Is that how it works?

that you know, needed to like cook and use your stuff. Like you're like, you know, you wanted to wash clothes and stay in one of your rooms. Like you'd walk that person around and you'd be like, hey man, this is um, this is recycling here, this is where trash is at. Uh like here's here's where you get water. Like uh the the trash goes out every you know, every Monday.

Uh, you know, this is uh I you know, we do blankets uh on the couch but like you we want'em in the c in the cabinet when they're done, you know what I mean? Like or we want these all folded and put over here like like all these things you have in your home that are like uh you know, important and um

And you're like just like you would like walking somebody a human around for the first time. That's what you'll do. And the robot will uh semantically under like will A have like will remember all of this. And uh and it will like it will learn based on w what you want and your preferences, like what to go do. Uh shit.

Uh it's just um we'll see. I mean, we're this is like it this is like solving like the holy grail of robotics. This is like solving for a good general purpose humanoid robot. Um maybe we don't solve it this year. Maybe we solve it next year. Maybe we don't solve it next year, but it's 2020. I don't know. Like we're we're close.

we feel like we're in the red zone with like we feel like we know the architecture, we have the hardware, we know we know how to get the data. We put the data in, the robot does it. We need to like now like learn how to generalize.

we need not like like move deeper into pre training for we like we know the directions we need to go ahead, we think to solve this and we're seeing a lot of both positive transfer and um a lot of just like uh W we're seeing internally the we think the right direction to make this work.

We'll be we'll be safe when or in to me it will be safe when I feel comfortable putting the robot around my kids. I have a one-year-old and

That's another hurdle. It's like getting to general like solving general purposeness, getting safety to work and then making enough of them. Those are kind of like the equations from here. Um we listen, we have a good plan on like what to go do here, but now it's like execution that we gotta go do to show that works.

Most people blame stress, sleep, or just getting older when the energy starts to fade. the brain fog, the slower recovery, the feeling of running on empty by midday. But underneath all of it, there's often a cellular Your body is running low on NAD, the molecule that powers energy production at its core. That's why I've been taking Roe Nutrition's liposomal NAD plus. And the difference is real. Clean, sustained energy, sharper focused, and better recovery without relying on caffeine or stimulus.

NAD supports mitochondrial function, the powerhouses of your cells that keep you energized and your mind sharp. Until recently, replenishing NAD meant expensive clinic visits. But Roe offers a simple at home option just one teaspoon a day. Right now, you can try it risk free with a 60-day money back guarantee and get 20% off with code SRS at rownutrition.com slash discount slash srs. Again, that's SRS for twenty percent off at RowNutrition dot com slash discount slash SRS.

Want more from the Sean Ryan show? Join our Patreon today for more clips and exclusive content. You'll get an exclusive look behind the scenes where you can watch the guests interact with the team and explore the studio before every episode. Plus, additional questions. Our new SRS on-site specials and access to an entire tactical training library you will not find anywhere else. And the best part?

Patreon members can ask our guests questions directly. Your insights can help shape the show. Join us on Patreon now, support the mission, and become part of the Sean Ryan Show's story.

We have better or basically our fifth generation hands on board that have a camera, tactile sensors, basically improved grip. We also have on the robot like basically more compute on board for running our Helix Neural Network. We also spent a lot of time on uh just basically making the robot more safe. So they all have kind of the squishy layer of foam on it. So go ahead.

Clothes can be put on by like uh any person, so we can basically unzip it, um, take clothes off, put clothes back on. We we don't need tools to do so, and uh Uh yeah, basically it's the torso.

Every every year, we're roughly building a new robot every year. We'll just get like dramatically better than this. Damn. Yeah. Our step up from here even to the future robots will be I think perhaps m the most dramatic step up we ever make. Yeah.

The hands are really good. Uh we uh honestly we've working on hands now for like close to four years. It's it's probably one of the hardest engineering problems we have on the on the hardware side. It's probably as hard. And we have our next generation hand that we kind of teased a couple weeks ago that has like basically full I I think it I think it gets a full human level dexterity with this hand.

You know, and then we're gonna still have a lot of trouble getting to hu like full human range of motion. Like small things, like you w reach inside of a um uh you know a washer and you kinda like move your head as you're like getting in or sometimes like some people get on like you know uh get down to the ground and like kind of get in the washer to grab some on the back there's like we do a lot of crazy stuff.

Well, I maybe I'll give you the background. Uh the goal was to like try to advance AI models for humanoid robots together. And um You know, they're they have some like great folks that have worked on like LMs and chatbots and things and um

And at the time we had like a we, you know, we still do, but we had our we had a full like AI team internally. So we were basically working weekly, uh daily on like basically how do we advance uh state-of-the-art kind of like language models for robotics. And um You know like uh Yeah, I I I ended up firing them.

A year a year later, but uh uh in splitting ways. But like listen, th they are a great team. I I like the senior leadership and everybody there, Sam included, like were great to interact with. Um The issue lied for us of like um There's nobody that's ever put like advanced like language models into these systems. And made it we have to like produce out like action output of the robot and it's like a very different thing than like um next token prediction for like language models.

Um we ended up finding that the team we had in place, you know, my my team lead, the folks we have here all from Google DeepMind or certain areas of like, you know, top AI programs, and they're they're really good. The team now we have is over fifty or so on the AI our Helix team internally. We just found that like that team we had internally, uh, we just we just found like kind of circles around them.

And we had a we just had a hard time getting him in the office. We had a hard time like basically like uh like basically you know advancing stuff together as a team. Uh ended up we like the strategy we had internally and the team we had was just like

Complete superstars. They're the best robot learning folks on the planet, they sit a figure. And um it got to a point where uh you know, I got a call one day. It just like, you know, we were like also week to week like showing them how we were doing all this work. And I got a call one day saying like, hey, we're like You know, we've been watching your progress, it's unbelievable. And um, you know, we're thinking about doing robotics work internally.

And I was just like, uh, this is over. Like I yeah, I just get out of here. Like this is like we're tr we're like teaching you how to do like robot learning. You're seeing our progress. We had like a couple of the Sam and a couple of the co-founders on site at one point right before this, and they saw it and they were like, Wow, this is like it was doing like this neural network on table, and they were just like, Jesus, this is amazing.

So we ended up, you know, we're gonna chart our own territory, we're gonna do AI ourselves here. It was also just became like a to be frank, like it became like really hard to recruit. We were like, you know, I I had to spend a lot of my time hiring like like on the AI team and um we'd bring candidates in. And they'd be like, oh, you guys are the robot opening ideas and model.

And I'm like, oh no, not really. No, we like a whole AI team internally. We do model development here ourselves. Um, you know, the we like we're advancing all of ourselves and it just wasn't the perception from the outside. It was just hard. Uh so that also wasn't helpful for us. Um both hiring was not great and we were like, you know, there was like an like information uh passing back that I think wasn't really helpful for us uh long term if we're gonna be competitors.

So uh we decided to split ways. Um I decided physically to split ways. Um, but they have like a great team. I think they're doing robotics now internally. Um

that they were you know they like open AI started out as a robotics program. They were trying to solve AGI through the first three, four years they were just like all in on robots. If you Google like OpenAI robotics, it's like old 2016, 2017, 2018, 2019, like like you know, maybe like yeah, maybe like twenty nineteen, twenty twenty, something like that, they end up pivoting into like large language models, maybe twenty twenty-one, something like this.

But they're in robotics from I think twenty sixteen, twenty seventeen for like many years, maybe three or four years, um, trying to solve like AGI through robotics. There's um You know, there's this other we we don't want to get into it, but like it's it's you know, it's unclear if you need an embodiment or not. Or, you know, at the time it was unclear whether you need an embodiment or not to like truly truly get to like above a peak human intelligence.

Um, and um and they they had a hard time in there, but there was like part of their thesis was like get back into robotics at some point. And I think we just we accelerated that here at figure. Um, and you know, I think to be fair, like to be hum like somewhat humbled, is like it's we made like I think we made like ten

I don't know, five to ten years of progress in like three years. Four years. Like we just like it just felt like uh this should have taken ten. Even right now, it feels like we're not even four years old yet. Full years old and end of May or something like that. Like

I d we like I couldn't believe when we started the company three and a half years ago we'd be at a point where you can get a humanoid robot even here doing the stuff it's doing here, but like let alone like the real stuff it's doing now and like twenty four seven commercial work in the home, like It's neural net driven. Like uh we can make'em every ninety minutes at the you know, when our lines are up. Like it's just like uh it's crazy. So Yeah, we started spar par we started us part ways.

Um, I think there's a huge opportunity like to save lives and help on the military side. Um, but I think it becomes like, you know, we we do have. You know, we do have like a very advanced system here. The system can, you know, unlike a car, if a car became sentient, like You know, you can like w walk in your house, walk upstairs, go in your room. It's like not gonna come chase ya. Like a robot will just walk right up your stairs and open your door.

The humanoid robot. You know, this is a very different technology. We've got to be very careful with it. Um, so I think s because of some of that and some other things, uh, we like we we know we've drawn a line here to say, like, uh, you know, we want to stick with uh, you know.

consumer market, commercial market, and go just harden the paint with that. Um I think there are and will be incredible opportunities for companies like to go in uh into the military to to be frank, is these these robots would be great. Like they can just like they can, you know, like some of the most dangerous missions are like, you know, uh going to close quarters and houses and, you know, um

Yeah, that that stuff is like extremely dangerous. Humanoids would be great at that stuff. Like opening doors and just making sure the house is, you know, cleared, like cle clear house. You know what I mean? It's like some.

We can't get them through without the CEO of like these major companies like coming to see the robots and saying, We're gonna announce this relationship with Figure and we're gonna announce humanoid robots in our facilities. And it's just like a it's a very, you know, it's like a There's you know, it's like if you want to be a little bit more than a little bit.

manifest in the future. So I think that's a real thing. Like I think the the robots can do human-like work. And it will can continue every year to do more and more human-like work. So, but like that, you know, we just gotta we we just wanna be very careful about how we position this and what we do and and also how we communicate it.

There's also a business where if you don't want to solve that, you can definitely ship robots. You could ship them in the commercial workforce, you could ship them in the military, as you mentioned. Um there's like there is there is a path to go like build a business doing that. Um but the biggest business in the world is if you solve like like general purpose robotics. Where uh just through speech and talking to the robot, it'd be it'd feel like you had like a human in a bodysuit.

They can like understand you, nod, like go off and do things now after task. Like that's the problem we want to solve at Figure. That's like a large scale, like it's it's like an AI lab problem at this point. We like we we talk a lot about how we're trying to like we're we're we're like we're trying to give AI a body here at Figure.

And uh so we have this embodiment. We need to put like a really sophisticated AI into it to be able to be able to command it. And that's the that's that's the biggest problem we're trying to solve. If if you're if you're with me in the office every day. I am um working that down with no sleep basically as like as hard as I possibly can. And it's a very, very difficult problem. At this point, it's um it's largely constrained by getting the appropriate data into the into the network.

But when people come to the office and experience it and see the robots and you can talk to them and some of the stuff you're doing here today, it's just like a full like emotional experience that is like really hard to convey. It's it's and

uh more and more. Like uh you feel it when you come to our office and you feel it when you're next to the robots, but it's like hard for the It's we're such early innings about about this yet uh uh for takeoff that it's uh it's hard for the whole world to really feel this.

uh like to sub in. It will like come walk walk over, sit right behind it. And then when the robot is ready and knows that it's there, it will then back away and then the robot will go in to do operations and do work. That other robot will then go over and and and start charging. If any of those robots have any problems throughout, it could be hardware or software, they will go and like uh go to like a basically like the hospital.

in our office. So they'll go to a certain place. When they when they get when they get um when they know they're going to the hospital, we have another robot coming in to the main docks to start subbing in and getting ready to go. All this communication is happening like robot to robot. And uh it's unbelievable. And the robots are getting really robust. We can like um

A year or two ago, we would like there would be like certain motors that you would lose communications with or other types of comms or could be hardware failures or software failures, whatever. Let's say it's the knee. Lose your knee, you like can't. can't stand anymore. You know what I mean? You get like you fall. Um today it doesn't happen. We can lose a knee, we can um hold its pos like we lose full full calms of the knee. We can stiffen the joint and we can limp off.

Holy shit. Yeah. Uh actually I'll I'll post some of the next like week uh publicly about this. Uh I've never it's like holy shit. So we can lose like a lower body motor. And it literally limps off stage, like off like the, you know, the main like uh line it's on.

Headed to the hospital. It'll limp all the way there. While it's limping there, another group from like the healthy part of the hospital will then come in and it's resub it in from the on the dock while the other one undocks while it just lost his knee to go in and do work. All that's happening through robot communication levels. We we you can be like literally asleep while this is happening. We run them twenty four seven. It could be at three in the morning and it will happen.

Uh it's it's it's it's insane. Um this is happening like I saw this in the last like few months. It's happening right now. This is not even like the future stuff. Future stuff is gonna be robots building robots. We're designing robots w we we will have robots building robots here. And then they will go out and they will just do autonomous work. And

they will like charge themselves, they will go do work. You'll speak to them sometimes. Sometimes you won't need to do and they'll just do work and they'll just be like everywhere. Um I I say this again, but I I I think we'll walk out it'll happen first in probably the Bay Area. We're based in the Bay and A lot of companies are in there for robotics. But I think you'll go to the Bay Area at some point and you'll see more humanoids than humans in the next ten years for sure.

Um when I'm over when I'm overseas I so I've like lalked a lot of like the high volume consumer electronics lines and stuff overseas. Some of the most impressive things I've ever seen in my life. It's like you walk these lines and they're just shipping out electronics like crazy. And they have every line they have like this box of automation inside of it, like a little tiny robot inside of there. It's moving some like

whatever, a phone enclosure or something like that. And it's doing it with through an automated way and moving it around a little conveyor and it's moving to the next station. Maybe a human's doing something and it's going down the line, it's going to a next station that's got a robotic system in there.

completely customized and different from what you just saw and they have lines and lines in in floors and floors of this and then buildings and buildings. And you're like, holy shit, each one of those boxes is like a figure style complexity. And they have like hundreds of them. Wow. And they're been they need to run them at a high rate. It's just like, it's unbelievable actually. It's not trivial. It's very complex. And they've been doing it for several decades on these lines.

So I think one is like I don't think that stuff we want to move back. I think we want to move back the the high end robotic stuff that's gonna be like super transformative for us in the future. Yeah, we want to bring back flying cars. I want to bring back like uh like humanoid robots, like the stuff that's like highly dynamic, very intelligent systems, like the next generation, like manufacturing 2.0.

stuff. Gotcha. So we're doing that right now in California on a on our campus. We have a fairly large campus in um in the Bay Area and we manufacture right now like whenever ninety minutes or so. And that we'll we'll continue to spin that up and then we'll put, you know, uh we'll we'll we'll talk about more about it. But we'll we have like we'll put more investment here. into US manufacturing for the future. Right on. So we're we're going to design humanoids here.

So they're they're self looking at each other, self calibrating. They're doing they're doing like burpees and other shit to make sure they like they're okay. If they fail, they go into a triage place. We understand why to fail. Like that shouldn't happen. We should always fix that and it should not fail again. Like how do we fix the manufacturing process so

the next one doesn't come out and ever have that failure. And now we've gotten that process really dialed I mean, dialed in. We still have issues, but like it's fairly dialed in. And so the robots come out, do a couple hour check, and then when they're done, they just walk over. Um And uh at some point we'd love to get like for them to get inside their own box. And another one like get it ready to go and put it on a pallet and we can just start shipping them out.

on this manufacturing conveyor system that like you you grab it, it's literally moving like it's moving because the conveyor is moving down and then the packages are squishing each other. And then the package itself is moving because it's plastic when you're grabbing it. Those are the hard things that are compliant that are like really difficult for robotics.

Because they're not like they're not like stationary when you touch them. So those are things that we're like, man, that's gonna be really tough to fold laundry. And for with code, it's been impossible. The reason you haven't seen like package logistics and stuff, some of the stuff automated is because like these bags are just like hard, they're compliant, they're just tough. You can't model them. Yeah. And now we have like we put it all in a neural net. They basically instantly

When we were working with our we have a logistics customer we're working with. They have like soft packages. And we sign them. They're like, we want you to move these packages on the scenario system. And we've put videos out about it and stuff. And the first month we um signed them. Uh Inga who runs, you know, accounts was like, We need we need to we need to uh we need to do we need to do this for them or they're gonna be brilliant happy. And I was like I was like, damn, that's like uh

compliant material that uh is moving while you're touching, some of them touch and they're the there's something hard inside, some of them are squishy. They're like there's tons of them. We're gonna move every three seconds. We've got to find the barcode, put it down, and put it in the middle of the conveyor. Every three seconds a packet. I was like it was fifty-fifty shots worth.

And it's it's gotta be within neural net. And we got a bunch of data, trained it, policy, and it right away it worked. And I was like, holy shit, this is like it worked really good. And for some reason, the neural nets do extremely well under those like high variability environments that's like extremely diverse. They can learn the representations extremely well across like a wider distribution. And um, they just love it.

folding t shirts, towels, like packages, like no problem. Wow. Stuff that would like You're replanning very fast as you're move as these things are all moving, just doing that in real time. It's just like It just works. Deep deep deep learning just works on on humanoid hardware. Yeah.

It's hard to know who the right stakeholders are, where the decisions are made, or how funding actually moves. Even when you get the right conversation started, you often hear the same responses. We like the technology. But funding is already out. or check back next fiscal year. That leaves a lot of capable teams with strong products but no clear path forward.

That's where my friends at SBIR Advisors come in. They've built a team of over 60 former acquisition officers who spent their careers inside that black box. They help you find the right buyers in the Department of War, find the money, and write winning proposals so you can get on the right contract and fast. Since 2020, they've helped small businesses like yours win over$600 million in government contracts. And they're a 100% veteran team dedicated to one thing.

getting the best technology into the hands of the people who need it the most. The warfighters. If you're serious about selling to the Department of War, go to SBIRadvisors.com. That's Sbiradvisors.com. And if you mention my name, you'll get the first month free.

Let's talk about your venture to save kids in schools. Ready to move on to that? Let's do it. I love this. Yes. Yeah. Yeah. Can you give us the synopsis? Yeah.

Yeah, you know, it's really hard to track, but we went from like thirty to forty events per year in the US to like three hundred. And that was over like a span of ten years. And um it's also really hard to understand why. That's like another thing that we could like spend time on, but it's

There's just like uh you know, a 10X, mostly in the US. You didn't really see this like a lot internationally. And um, you know, we I sort of looking at it looking at it, uh I basically you started reading a bunch of like research reports and other things and I found I stumbled upon

this technology, uh this like basically technology in kind of like terahertz radar. So basically like uh or sometimes also called millimeter wave technology where it's basically a like um Uh basically high frequency like uh like uh it's basically like r it's rad it's radio uh RF, it's like radio frequencies, but basically done at like very high frequency. Uh in the two to three, four hundred gigahertz.

And um it's basically like similar to, you know, when you're in an airport and you go in there and you like hold your hands up and like the LG systems scan you, like a couple feet away, they can see like anything, anything you have. Uh like, you know, if you have knife, gun, vape pen, whatever. Um, I read a research report that showed in my my goal is like if you want to put in schools, you you can't scare the kids.

You have to be able to so sorry, back up. My view in schools is if you want to solve it, you have to solve it from a perception perspective. Meaning you have to see if people have you have to understand if people have guns on them or not. You can like change like there's like a regulation side some people chase, which uh we're not chasing.

Uh, and then there's like a how do we actually like know if people have guns on them? Because if you know a kid has a gun on them, you can go like take it away.

And that is majority of all thing uh all gun events. The ones where you see like uh a planned event that's like on like CNN where somebody's like coming in with a machine gun or automatic weapon, there's a f it happens like one or two times a year. It's on the front page of the news.

The majority of all the cases, like 90 something percent, is all happening from unplanned. Folks are bringing in guns all the time, and then they're shooting it. So you basically what you can do is you can stop all those. The the the planned ones are very difficult and maybe un impossible to stop. But the 90 some percent of all other shootings, you can actually st avo I think you can avoid those, meaning like uh you can prevent them by knowing if somebody has a gun on them.

You can do it the old fashioned way, which is like metal detectors and all this sort of stuff but like it's just like that's not we don't want the kids to go to school like that. That's just like not how I want my kids growing up. Um

So basically uh the reason why I got obsessed with like uh terahertz imaging is you could basically do this at like uh at a at a larger offset, 10, 20, 30 meters away. You can do it at a high frame rate, and you basically get back a point cloud. You basically get back an image. It's like a it's like it's like a three-dimensional camera image almost, but it's done in like a radio frequency. Um it's just you could like look at like almost like an optical image.

And um the reason that's interesting is because like if it's basically people bringing guns in habitually and you can scan them at entrances, you're always coming in through a few doors at a school. You're not going in anywhere anymore. And the schools also have all procedures now for this, like for this stuff.

You basically can do like offset scanning at, you know, five or ten whatever meters away. You can scan people as they're walking in passively. Like just like as like, you know, just walking in, don't need to stop anybody. And you can basically scan You know, most guns that are brought in schools are either in your pocket, waistband, or backpack. It's like most of all guns are being brought in there. And you can basically find'em. And if you know that, you can basic like stop

It was funded by the U.S. government. We worked on it for 10 years. And when the war stopped like funding dropped to zero and we like we were done. We didn't work on And I'm like, oh sucks.

It's like his tarp over this uh machine. Took a tarp off. They had a guy with a manic like a mannequin that's sitting there with a gun underneath his shirt, like I don't know, three or four meters away. It turned this machine on. It was built like ten years ago. It had like a computer tower inside of it. And and then it had like a little screen next to it. So you started this machine.

And they basically moved over to the screen. And the screen showed like as clear as day, like a photo of the you can see the exact guy. You could see it in 3D, you could see it in 2D, you could see it in power. There's a bunch of other ways we can look at the data, but it's just like crystal clear.

I basically made the decision, you know, long story short, I ended up chasing Archer at the time. I went and built Archer. And at the time I only had like a like you know, this was a big endeavor for me, like going from software and like, you know, deep tech hardware. So I basically decided to put cover on hold and um you know chase chase archer. And then about two years ago, somebody came to my office.

One of my investors and was like, hey, I'm like looking at like trying to solve school shootings. I was just back from LA and I'm like trying to solve it with CCTVs, like the security cameras.

He's like the problem is like you can't you won't know until the gun goes off. And you you won't like brandish a gun, you won't pull the gun up until you're like trying to like shoot it. So it's just like way too late. And I told him a story about how I went down this path and here and he kind of like looked me dead in the eyes. I have kids and you have kids, like you you have a fiduciary duty to go build this.

And it was right when my daughter was also applying for first grade, and we were worried about it at schools. You know what I mean? Just looking at like the

And we've been working on this now for two years. I've been self-funding the whole thing. And we will have we have a prototype that already works uh last year and we'll have a full scale prototype out like

I hope by summer, like in our lab. And then we hopefully if all goes well by end of year we're beta testing in the school. Wow. And we'll put him at figure campus first, Stephen. Wow. This is an AI this is like an optical play, uh can you see it? This is an AI play saying can you detect it now? Um there's 130,000 K through 12 schools in the US. There's like 60 or 80 million K through 12 students. It's uh it's huge. And um but it's not just schools, it's stadiums and airports or

But it was just too expensive. The systems we were using were like certain parts on it were like fifty, sixty thousand dollars. So we moved all of that into a into a chip. And we spent last year and a half doing that work. Those chips are in our office now and working. Those chips are like seven dollars instead of fifty thousand dollars.

Yeah. There's a only a few groups in the world that could make them and design them. We co-designed them, we worked on the design with them, made them, fabricated them. And we have them now in our office. They work. We need like a lot, you know, we use many different like we use like a lot of chips, but they're like really cheap.

And um that's important. So we we like, you know, K thelves don't have like a large budget and we need to be able to uh get the cost down to make it affordable for every school.

Uh there's a lot of cash there. Uh the schools also spend a decent amount per student. And I think we get I think we get the cost on a a reasonable amount per student that w that both public and private schools can afford. Um but it's a rat or like r like we we could have already had our systems beta testing in some schools by now, if we didn't pivot a year and a half year ago, we spent the last year trying to like 90% decrease like decrease the bill of materials, like the cost.

Wow. Um th it's just like nuts needed to go big and make this really work well.

You're upgrading to America's number one choice for banking with a Chime checking account. Chime is not just smarter banking, it is the most rewarding way to bank. Join the millions who are already banking fee-free today. Head to Chime dot com slash SRS. That's Chime dot com slash SRS. It only takes a few minutes to sign up. Chime is a fintech, not a bank. Banking services for my pay and chime card provided by Chime's Bank Partners. Optional products and services may have fees or charges.

Dated annual percentage yield and cashback for Chime Prime only. No minimum balance required. Checking account ranking based on a JD power survey published october twentieth, twenty twenty five. For more information on APY rates, my pay spot me and travel perks, go to Chime.com slash disclosures.

Because guns are like they're hidden somewhere. They're concealed. People are not walking in with like handguns and shotguns like into school. They're like they're in like in a waistband, in a pocket, backpack. We can be really thoughtful about if somebody, you know.

clearly doesn't have anything in like uh anything in their pockets when they're walking in, but they have a backpack. We can be thoughtful about like we probably need to scan a backpack. Um so and we maybe need to spend more time uh ex uh uh getting higher frame rate on this area.

Um, and then as you mentioned, like a lot of understanding about like uh does is this person belong here or not? Uh is this like a is this a weird time for somebody to be like w leaving and walking back into the school. So there's just a lot of

semantic grounding we can put into the models to really help like understand if there's threats or not. The schools are set up really well to do like random locker checks now and like uh okay, this doesn't look okay or not. Like the schools are really well equipped for that. It's just like we don't know what's happening. We actually think now that there's probably like

Like perhaps like tens of thousands of guns that are being brought into schools in the US across, you know, 130,000 schools every year. I think we're finding what we're finding now is a very, very small percentage of them are found. That are brought in. And then from there, what we're also finding is actually even a similar small percentage are actually being reported. Because if you report like a student that has a gun. They're going to they're going to juvenile.

what we should do here, you know, for that. That's like that's terrible. Um, but we think there's like we think there's like maybe tens of thousands, maybe hundreds of thousands of guns that are being brought in every year through the schools. Wow. We're finding we're you're reporting thousands. And you're seeing hundreds of shootings. So our view is that that we think it's actually happening like, you know, as a percentage it's low, but as a like a absolute number, it's quite high.

Yeah, so I'm I'm excited about this. We um But it's in some way like I, you know, I I write this prediction every end of every year for like what you know will happen in the spaces I'm in, which is like flying cars, like robotics, like AI and weapon detection and stuff like that. And like uh I did a post. In December, like here's what I think on these four areas. And like overwhelmingly, like the most support I got, like publicly, um, was just for cover.

There it just like um I think it I think it you know, it just like hits I think it hits in a really good way with a lot of folks, maybe especially parents. So um

like feeling like we were secure before, but we lost it now. And now we can definitely see it and feel it. And it's like we just never we're never gonna be able to get back. Yeah. At the at the and I've had like, you know, I've had like some close people I know that have been involved around this stuff. And it's just like, um, it's terrible. And

So you know, my my agenda here is I I think it can be prevented. I don't know if you're gonna prevent all of them. I think you can prevent a lot of them. And then if and then if you even have like there's no real security there at all right now, but even if you have security, there's also like a sense of like

Shit, I gotta bring a gun in here now. There's like real sophisticated AI that's in all these schools that can catch it. Yeah. I think that's another big thing. Uh you have that at like TSA when you go to preach it. Yeah, it's a deterrent. So you have like that, but we can also find you find it. We can see underneath uh through backpacks and stuff. It happens at specialized um like uh like radio frequencies.

Uh it happens at like, you know, two hundred to three hundred gigahertz, then it happens again at six hundred gigahertz. And in between those bands, there's either F C C rules that prevent you from doing it or there's ap atmospheric attenuation, meaning um Sometimes uh there's enough moisture in the atmosphere at certain radio frequencies that like the the radio frequencies don't do well and perform well. Um they perform well at these certain uh radio frequencies for the imaging stuff we do.

So it's actually quite a hard technical feat. I one of the reasons I didn't do it and did Archer is because I thought the cover stuff was actually harder than doing

They're doing work in weather and space and they're not doing this for like shootings and like security and stuff. There's a there is no industry for this. And luckily we have like the world's best terahertz experts. at cover that are working every day on this and they're really passionate

Uh they're probably not getting paid enough and they're just like super passionate about solving this problem. And um so anyway, I think um I think the through line for covers, I think it'll work. I think we'll be able to demonstrate it hopefully by end of this year. Like we'll be able to say like we have it at like at we'll have it at figure campus first.

And then we'll put it in uh in like schools, like hopefully on the West Coast and maybe a maybe one or two and we'll see how it goes. There's um, you know, there's a m there m how do we market this? Like what do we tell the students, like teach like parents? There's like, you know, there's a lot of stuff here we need to get right. But if it go it goes well from there. And we're getting low false positives. Like what we really wanna do is make sure we don't freak the kids out. We don't wanna

you know, think it's a gun but it's a crayon box. That'd be terrible. So like that that's a really an AI problem. So we basically want to make sure like uh we have like low false positives around the whole sys stack. Uh that's a really hard problem to solve, especially for us, where you can be like partially occluded on certain areas.

uh of the person or the weapon and then we need to like to know what we see. It's actually is is it w is it real or not? And so funny enough, if you come to our lab, we have like a just guns everywhere. And um they're all they're all bricked. You can't like actually shoot'em. But um but we we all we all day we we try to figure out how to put guns on humans or mannequins and we try to figure out how to detect'em.

In an order of like twenty X or so, like the the radio frequency level, it's like operating in a few gigahertz, something like that. But we want we operate at much higher frequencies, like three hundred gigahertz. Um, so you want to like whatever, call it 50, maybe it's 50, 100 times this. And um, and then basically it shoots this out uh and this waveform comes back.

And we we review it and we look how long uh like how long it took to come back. Um and we use beamforming a couple of other techniques to figure out um what happened. But you're basically it's same as like traditional radar technology. Um but we can shoot it out, comes back, it's not ionizing, it won't hurt you. Like it's perfectly fine to be around like your Wi-Fi. And we can basically get uh we can get both um a 2D image, what's happening, and a 3D point cloud.

The 3D point cloud is what's really important. So if you have like a weapon on you, like in your uh pocket for whatever, or say I have one on like my chest, for example. We will start getting back the signals back from the top surface of the gun before we get your chest certs, chest stuff back.

And in the case of your chest, you have a lot of like water in your skin, and it'll it'll somewhat attenuate in your chest. So then we'll get back an image from this and we'll re- reconstruct it really fast.

And we can reconstruct into like somewhat of a three-dimensional point cloud like you do a camera. So you can basically get like almost like it look like like an op like I showed you earlier today the the op the the vision from this. It looks like a kind of a camera image is what you get back. And um so from there you can kind of like visually see what's really happening. In the case of the gun, you can like you can see the gun. You can see the trigger in some cases.

Yeah, and sometimes it might be just be like the handle or the um a side of a gun or different places of it, but you can see it through materials. Like it could be backpacks, it could be clothing or a jacket. Um but most guns are all in the waistband pockets or backpacks.

Which makes sense, right? You're not like wearing it around your neck on a outside your shirt or um like things like this. So uh that's where that's where like most weapons are entering the school. We've done a lot of we have uh probably one of the best like data scientists in the world that Is obsessed with school shooting and he puts out the best school shooting analytics. He does it daily. He's done it for five years.

He's he's working with us through this and we've done so much work on how people how how students enter schools, how they exit, emergency responses, what solutions are on campus now for this. uh like uh where what what data uh we find on like where guns are at, what type of guns and weapons are there. There's um I think it was like 200 nice stabbings last year. It's like it's so high and so dangerous, like we're trying to like we can detect night. Like there's um

You know, vape pins, whatever whatever. It's not a metallic thing. It's like we can, it just doesn't matter what the object it looks like. Different metallics will actually like um uh like come back to the radar system a little bit differently. So you can kind of maybe sometimes tell if there's like a like a gun, like a metallic signature or not coming from the material.

But the technology is like really kind of straightforward in the sense of like it's RF technology, like radio frequency technology. And you get back like an image. And we can use that image to train like to build like a neural network to then look at it and say like what is this thing? What time of day is it? Who is this human? Like is this a dangerous threat or not?

And we need to do a really good job of making sure we're accurate in those readings or not. Uh if we're not, we're gonna cause havoc. And we're for right a lot of times we could basically start saving lives. And that's I mean, there's there's a on average a one shooting every every single day in the in like more than that. Like

There's you know, there's over three uh rough three hundred or more or so shootings roughly a year if you look back the last couple of years. So like every single day, I mean there's less school days in a year than three hundred and sixty five, but like roughly every day is a school shooting in the U. Um that's just at K through twelve, not colleges. Uh but that's about that's looking at the hundred and thirty thousand K through twelve schools in the US.

you know, several meters wide, multiple doors, sometimes double doors to get in. Like we need to scan all of that all the way through. So it's like a natural aperture that students are walking into, which is good. You're not walking inside of a building, you know, like through a brick wall. You have to walk into a door entrance. And we're trying to, yeah, basically we're trying to uh get that fully complete this year.

that I've been my my team and I have been working through day and night for the last four years. So it's like, okay, we want to go like build like a crazy sci-fi future with the with like flying cars, AI humanoids. And then on my other half of my life, I'm like using like an AI chatbot. Like a frontier lab, like Jim and I or ChatGPT. And it's so stupid.

It doesn't know me at all, doesn't remember anything I'm saying, can't see what I'm doing. It can't use tools very well, uses the internet like really poorly, can't even order me a sandwich if I needed one right now. Um and like uh It doesn't feel very futuristic. It felt futuristic three years ago, but now anymore it's just like it's just not very good.

It feels like I'm like in an incognito window searching Google. That's all I can do. Doesn't have access to my accounts, doesn't know any of this stuff. Meanwhile, I think like for me, like I I was just been sitting here for three years thinking like, we're we're gonna get like Jarvis out of this from Iron Man. we're gonna get something crazy out of AI. It's gonna move to a point where it can like it can like listen and speak.

Naturally like a human, it can see the world, it can do uh it can use tools like a browser and terminal, it can do real work for you and help you out. It'll know you really well and a shot, it'll know everything you're ever doing, all your stuff, and be really personal to you. We don't have anything like that now. I got like this stupid chat by that doesn't remember the last thing I said to it. And So I decided to like I I decided like there's two things here that are extremely broken.

One on the AI side, we have like extreme we have like we have like we have like a lot of gaps to get to to get to like um like like uh extremely personalized like AI intelligence. There's just a lot of there's a lot of r there's a lot of like like missed opp like missed opportunity now last like few years a lot a lot of like um a lot a lot of gaps there. The second thing is we're like interacting with

these AI systems feel like old pre-AI computers. Like you're putting up your phone or your Mac or your c your computer. It's like they're all designed like 20 years ago. So they're like the it's like a really old interface. The chatbot's an old interface. It's the wrong interface to AGI. You're not going to get the Jarvis with those. So we have to go rebuild uh all the hardware from scratch.

Act as a coach and say like, Hey, you said you do this over ninety days and you're not doing this over here. It'll just like it'll Yeah. Yeah. Yeah, we have we've been We have hardware in our lab, we have we work on AI models now, like stuff is crazy cool. And um We're yeah, we're gonna uh We're gonna I think we'll probably come out of stealth by the time this thing airs here between you and me.

you know, doing the annual like college trip with like my friends and stuff. Like I don't do that anymore. I spend I have like my family and I have my companies. And that's all I do. And um And I rarely I do like a few podcasts a year, not a much. I'm excited to come here'cause like I'd love your show and get the story out too. You're great at it. And um

So I you know, I just like protect my time and just like I go all in on these things, like my you know, my kids and I have my work kids, you know what I mean? Like and so like, um which are like, you know, these are my like like they're like kinda like babies. I go, you know, make'em get constant care and attention. Uh so I have like this family and I need to like uh I go all in on it and I do everything else less good. You know what I mean? I'm like a shitty

college friend if you like, you know what I mean? If I haven't seen you in a little while, like just like not gonna spend the half a day with you on Saturday if you're in town and I haven't seen you in ten years. So which is unfortunate. I wish but like, you know, I care about these things more. I care about doing this stuff really well. And I I'm really happy at it. Uh, you know, I'm happy with family, happy with like things going to work. And, you know, to be frank, I just I

I was born and raised on a farm, man. And I get to do like work on this cool shit every day. And um and you know, I I got billions behind it, make like going for it. Great teams that work like their asses off. Like teams that are, you know, here came with. And uh it's great. And I like um fired up to come every day to work to try to make this thing happen and I hope these things all work. It's just but like uh I don't know, these are also hard businesses, so

The stuff I'm working on, I feel like is very important for the world, which is also great. I didn't, you know, you and vetery is like there was a part of me saying, like, okay, is this like the the thing I want to spend my whole life doing. And uh I have that here, which is great. These are like the things I wanna spend all my time on for the next like twenty, thirty, forty years. Um, so it's good. I'm just like, uh I just don't wanna don't wanna screw it up now, you know? Make them work.

All right, we're wrapping up the interview. I got a hot question to ask you. You ready? Let's do it. For decades, movies taught us to fear robots becoming self-aware and turning on people. But in the real world, we still don't have public evidence of conscious machines. What we do have are real cases of robots harming people from Robert Williams being killed by a Ford industrial robot nineteen seventy nine to the viral twenty twenty-five

Unitry H1 malfunction that showed how violently a humanoid system can lose control. Plus, long standing research warnings. That robots and homes can create privacy and security vulnerab vulnerabilities in ongoing global debate over autonomous weapons. So is the bigger threat not conscious machines at all, but obedient machines that can still malfunction, be hacked, surveilled through remotely controlled or turned into tools of intimidation, assassination, or state power?

Are we gonna have like, you know, bumps along the way, like for sure? Uh are they gonna, you know, hurt somebody at some point? Like I think that's bound bound to happen at some point with enough scale. Um, but I think like the this this the this the spirit here for humanity to get this to get this done. I I think is here and I think it's gonna be one of the most important technologies of our lifetime. Like I think in

Someway this AI stuff of like we're like we're generating AI systems that can be embodied and can use computers, like it's gonna be like one of the most transformative technologies we've ever been through. Like we we're building synthetic humans at scale.

It's it's both scary, but also like very um. I'm like very excited about that future. Um So I think my view here is um yes, there's like a lot of uh like really difficult things that could go wrong that perhaps could like maybe will go wrong, but I think uh we need

Uh, just like we need cars. And I think uh just like we need like um, you know, uh a lot of things in life. Um, airplanes and things. I think these are like important technologies that really move society f few forward. Um So anyway, I uh I happen to believe that this is like uh e extremely important, will save lives and like I think increase prosperity across like all of human civilization and I think uh

I'm excited to be working on it, but I think there is a lot of truth to what you know, like I said, it's gonna be a really hard road.

There's like different paths to go down uh from here that could be good or bad. Um, I think my bet's on high probability of really great. Um, there's obviously always path that could like uh not go well, but like being conscious of that. And like basically doing everything possible to steer it in the right direction is like what we like what we have to do at this point. Like this is not like something we can turn off.

you know, turn off the internet. Yeah. You're gonna stop people from trying to build like systems that like make us more productive and do more work. I don't think that's it's not happening. So like um all we can do is basically do it the right way that has the best positive effect on the

too about advice. Certain people. And i i I think that's a big fear of a lot of a lot of folks too. It's already happening with Chat GPT and all these other cloud and all these other things anyways, but Who are they getting for vice from before that? Probably I mean you know what I mean? It's it's I think it's the caliber of person.

Uh just go learn. I think um there's never been a never been a situation where I haven't like done something and then learned a bunch and then uh haven't reset from that feedback. So almost like um like little stairs I'm climbing over and over throughout time. And um so if I just wouldn't have started and wouldn't have moved, like I wouldn't have learned this information.

Um, so it's like a lot of information coming in recursively self-improving and getting better over time. This could be simple things like hiring and doing accounting or running an engineering team or like trying to ship a product or getting feedback from customers. Like, I'm just getting I I think I'm getting it's like a s it's like a it's like a sports player. You're getting better with more practice. And so I think the most important thing is just like just go. I also think um

The thing I learned a lot in my lifetime is like what you work on is really a defining moment for like for founders. And it could be founders of any in any industry, tech, non tech or whatever. You're generally gonna go and just try to like have this like have this kit.

that needs a lot of attention. And then at some point it's like you just can't abandon this thing. And you got to keep like spending more time with it. And it needs a lot. And it's like constantly working on the problems with it. So it's like not the fun things. You're working on all the hard things. It's like this problem funnel I have where I'm working on the hardest, most pernicious problems at the company. So you gotta like really love it.

And it's not like you can be there for a year or two. You have to be there for sometimes a really long time. And even if you're successful, and even if you sell your company or whatever, go public, you're getting your stock locked up or you're you're vesting out over a minute period of time. It's you're gotta be in it for quite a while. And I find that for me, the things I work on as like probably the most important things I could be doing with my decision making.

And that's happening at a micro level inside the companies with what I work on week to week, month to month, but it's happening at a macro level where like where do I spend my time? Like I'm 39 right now. Like where do I spend my time as 39-year-old Brad? And where does like 20-year-old Brad and 25-year-old Brad spend my time as an entrepreneur? And I generally have this philosophy that harder things are easier. Like meaning um there's uh there's like a nonlinear effect here.

For like uh starting companies that are like easier versus harder. Meaning um starting something that could be like a hundred times higher outcome is generally not a hundred times harder. Um, so like doing figure is not a hundred times harder than doing another robot company. It's probably like

three times harder, maybe five times harder. But the total reversible market, the opportunity is probably millions of times bigger than another like robot that's like on an assembly line moving back and forth. And so I think there's like this nonlinear effect to like decision making here that is really important where harder things that have like larger outcomes are like usually easier to recruit the best talent in the world. That gives you a better lift to build a better product.

and a better team. That team and better product and maybe even a bigger industry because it's harder will give you like more capital coming at you for disposal. uh to be able to like make the right investments you need uh into the right say equipment or people or personnel or whatever marketing to basically make you more successful. And then you're generally worth like working inside of bigger dressable markets like TAMS that

you know, potential acquirers or public markets or other folks like really want to see and have like basically a disproportionate outcome. They wanna they want a high risk reward. They want to You know, investors and things and even people, they want to like go in and like if it works, they want to like a hundred X or a thousand X. They don't like a two X.

And generally for venture, like 90, 95% of people fail. So if it works, you really want to go, you want to hit a grand slam. So I think my philosophy is like um spin like choose wisely what like hey young brat spin choose wisely what you work on young entrepreneurs and um and then I would try to be as ambitious as possible.

Um, there's capital for that and there's there's humans for that that wanna work at really crazy shit. Uh, we have'em at my companies and they're they're they're incredible. You met some of'em today. They're just like you know, my design lead and

uh uh a bunch of other folks here that are just unbelievable at what they do. They're the best in the world at what they do. But they want to come here and they want to try to do something like that's never been done before. They don't want to go off and design the next

car like you know or do the next AI product everybody else is doing. They want to be here signing in something revolutionary. So I think that's like um something don't stress enough. And I think the last thing is like there's no rule book for this. Which is like really unfair. And there's There's a lot of people out there that will teach like here's what to do. And they're generally coming from folks that haven't haven't done it before. And the signal of noise out there is just so high.

Or so low. Like I mean you get a lot of noise out of like in the market. It's very noisy about like what to do and what's what successful means for building a team or hiring engineers or like uh executing a product. It's very difficult. Um And very few people in the world know how to do it really well consistently. And so I found over time it's been really hard for me to get the right advice. And um So I think it's been a lonely path. And for folks out there that are on that path.

It's lonely, but I believe in you. And you can do it. And I think that's um I've never had somebody for 20 years I could call and just like, what should I do in this situation? I never have had it. And uh I wish I had. Uh there's no book. There's nobody to call.

You are. And it's like it's the great filter. It's the folks that you know. You went through butts. Like it's the great filter. It's 95% of everybody will fail, and you'll devote your life into it and time and maybe all your money and your brand and you'll be embarrassed and you'll you'll fail. Most people will fail. And it's uh it's only for the folks that will like, I will like, you know, I will do whatever it takes to go make sure I make this happen. There is no failure.

Those are the folks that do well here. And you can bend the world and uh you basically can mold the the future to how you kind of want to if you try hard enough. And the the end of the goal, the goal at the end of the day is just to not die. Yeah. you won't die. So like uh anyway, I think um

I think it's uh a listen, been playing this now for twenty years, still playing it. I feel like I'm in the early innings of my career now. I wanna go ship at scale these systems. I haven't done that yet. We're like an inning we're bred inning one. And so like uh but for everybody out there that's in that, I just think it's um I believe in you here. Yeah.