Latest / What's Your Problem? / Teaching AI to Build Stuff in the Physical World

Transcript

Teaching AI to Build Stuff in the Physical World · What's Your Problem?

• 0:15
  Speaker 1: Pushkin.
• 0:20
  Speaker 2: AI works amazingly well. It works terrifyingly well even for
• 0:26
  Speaker 2: virtual things, for words, for pictures, for videos. This is
• 0:32
  Speaker 2: true in large part because of the Internet. The Internet
• 0:36
  Speaker 2: provides this wildly abundant, readily available source of words, pictures,
• 0:41
  Speaker 2: and videos to train AI models. But there is no analogous,
• 0:48
  Speaker 2: wildly abundant, readily available data set for the physical world.
• 0:53
  Speaker 2: There is no gargantuan Internet like repository of data that
• 0:58
  Speaker 2: describes how things move and bend and break in real
• 1:02
  Speaker 2: physical space. And as a result, we do not yet
• 1:07
  Speaker 2: have robust AI for the physical But people are working
• 1:12
  Speaker 2: on it, and if they succeed, they'll change the way
• 1:16
  Speaker 2: the world works, not just the world as it appears
• 1:19
  Speaker 2: on our screens, but the actual physical world, the world
• 1:23
  Speaker 2: where if you drop something on your foot it hurts.
• 1:33
  Speaker 2: I'm Jacob Goldstein and this is What's Your Problem, the
• 1:35
  Speaker 2: show where I talk to people who are trying to
• 1:38
  Speaker 2: make technological progress. My guest today is Edward Mayer. He's
• 1:43
  Speaker 2: the co founder and CEO of Machina Labs. Edward's problem
• 1:47
  Speaker 2: is this, how can you use AI to turn robots
• 1:51
  Speaker 2: from dumb, inflexible machines into skilled versatile craftsmen. Before he
• 1:58
  Speaker 2: started Machina Labs, Edward worked in the rocket ship business,
• 2:02
  Speaker 2: first at SpaceX and then at a company called Relativity Space.
• 2:07
  Speaker 2: And in the rocket business, the word solw firsthand the
• 2:10
  Speaker 2: problems of traditional manufacturing. It's the kind of problem he's
• 2:14
  Speaker 2: now trying to solve with AI and robots. It's a
• 2:17
  Speaker 2: problem called the rigid factory problem. So I've heard you
• 2:24
  Speaker 2: use this phrase that's interesting to me, and it's the
• 2:27
  Speaker 2: rigid factory problem. What's the rigid factory problem?
• 2:32
  Speaker 1: That main problem with the factories today is that rigidity,
• 2:36
  Speaker 1: meaning that if you have to build a physical product,
• 2:40
  Speaker 1: you pretty much have to build a factory that's designed
• 2:42
  Speaker 1: for it and built for it. There's a lot of
• 2:45
  Speaker 1: components that goes into the factory, from machinery all the
• 2:49
  Speaker 1: way to the tooling that is required to build products
• 2:52
  Speaker 1: that are specifically designed for the geometry for the material
• 2:56
  Speaker 1: that you're trying to use. The moment you want to
• 2:58
  Speaker 1: change that, you have to change your factory, which is
• 3:00
  Speaker 1: a huge investment. You know, I always give an example
• 3:03
  Speaker 1: from when I was at SpaceX. You know, you think
• 3:06
  Speaker 1: of SpaceX as a very innovative and it is you
• 3:10
  Speaker 1: know on the edge of a hardware space in terms
• 3:12
  Speaker 1: of innovation in the past twenty four years twenty three
• 3:17
  Speaker 1: four years that they have existed, they have two rocket families.
• 3:21
  Speaker 1: There's Starship and there's Falcon, right, because at the moment
• 3:24
  Speaker 1: you decide on diameter of for example, Falcon nine or
• 3:27
  Speaker 1: the Falcon family in general, the diameter of that core,
• 3:29
  Speaker 1: it's very hard to change it. A lot of tooling
• 3:31
  Speaker 1: and machinery specifically built for that diameter. And that's why
• 3:35
  Speaker 1: for Starship they had to start from scratch.
• 3:37
  Speaker 2: Start from scratch, meaning like not just design, but like
• 3:40
  Speaker 2: the factory itself, like the Factorily had to build a
• 3:43
  Speaker 2: whole new factory because they wanted to make a different
• 3:46
  Speaker 2: sized rocket.
• 3:47
  Speaker 1: Yes, different size, different material, All the tooling has to change,
• 3:51
  Speaker 1: right almost almost, Yeah, you have to basically assume building
• 3:54
  Speaker 1: from scratch, ground up factory. Why does it need to
• 3:57
  Speaker 1: be there for us to build this new product?
• 4:00
  Speaker 2: I heard you describe was this from your own experience
• 4:03
  Speaker 2: the sort of era at SpaceX when the fact that
• 4:08
  Speaker 2: you couldn't make the rocket wider led to all these
• 4:12
  Speaker 2: kind of difficult things people were trying to do to
• 4:16
  Speaker 2: be like, how can we do all these things under
• 4:18
  Speaker 2: this fundamental constraint, Like, can you talk a little bit
• 4:20
  Speaker 2: about that.
• 4:21
  Speaker 1: Yeah, this is a lot of conversation happening in twenty twelve,
• 4:25
  Speaker 1: twenty thirteen, twenty fourteen time when the diameter of the
• 4:30
  Speaker 1: Falcon nine could not get any larger, and if you
• 4:33
  Speaker 1: look at actually different Falcon versions, the height of that
• 4:36
  Speaker 1: vehicle kept going higher, the diameter could not change. So
• 4:40
  Speaker 1: it was about where what space can you find to
• 4:42
  Speaker 1: put new features and new designs that exist within the vehicle.
• 4:49
  Speaker 1: So there was a lot of stuff basically being crammed
• 4:51
  Speaker 1: into the space that you have already got.
• 4:53
  Speaker 2: So that's true for building rockets. I mean, what are
• 4:55
  Speaker 2: some other just you know, different kinds of manufactured products
• 4:59
  Speaker 2: where that kind of rigidity is a problem.
• 5:02
  Speaker 1: Yeah, I think it is just common almost in all manufacturing.
• 5:06
  Speaker 1: That's why this phenomenon. I think it's kind of funny.
• 5:08
  Speaker 1: People take it for granted that a thing called economies
• 5:12
  Speaker 1: of scale, uh huh, Like people take it for granted
• 5:16
  Speaker 1: as if it's rule of nature. It's actually not.
• 5:19
  Speaker 2: Just to be clear, it's basically, the more you build
• 5:21
  Speaker 2: of a thing, the cheaper each one of those things gets.
• 5:23
  Speaker 2: If you build one, it's really expensive. If you build
• 5:25
  Speaker 2: a million each one's.
• 5:27
  Speaker 1: A lot cheaper, yes, exactly, But then people don't think
• 5:29
  Speaker 1: about it's like, oh, okay, intuitively makes sense, but why
• 5:32
  Speaker 1: it's actually not that intuitive. It's actually a limitation of technology. Right.
• 5:38
  Speaker 1: Why why economies of scale is a thing is because
• 5:41
  Speaker 1: you have to make a huge amount of investment to
• 5:43
  Speaker 1: make the first thing, and the moment you make the
• 5:46
  Speaker 1: second thing and the third thing, then you can break
• 5:48
  Speaker 1: even your investment onto more products that you're going to
• 5:52
  Speaker 1: come out of it. But that's only true if the
• 5:54
  Speaker 1: second product can be built for the first investment. You
• 5:57
  Speaker 1: had to turn this concept and say, oh, this is
• 5:59
  Speaker 1: a given. This is an axiom of the world that
• 6:02
  Speaker 1: economies of scale is a thing, but in reality it
• 6:04
  Speaker 1: is a technological challenge. Right. It means that you build
• 6:08
  Speaker 1: a car, you ask what application you Once you build
• 6:10
  Speaker 1: a factory for a car and all the toolings and
• 6:14
  Speaker 1: dies that goes into stamping of the panels of that car,
• 6:17
  Speaker 1: that's one hundred and fifty million dollar investment just for stamping.
• 6:21
  Speaker 1: And this is a number for example from Tesla. Tesla
• 6:23
  Speaker 1: spens one hundred and fifty million dollars in a stamping
• 6:25
  Speaker 1: plant they have in Giga factory in Texas, right, and
• 6:29
  Speaker 1: that can only make Model Y or Model three right
• 6:33
  Speaker 1: the moment you have to change that. That means go
• 6:36
  Speaker 1: through every eighty two hundred and thirty sheet metal panels
• 6:40
  Speaker 1: that exist on that car and design a new tool
• 6:42
  Speaker 1: for it. And each of these tool is going to
• 6:44
  Speaker 1: be a few hundred thousand dollars to sometimes a million
• 6:47
  Speaker 1: dollars or a million and a half million dollars. And
• 6:49
  Speaker 1: you're talking about like eighty t one hundred and thirty
• 6:51
  Speaker 1: tools per vehicle.
• 6:53
  Speaker 2: And like all you're doing you're not reinventing the car there,
• 6:55
  Speaker 2: you're just making a car that's a slightly different shape.
• 6:58
  Speaker 1: Basically, yes, maybe you may get sedan, and you're not
• 7:02
  Speaker 1: trying to do a slightly longer version, a slightly bigger version.
• 7:06
  Speaker 1: And that's why economies a scale of the thing you saying, Okay,
• 7:08
  Speaker 1: I made a fact. Now it only pays back if
• 7:11
  Speaker 1: I make a million of this car, right, because I
• 7:14
  Speaker 1: had to just drop one hundred and fifty million dollars
• 7:16
  Speaker 1: on just a stamping plant. So yeah, it's all over manufacturing.
• 7:20
  Speaker 1: We abstract this whole concept and gave it the name,
• 7:22
  Speaker 1: says economies, of scale.
• 7:24
  Speaker 2: Yeah, so you left SpaceX and you went to Relativity Space, right,
• 7:30
  Speaker 2: a company that was also in the space business that
• 7:33
  Speaker 2: was using three D printing.
• 7:34
  Speaker 1: Right.
• 7:35
  Speaker 2: That was the idea of the company, which seems like
• 7:37
  Speaker 2: an approach to this problem that you're talking about.
• 7:40
  Speaker 1: Right.
• 7:40
  Speaker 2: An advantage of three D printing is that it is
• 7:44
  Speaker 2: much more flexible and less rigid than traditional manufacturing. Right,
• 7:48
  Speaker 2: So tell me about that.
• 7:50
  Speaker 1: Yeah. So, yeah, we saw this challenge at SpaceX and
• 7:54
  Speaker 1: I joined Relativity very early on. I was the fourth
• 7:56
  Speaker 1: person on that team. And the goal over there was, Okay,
• 8:02
  Speaker 1: let's just think about this fundamentally, can we build a
• 8:05
  Speaker 1: rocket that all built with flexible technology and a time?
• 8:09
  Speaker 1: Three D printing was that forefront of everybody's minds because
• 8:14
  Speaker 1: people were already starting to build that. NASA SpaceX people
• 8:18
  Speaker 1: were already starting to build engines out of three D printing,
• 8:21
  Speaker 1: and the concept was like, well, that's great, it's very flexible.
• 8:24
  Speaker 1: Three D printing has this promise of geometry agnostic, material agnostic.
• 8:27
  Speaker 1: You can just feed it a design and can build
• 8:29
  Speaker 1: a product for you. And it worked very well with
• 8:31
  Speaker 1: rocket engines. I think probably the future all rocket engines
• 8:36
  Speaker 1: would be would be three D printed, And the concept was,
• 8:39
  Speaker 1: can we take this and scale it to the whole vehicle, right,
• 8:44
  Speaker 1: can we build the whole vehicle with a process like
• 8:46
  Speaker 1: three D printing so that it is flexible Today, if
• 8:50
  Speaker 1: you want to build a rocket with twelve foot diameter,
• 8:53
  Speaker 1: we can do it. And then if our calculation changes
• 8:55
  Speaker 1: and we wanted to go to another orbit or do
• 8:57
  Speaker 1: a different type of emission, then we can change that
• 9:00
  Speaker 1: diet twelve diameter to twenty diameter twenty fie.
• 9:03
  Speaker 2: Don't have to build a new factory, don't have to
• 9:04
  Speaker 2: build new machinery, just change three D printer.
• 9:07
  Speaker 1: Yeah, exactly. So that was a concept behind relativity. That's
• 9:12
  Speaker 1: a thesis behind relativity, and that was the goal there.
• 9:15
  Speaker 1: The goal was, you know, three D print a whole
• 9:17
  Speaker 1: rocket so they can be flexible.
• 9:19
  Speaker 2: But it hasn't. It hasn't worked at least in the
• 9:22
  Speaker 2: kind of maximalist version, right, Like, they just haven't been
• 9:25
  Speaker 2: able to do it. They've they've sort of backed off
• 9:27
  Speaker 2: of that that big dream, as I understand it.
• 9:30
  Speaker 1: Yeah, yeah, So I think the challenge was that three
• 9:36
  Speaker 1: D printing is just one process and it's necessarily not
• 9:39
  Speaker 1: good for every type of part. You know, manufacturing is
• 9:44
  Speaker 1: very versatile. You do different types of geometries, different types
• 9:47
  Speaker 1: of material, and three D printing has a very small reach.
• 9:50
  Speaker 1: There's certain type of parts like rocket engines, very good fit.
• 9:54
  Speaker 1: You're building a tank, maybe not so right. So yeah
• 9:58
  Speaker 1: it's good for certain type of parts, but there as a
• 10:00
  Speaker 1: whole lot of other parts. Like I said, you know
• 10:01
  Speaker 1: you're building a fuel tank, which is basically large sheet
• 10:05
  Speaker 1: metal or thin walled structure, then maybe three D printing
• 10:09
  Speaker 1: is not as good as a fit because it takes
• 10:11
  Speaker 1: a long time, and also because it's thin, you have
• 10:14
  Speaker 1: a lot of physical challenges in terms of controlling the
• 10:16
  Speaker 1: geometry and the tolerances. So we realize soon that maybe
• 10:23
  Speaker 1: other processes are also need to be automated the same
• 10:27
  Speaker 1: way three D printing is. We need to have more
• 10:30
  Speaker 1: flexible processes that are not just one process, more flexible platforms.
• 10:34
  Speaker 1: They can do different types of processes, not just three
• 10:36
  Speaker 1: D printing, to be able to cover a whole variety
• 10:41
  Speaker 1: of products in a flexible manner, the same way the
• 10:44
  Speaker 1: three D printing that's for certain type of products. And
• 10:47
  Speaker 1: that was actually the thinking behind MARKETA Labs is that, okay,
• 10:50
  Speaker 1: can we step back and say, what do we need
• 10:53
  Speaker 1: to build? What is this flexible platform that can do
• 10:57
  Speaker 1: three D printing if needed, or it can do sheet
• 10:59
  Speaker 1: forming if it's needed, It can do machining if it's needed,
• 11:02
  Speaker 1: but chooses the right operation, right flexible operation for the
• 11:06
  Speaker 1: right part, but still very agile and doesn't require a
• 11:09
  Speaker 1: lot of tooling and it's not inflexible.
• 11:12
  Speaker 2: So it's it's sort of zooming out more. It's saying
• 11:15
  Speaker 2: three D printing is not going to do everything the
• 11:18
  Speaker 2: way manufacturing works now. It's just too rigid, too hard
• 11:21
  Speaker 2: to change things, to rely on on scale to make
• 11:24
  Speaker 2: the economics work out. So like that's a very big,
• 11:28
  Speaker 2: very abstract thought. To start a company, you got to
• 11:33
  Speaker 2: make something or you got to make something that makes
• 11:36
  Speaker 2: something like what do you what do you actually do?
• 11:39
  Speaker 1: Yeah? So it was interesting, right, you know we actually
• 11:44
  Speaker 1: the solution was in our past. Right if you look
• 11:47
  Speaker 1: at like.
• 11:48
  Speaker 2: The lesson in a movie, it's like the Wizard of
• 11:50
  Speaker 2: Oz or something exactly.
• 11:53
  Speaker 1: If you look at manufacturing, I mean up to Industrial Revolution,
• 11:56
  Speaker 1: it was arts and crafts. Right, it was basically humans
• 12:01
  Speaker 1: trying to figure out how to conquer nature, right, Like,
• 12:03
  Speaker 1: how am I gonna use my hands? On my brains
• 12:06
  Speaker 1: and very few primitive tools to deform a product or
• 12:11
  Speaker 1: shape a product from raw material. Right and to this state,
• 12:16
  Speaker 1: if you are in a very high mixed manufacturing still
• 12:19
  Speaker 1: a lot of that creativity exists. There is a person
• 12:22
  Speaker 1: at Space IX. His name is Big John. I don't
• 12:25
  Speaker 1: think he's there anymore, but there was this guy. It
• 12:28
  Speaker 1: was like, you know, a very skilled maker, a craftsman.
• 12:32
  Speaker 1: You could figure out how to use simpler tools to
• 12:35
  Speaker 1: build different things in a creative way. Maybe it's not
• 12:37
  Speaker 1: a repeatable way like you know a stamping works or
• 12:41
  Speaker 1: ejection molding works, but you can be flexible. You can
• 12:44
  Speaker 1: do different types of things. You can be creative about
• 12:46
  Speaker 1: it and do different type of things. So the inspiration
• 12:48
  Speaker 1: came from how actually humans used to do manufacturing but
• 12:52
  Speaker 1: realized in order to be flexible, you actually need two components.
• 12:56
  Speaker 1: You need intelligence and you need set of simple tools
• 13:00
  Speaker 1: with a lot of kinematic freedom. Now you can pick
• 13:03
  Speaker 1: up those simple tools, and as long as you have
• 13:05
  Speaker 1: the intelligence on how to use tools and what sequence
• 13:07
  Speaker 1: and what kind of a process how to use those tools,
• 13:11
  Speaker 1: you can actually do a whole variety of projects.
• 13:14
  Speaker 2: And so when you say kinematic freedom, you basically mean
• 13:16
  Speaker 2: like robot arms that can move in lots of different ways.
• 13:20
  Speaker 2: Is that practically what kinematic freedom means in this context?
• 13:23
  Speaker 1: Yes, basically can apply these tools in a lot with
• 13:26
  Speaker 1: a lot of freedom to the material, right the same
• 13:28
  Speaker 1: way humans humans do it, right, you know as a human,
• 13:31
  Speaker 1: you know, if you think about it, you can pick
• 13:33
  Speaker 1: up a welder and weld something, and then you drop
• 13:36
  Speaker 1: the welder, and you pick up a drill and you
• 13:38
  Speaker 1: put a hole in it, and you drop to drill
• 13:40
  Speaker 1: and you pick up a you know, hammer and maybe
• 13:43
  Speaker 1: hammer it into shape. So you actually have a few
• 13:46
  Speaker 1: set of tools, but you have a lot of good
• 13:48
  Speaker 1: kinematic freedom and most importantly, very creative mind too tells
• 13:52
  Speaker 1: you how to apply these tools to the material, so
• 13:55
  Speaker 1: they can actually get very complex set of products and
• 13:58
  Speaker 1: a lot of diversity.
• 13:59
  Speaker 2: So plainly, instead of big John, you want a robot, right,
• 14:03
  Speaker 2: That's where that's the kinematic freedom. The tools are kind
• 14:06
  Speaker 2: of like old tools, but optimized for the root. And
• 14:10
  Speaker 2: then when you say intelligence, that's the one where it's
• 14:13
  Speaker 2: like feels more frontier ish, like does that mean like
• 14:15
  Speaker 2: clever engineers figuring out how to automate the robots doesn't
• 14:19
  Speaker 2: mean AI? Does it mean both?
• 14:21
  Speaker 1: Yeah? So I think yeah, you're basically getting to the
• 14:24
  Speaker 1: crux of how do you scale it? Right? You need
• 14:27
  Speaker 1: to have those three components, and how does the intelligent piece,
• 14:31
  Speaker 1: which is the most important piece, comes into play in
• 14:34
  Speaker 1: an automated fashion. So early days we started from basic
• 14:37
  Speaker 1: intelligence of humans. But then we had a plan to
• 14:40
  Speaker 1: capture data and train AI so that you can replace
• 14:44
  Speaker 1: the thinking and the creativity that human had to put in.
• 14:47
  Speaker 2: What's the first thing you decided to try and build.
• 14:50
  Speaker 2: What's the first sort of problem you want to solve?
• 14:52
  Speaker 1: Yeah? I think so. I left relativity in twenty eighteen,
• 14:55
  Speaker 1: and the idea when I left relativity was there, right.
• 14:57
  Speaker 1: I was like, okay, we need to build basically what
• 15:00
  Speaker 1: I had in My mom called it robot craftsmen. Robocraftsman,
• 15:03
  Speaker 1: we call it a time. How can you build a
• 15:05
  Speaker 1: robot system? To your point, you can pick up different tools,
• 15:07
  Speaker 1: has the same king mean, but also have to have
• 15:09
  Speaker 1: the intelligence. The challenge is you know you said, in
• 15:13
  Speaker 1: order to train these robots with AI, you need to
• 15:15
  Speaker 1: have a lot of data. And this is not the
• 15:18
  Speaker 1: data you can find on internet.
• 15:20
  Speaker 2: Right, this is the AI robotics problem, it seems right,
• 15:23
  Speaker 2: like unlike with large language models, like that's why we
• 15:26
  Speaker 2: have large language models and not AI robots, right because
• 15:30
  Speaker 2: because we have the data just sort of randomly sitting
• 15:33
  Speaker 2: around on the Internet, and we don't have that physical
• 15:35
  Speaker 2: world data for robots.
• 15:36
  Speaker 1: Right exactly. So basically the problem narrowed down into Okay,
• 15:41
  Speaker 1: how can I generate enough data? How can I create
• 15:44
  Speaker 1: a business that has a sustaining way of generating data
• 15:47
  Speaker 1: so I can actually build these models, I can build
• 15:49
  Speaker 1: this intelligence for these robots. And the thinking was, Okay,
• 15:54
  Speaker 1: I need to create a solution that can scale in
• 15:58
  Speaker 1: the industry with limited amount of data and some heuristic.
• 16:04
  Speaker 1: But then because it's scaling, we can generate a lot
• 16:07
  Speaker 1: of data and it starts building AI mods.
• 16:09
  Speaker 2: Right. You need a first thing that you can actually
• 16:12
  Speaker 2: do before you really have AI, to generate the data
• 16:15
  Speaker 2: that will get you to.
• 16:17
  Speaker 1: AI exactly exactly. So we're thinking about, Okay, it needs
• 16:22
  Speaker 1: to be a large enough market right where we can
• 16:25
  Speaker 1: get mass adoption, and we need to solve a problem
• 16:28
  Speaker 1: that's big enough it's ten times at least better than
• 16:30
  Speaker 1: the current solution so it can actually get adoption, right.
• 16:34
  Speaker 2: Meaning you can't just do something as well, you have
• 16:36
  Speaker 2: to do it ten times better.
• 16:38
  Speaker 1: Yeah, Because I think what we realize is that through
• 16:42
  Speaker 1: the last two companies, if something is not ten times better,
• 16:45
  Speaker 1: cannot overcome the inertia that exists in an industry for
• 16:49
  Speaker 1: adoption because you know, if you're doing something for the
• 16:51
  Speaker 1: same way, and in manufacturing, people have been doing things
• 16:54
  Speaker 1: the old way for hundred of years, right.
• 16:56
  Speaker 2: Yeah, and it's a risk, right if they're going to
• 16:58
  Speaker 2: try working with you, they're immediately taking a risk. And
• 17:01
  Speaker 2: if it's only going to be a little better, why
• 17:03
  Speaker 2: should I take that risk?
• 17:05
  Speaker 1: Exactly? So the idea was, Okay, we need to find
• 17:08
  Speaker 1: it large enough market for our first application, and we
• 17:11
  Speaker 1: need to have a solution that at least ten times better.
• 17:13
  Speaker 1: So that landed us. We actually looked at a lot
• 17:15
  Speaker 1: of things, from three D printing to forging to a
• 17:17
  Speaker 1: lot of things, and then landed on sheet metal. So
• 17:20
  Speaker 1: sheet metal is the largest metal processing sector out of
• 17:23
  Speaker 1: all It's a two hundred and eighty billion dollar industry today,
• 17:27
  Speaker 1: and forming complex sheet metal shapes is very tool intensive.
• 17:31
  Speaker 1: So so what we started to do was, okay, can
• 17:35
  Speaker 1: we make our robot craftsman's first operation to be forming
• 17:37
  Speaker 1: sheet metal, basically forming sheet metal the same way a
• 17:40
  Speaker 1: sheet shaper hammer is a sheet into shape.
• 17:43
  Speaker 2: And when I think about sheet metal, I mean I
• 17:45
  Speaker 2: don't know anything about sheet metal. I think of like
• 17:48
  Speaker 2: I think of cars, I think of planes, right, I
• 17:50
  Speaker 2: think of like you know, detroit, like stamping.
• 17:53
  Speaker 1: Is that?
• 17:53
  Speaker 2: Am I thinking about the right thing? Am I missing
• 17:55
  Speaker 2: huge or huge sheet metal universe?
• 17:57
  Speaker 1: Like?
• 17:57
  Speaker 2: What's the sheet metal universe?
• 17:59
  Speaker 1: Yes? So sheep metal almost is everywhere. I think is
• 18:01
  Speaker 1: the most common metal part that you see on day
• 18:06
  Speaker 1: to day, right, because most of the time we use
• 18:08
  Speaker 1: metal to be a container for other things. So it's
• 18:11
  Speaker 1: usually a thin metal structure that's formed in complex shape
• 18:15
  Speaker 1: to hold something else. Now you know it can be
• 18:17
  Speaker 1: from case of a computer. Uh, you know, to a car, right,
• 18:23
  Speaker 1: you know you're sitting in a freeway you're in to
• 18:25
  Speaker 1: see a sheet metal or to a to a airplane
• 18:28
  Speaker 1: you're in a sheet metal can to a rocket body.
• 18:31
  Speaker 2: Uh.
• 18:32
  Speaker 1: For for a lot of rocket someone will composites with
• 18:33
  Speaker 1: a lot of a machine metal. And to agricultural heavy
• 18:39
  Speaker 1: equipment machinery you think of combines, tractors to even building equipments.
• 18:45
  Speaker 1: You look at your h ract ducts are all sheet metal, right,
• 18:50
  Speaker 1: because it just makes sense. It's we mostly use metal
• 18:53
  Speaker 1: parts to contain other things, and we give it complex
• 18:56
  Speaker 1: shapes and that's where she forming comes into play. So
• 18:58
  Speaker 1: you pretty much see it everywhere. But the challenge is
• 19:01
  Speaker 1: that in almost in all cases, you have to create tooling.
• 19:04
  Speaker 1: It goes back to that first problem. He said, you
• 19:06
  Speaker 1: have to create tooling for each of those geometries. And
• 19:09
  Speaker 1: that's why you know a Ford needs to make sure
• 19:13
  Speaker 1: they can sell a million of an FUN fifty before
• 19:15
  Speaker 1: they can invest in a plant that makes a new
• 19:18
  Speaker 1: version of FN fifty, right.
• 19:19
  Speaker 2: Because you basically have to build a bespoke factory just
• 19:24
  Speaker 2: to shape sheet metal in a new way exactly for
• 19:28
  Speaker 2: a new geometry, for for a new design. Exactly where
• 19:30
  Speaker 2: is that a particular problem? Like where is it? Where
• 19:34
  Speaker 2: is that? Where does that acutely bind the fact that
• 19:37
  Speaker 2: sheet metal is so hard to do if you're not
• 19:39
  Speaker 2: working at.
• 19:39
  Speaker 1: Scale so expensive? Yeah, so I think now you're coming
• 19:42
  Speaker 1: to the even the third stage of how do you
• 19:45
  Speaker 1: scale this technology? You need to first find you know,
• 19:48
  Speaker 1: you said you need to be ten ex better we
• 19:49
  Speaker 1: need Right, you're in an area that has a lot
• 19:53
  Speaker 1: of pain with today's time.
• 19:54
  Speaker 2: I was like, oh my god, thank god, you've walked
• 19:56
  Speaker 2: through the door. We've been waiting for you.
• 19:58
  Speaker 1: Yeah, So end up being very much defense in airspace. Right,
• 20:02
  Speaker 1: So think of you know, think of our military for example,
• 20:07
  Speaker 1: right today, they we have fifty sixty different weapon system
• 20:13
  Speaker 1: or defense systems you can basically think of like aircrafts
• 20:17
  Speaker 1: that they're maintaining. And some of these systems have been
• 20:19
  Speaker 1: built from sixty seventy eighty years ago, like think of
• 20:23
  Speaker 1: B fifty two C one thirty like World War.
• 20:25
  Speaker 2: Two planes still flying kind.
• 20:27
  Speaker 1: Of still flying, yes, exactly, and they have like you know,
• 20:31
  Speaker 1: thirty of one, fifty of another, one hundred of another one.
• 20:35
  Speaker 1: And these things get break down, right, and unlike a
• 20:38
  Speaker 1: Ford factory, there is no factory for seventy different products
• 20:42
  Speaker 1: that they're carrying.
• 20:42
  Speaker 2: Right, and presumably the factory they built in nineteen forty
• 20:46
  Speaker 2: one to build this plane doesn't.
• 20:47
  Speaker 1: Exist any It doesn't exact. Even the vendor might completely
• 20:50
  Speaker 1: have disappeared, right, that made that misspecific component. So they're
• 20:54
  Speaker 1: constantly battling with this challenge of an aircraft goes down,
• 20:58
  Speaker 1: how can I fix it? How can I find the part?
• 20:59
  Speaker 1: And there are thousands of parts in each of these aircrafts, right,
• 21:02
  Speaker 1: so any of them can go down, and that's a
• 21:04
  Speaker 1: huge challenge. I mean, if you look at you know,
• 21:06
  Speaker 1: government of a government accountabilit The office put this report out,
• 21:11
  Speaker 1: I think it was a couple of years ago or
• 21:13
  Speaker 1: a year ago about how ready each weapon system is
• 21:16
  Speaker 1: to defend the United States. Out of the forty eight
• 21:19
  Speaker 1: to forty nine weapon systems they look into only one,
• 21:23
  Speaker 1: only one in the past eleven years, every year was ready, right.
• 21:29
  Speaker 1: I think only top four had like at least half
• 21:32
  Speaker 1: of the years ready right. So that means in most
• 21:35
  Speaker 1: years these weapons are not ready to fight, like.
• 21:37
  Speaker 2: They're waiting for parts.
• 21:39
  Speaker 1: They're waiting for parts. Something is broken, something is damaged,
• 21:42
  Speaker 1: and we used to go deeper. Some of these components
• 21:44
  Speaker 1: take four years to be replaced. So if a plane
• 21:47
  Speaker 1: gets damaged, it needs to sit on the ground for
• 21:49
  Speaker 1: four years before it can be it can be replaced,
• 21:52
  Speaker 1: and the cost of replacement is building another factory basically,
• 21:55
  Speaker 1: So some of these parts, and think of it, a
• 21:57
  Speaker 1: landing gear door that goes on a plane will cost
• 22:00
  Speaker 1: them eight hundred thousand dollars for example, because they have.
• 22:02
  Speaker 2: To go make it because it's bespoke essentially, like buying
• 22:05
  Speaker 2: a bespoke suit or something. It's just like it's gonna
• 22:08
  Speaker 2: cost a lot.
• 22:08
  Speaker 1: Yeah, yeah, So the idea started there. I think that
• 22:11
  Speaker 1: was one of our first customers. Can we make defense
• 22:14
  Speaker 1: manufacturing more agile? Directly affects our national readiness for military
• 22:19
  Speaker 1: conflict and it's a huge problem. But then you know,
• 22:22
  Speaker 1: even in a broader sense, any defense product or aerospace
• 22:25
  Speaker 1: product usually has very low volume but high mix of products.
• 22:29
  Speaker 1: You know, even you know, you're building a missile, you
• 22:31
  Speaker 1: make like, you know, a few thousand a year, and
• 22:33
  Speaker 1: you might make five, six seven different versions of right,
• 22:35
  Speaker 1: So it's very unlike cars, where you know, you make
• 22:38
  Speaker 1: a million of the same car over and over all.
• 22:40
  Speaker 1: So that ends up being our first application, which we've
• 22:44
  Speaker 1: got a lot of traction with. But but you know,
• 22:47
  Speaker 1: even outside of that, you know, you look at companies
• 22:50
  Speaker 1: like Caterpillar, like John Deere's of the world. These folks
• 22:53
  Speaker 1: also are in the same book. You know, they make
• 22:55
  Speaker 1: two hundred combines, right, but they need to support them
• 22:58
  Speaker 1: in the field. And these folks have the exactly same problem, right,
• 23:01
  Speaker 1: you know, do I need to run a large factory
• 23:05
  Speaker 1: to support all these models at all time, and that's
• 23:08
  Speaker 1: will be very expensive to support, like one hundred vehicles out.
• 23:11
  Speaker 2: There still to come on the show. We'll talk about
• 23:17
  Speaker 2: the future of AI and robotics at Mocking the Labs
• 23:20
  Speaker 2: and beyond. And so you got the right market. Now
• 23:33
  Speaker 2: you've got to make a thing. You got to figure
• 23:35
  Speaker 2: out how to actually do the thing, how to make
• 23:37
  Speaker 2: your idea come true? Like how does that work?
• 23:41
  Speaker 1: So the idea originally was can we get rid of
• 23:44
  Speaker 1: a die right and do it the same way a
• 23:47
  Speaker 1: sheet shaper forms a sheet of metal? And what does
• 23:50
  Speaker 1: a sheet chaper do? Sheet chaper get starts from a
• 23:52
  Speaker 1: flat sheet of metal and it slowly hammers it into shape.
• 23:57
  Speaker 1: So what we wanted to do was have a robot
• 23:59
  Speaker 1: do that, right, have robotic system basically do that incremental
• 24:03
  Speaker 1: deformation into shape. We call it romophor.
• 24:05
  Speaker 2: So you're sort of bending it, right, I mean you're
• 24:07
  Speaker 2: hammering it. It's sort of like if you take a whatever,
• 24:10
  Speaker 2: cut open a luminum can and kind of bend it
• 24:11
  Speaker 2: into shape. Like that's a version of what's happening here, right, Yes,
• 24:16
  Speaker 2: exactly complicated way, yeah.
• 24:17
  Speaker 1: Exactly, you're right, I mean the same way a potter
• 24:22
  Speaker 1: forms a clay bowl. That's basically what our robots do.
• 24:24
  Speaker 1: They start from a flash sheet of metal and slowly
• 24:26
  Speaker 1: deforming in the shape the same way a potter, which
• 24:28
  Speaker 1: form it clay bowl or a sheet shape of hammets
• 24:31
  Speaker 1: and hammers a sheet into shape. Yeah, so I've seen
• 24:33
  Speaker 1: it right.
• 24:34
  Speaker 2: So there will be a sheet of metal like hanging
• 24:36
  Speaker 2: hanging up in whatever above the ground. And then you
• 24:39
  Speaker 2: have a robot arm on either side, right, like one
• 24:42
  Speaker 2: on one side, one on the other. And then what.
• 24:44
  Speaker 1: Happens Basically the robots come together from both sides the
• 24:48
  Speaker 1: sheet and they pinch the sheet in a certain way
• 24:52
  Speaker 1: so that that location that they're pinching slightly stretches and deforms. Right.
• 24:57
  Speaker 1: And if you start applying this pinching all over the
• 25:02
  Speaker 1: sheet and incrementally, you slowly start to form it into
• 25:05
  Speaker 1: a shape. Right. So instead of traditionally would use it
• 25:09
  Speaker 1: die and with sheer pressure of the press pushing the
• 25:12
  Speaker 1: sheet against the dye to give it a shape. Now
• 25:14
  Speaker 1: the robots are like a craftsman, like a trades person
• 25:17
  Speaker 1: coming in to slowly deform the sheet into shape by
• 25:20
  Speaker 1: just applying pressure. So one robot is pushing it the
• 25:23
  Speaker 1: other robot is supporting it, and by applying a pinch
• 25:26
  Speaker 1: you slightly stretch the material and you form it into
• 25:29
  Speaker 1: a shape.
• 25:30
  Speaker 2: So I mean, the way you describe it, it makes
• 25:33
  Speaker 2: sense and it sounds easy. I'm sure it wasn't easy,
• 25:39
  Speaker 2: Like were there things that just didn't work for a while.
• 25:44
  Speaker 1: So you should have been here when the first time
• 25:47
  Speaker 1: we actually tried to form a part, the part looked
• 25:50
  Speaker 1: like it was like a ghost of the geometry that
• 25:53
  Speaker 1: they wanted to make, and actually in the end it
• 25:55
  Speaker 1: tore right. So think about it. You have this very
• 25:58
  Speaker 1: flimsy sheet applying pressure to it, and if features apply
• 26:03
  Speaker 1: pressure slightly wrong right, it can potentially tear it. It
• 26:08
  Speaker 1: can form it into a different shape. And also the
• 26:10
  Speaker 1: whole sheet is moving the whole time you're trying to
• 26:12
  Speaker 1: move to form it. The whole sheet is moving because
• 26:15
  Speaker 1: it's very flimsy. It's not a rigid structure, right. So
• 26:19
  Speaker 1: the main challenge was how do you get this accurate? Right?
• 26:22
  Speaker 1: How do you get this process accurate? How do you
• 26:23
  Speaker 1: get accuracy? And the idea was what does the robot
• 26:27
  Speaker 1: need to do given all of these chaotic nature of
• 26:30
  Speaker 1: the process where the sheet moves and if you apply
• 26:32
  Speaker 1: it too much pressure, it will deform in a bad
• 26:35
  Speaker 1: way or in my tear. If you're probably not enough pressure,
• 26:38
  Speaker 1: it might just not form. So how do you come
• 26:40
  Speaker 1: up with the right set of robot movements and process
• 26:43
  Speaker 1: parameters to form the part? And that was the problem
• 26:46
  Speaker 1: we want to solve with AI right, but we didn't
• 26:48
  Speaker 1: have the data right right in the beginning. Right. The
• 26:52
  Speaker 1: idea was that if I form enough parts with this process,
• 26:56
  Speaker 1: and I can capture all the data throughout the process,
• 26:58
  Speaker 1: where did the robot go, how much pressure did it apply,
• 27:01
  Speaker 1: and what was the resulting geometry? That can start building
• 27:04
  Speaker 1: a model that says that correlates the inputs to the outputs,
• 27:07
  Speaker 1: and I can explore this and say, okay, in order
• 27:10
  Speaker 1: to get to the right output, I need these inputs.
• 27:12
  Speaker 1: But we didn't have them in the beginning. So the
• 27:15
  Speaker 1: idea was two things. One was maybe we can simulate
• 27:18
  Speaker 1: the data right and very early on we started doing
• 27:22
  Speaker 1: some simulation, physics based simulation, and we soon realized in
• 27:24
  Speaker 1: order to get an accurate result, the simulations are going
• 27:28
  Speaker 1: to be very computationally intensive. A simulation of a part
• 27:31
  Speaker 1: that took only fifteen minutes to form took us one
• 27:35
  Speaker 1: week on twenty seven core machine. Wow. Right, so okay,
• 27:39
  Speaker 1: stimulation not only is not accurate, it takes forever. So
• 27:43
  Speaker 1: we realized, okay, so that's not the right route. The
• 27:46
  Speaker 1: right route was like, Okay, we can also form a
• 27:48
  Speaker 1: lot of parts and gather the data. But in order
• 27:51
  Speaker 1: to do that we go back to that same problem.
• 27:53
  Speaker 1: We need to have a scale. We need to have
• 27:54
  Speaker 1: a lot of these machines for these parts and get
• 27:56
  Speaker 1: that data.
• 27:57
  Speaker 2: I mean, one of the big AI insights of the
• 27:59
  Speaker 2: last whatever decade is like, you need a ton of data,
• 28:03
  Speaker 2: which is easy if it's words, but hard if it's metal. Right.
• 28:07
  Speaker 1: Yes, we ended up doing was created a hybrid model.
• 28:11
  Speaker 1: We said, okay, what if we keep the humans in
• 28:15
  Speaker 1: the loop, so the human can give an instruction initially
• 28:18
  Speaker 1: based on herostricts, and then we look at the data
• 28:21
  Speaker 1: and human can adjust and then iterate on that. But
• 28:25
  Speaker 1: while we are capturing all these data, and over time,
• 28:29
  Speaker 1: as we're capturing the data, we start building the models
• 28:31
  Speaker 1: that will help the human do less trials. Right. It's
• 28:35
  Speaker 1: basically guided reinforcement learning, right, and a humans are actually
• 28:38
  Speaker 1: guiding it where to go, but it's exploring those areas
• 28:41
  Speaker 1: but after a while, once we started forming south thousands
• 28:43
  Speaker 1: of parts, then you can start feeding this data into model.
• 28:47
  Speaker 1: Then the model will be like, okay, human, you don't
• 28:50
  Speaker 1: need to do twenty five different trials. Now you can
• 28:52
  Speaker 1: do with five trials, you're going to get to the
• 28:54
  Speaker 1: right place, which is actually the number we are at
• 28:56
  Speaker 1: right now.
• 28:56
  Speaker 2: And that's happening in the physical world largely those iterations
• 29:00
  Speaker 2: like you're trying a piece of metal and it's bad
• 29:02
  Speaker 2: and it tears, and you do another piece of metal
• 29:04
  Speaker 2: and it's a little less bad and eventually.
• 29:06
  Speaker 1: Exactly exactly, and that initially would take twenty five parts,
• 29:10
  Speaker 1: like you know, before we find a recipe for that design.
• 29:13
  Speaker 1: But twenty five parts still was better than traditional alternative.
• 29:16
  Speaker 2: When you say twenty five parts, I mean twenty five
• 29:18
  Speaker 2: tries twenty five pieces of metal before you make the
• 29:21
  Speaker 2: part the right way exactly.
• 29:23
  Speaker 1: And that was like, you know, they would sit down
• 29:25
  Speaker 1: basically twenty five days in a row, so in a
• 29:27
  Speaker 1: month they could actually define a recipe where traditionally making
• 29:30
  Speaker 1: a mold would take at least three four months. Right,
• 29:32
  Speaker 1: So we were still better. But then now with over time,
• 29:36
  Speaker 1: when we generated the data and now the model can
• 29:38
  Speaker 1: tell the engineer, okay, maybe you want to choose these parameters,
• 29:43
  Speaker 1: is now becoming an advisor with down to five trials.
• 29:46
  Speaker 1: In five trials, we can actually get to the right
• 29:48
  Speaker 1: part and then hopefully in the future we get to
• 29:50
  Speaker 1: a point where you know, the machine will tell the
• 29:53
  Speaker 1: romans what to do and the human can be completely
• 29:55
  Speaker 1: out of the loop. Yeah. But the idea was like,
• 29:57
  Speaker 1: how do you kind of create that hybrid model that's
• 29:59
  Speaker 1: efficient so that we can generate the data until the
• 30:03
  Speaker 1: model is good enough to do the job itself.
• 30:05
  Speaker 2: And you find that the data is sort of generalizable,
• 30:09
  Speaker 2: I mean clearly, like making one kind of part makes
• 30:12
  Speaker 2: the model the AI smarter about making another kind of part.
• 30:16
  Speaker 1: Yes, you know, yeah it is. It's kind of interesting.
• 30:18
  Speaker 1: I think people don't think about it. I used to
• 30:21
  Speaker 1: do sheet shaping by hand, right, That was one of
• 30:23
  Speaker 1: the hobbies I had. I was working with this shop
• 30:25
  Speaker 1: in Pomona that we were actually hammer sheets into shape,
• 30:29
  Speaker 1: and we used to say, you know, if you spent
• 30:31
  Speaker 1: five years doing it, you're really good. You get really
• 30:33
  Speaker 1: good at it. I was used to think, you know, okay,
• 30:36
  Speaker 1: after five years of doing this, yes, you have this
• 30:38
  Speaker 1: intuitive understanding of you look at the sheet and be like, okay,
• 30:40
  Speaker 1: this this place needs to be hammered more. This place
• 30:42
  Speaker 1: needs to be hammered. It was, it was, it was.
• 30:44
  Speaker 1: It was intuitive. It was like you couldn't explains why
• 30:46
  Speaker 1: you're thinking this need to happen. There was no physical explanation.
• 30:49
  Speaker 1: None of these people who were she shaping got PhDs
• 30:52
  Speaker 1: in material science. Yeah, they just learned over time seeing
• 30:55
  Speaker 1: the pattern of how the sheet formed. Yes, craftsmanship, that's
• 30:58
  Speaker 1: craftsmanship right. Yeah, but really reminded me of Okay, these
• 31:02
  Speaker 1: people can know how to do it, but without really
• 31:04
  Speaker 1: being able to explain it, to do it for five years.
• 31:07
  Speaker 2: It's that kind of tacit knowledge.
• 31:09
  Speaker 1: Yeah, and reminded me of the same challenge we had
• 31:14
  Speaker 1: early machine learning challenge where they were like, okay, a
• 31:16
  Speaker 1: human can look at two pictures and say, okay, this
• 31:19
  Speaker 1: is a cat and this is a dog. Something happens
• 31:21
  Speaker 1: in their brain that knows which is a cat, but
• 31:23
  Speaker 1: they cannot really define why they're calling this cat and
• 31:25
  Speaker 1: this sort of dog. So that was where it starts
• 31:28
  Speaker 1: to click for me. If I can capture enough data,
• 31:30
  Speaker 1: five years worth of data right of a human, then
• 31:34
  Speaker 1: I should be able to get to a very good
• 31:36
  Speaker 1: sheet shaper, right, And you know it's funny. Back at
• 31:40
  Speaker 1: the end, I was like, okay, humans are you know,
• 31:42
  Speaker 1: receiving x amount of megabytes a second? Okay, how five
• 31:45
  Speaker 1: years worse of data? Is that much? So roughly, I
• 31:47
  Speaker 1: think once we get a certain amount of data, I
• 31:50
  Speaker 1: think we have enough data to be able to basically
• 31:53
  Speaker 1: replace a like not replace replace the mentality or the
• 31:56
  Speaker 1: model that the sheet shaper has in their mind.
• 31:59
  Speaker 2: So how how how many years of kind of human
• 32:02
  Speaker 2: level craftsmen sheet shaping data does the model have at
• 32:06
  Speaker 2: this point?
• 32:07
  Speaker 1: Yeah? No, so I think rastam I check? One year ago?
• 32:09
  Speaker 1: I checked around, we were like three fourths of the
• 32:11
  Speaker 1: way there in terms of the data that we have
• 32:14
  Speaker 1: for just she shaping. Right. So once we get to
• 32:16
  Speaker 1: I think full, I think and these at that point
• 32:18
  Speaker 1: we have no excuse. We have enough data. The model
• 32:19
  Speaker 1: should be good. We just need to figure out how
• 32:22
  Speaker 1: why it's not. Maybe far from it is.
• 32:24
  Speaker 2: It is interesting to analogize it to like human craftsmanship, right,
• 32:29
  Speaker 2: And I mean even if you want to zoom out
• 32:31
  Speaker 2: even more, the like fifty year history of AI, where
• 32:34
  Speaker 2: first everybody was like, oh, you just got to teach
• 32:36
  Speaker 2: the machine all the rules for to use your example,
• 32:39
  Speaker 2: like what's a cat and what's a dog? But then
• 32:41
  Speaker 2: you realize it's actually wildly hard to make a list
• 32:43
  Speaker 2: of rules that can reliably distinguish a cat from a dog.
• 32:48
  Speaker 2: And the weird thing that has happened in AI is like, oh,
• 32:51
  Speaker 2: you don't actually have to make a list. You just
• 32:55
  Speaker 2: need like image that you just need like a giant
• 32:58
  Speaker 2: database of images and a giant neural network and you
• 33:03
  Speaker 2: just throw it at it like and say figure it out,
• 33:06
  Speaker 2: and it figures it out, and you're sort of doing that.
• 33:09
  Speaker 2: But for shaping metal.
• 33:11
  Speaker 1: For metal, and then the only challenge was, like you know,
• 33:13
  Speaker 1: cats and dogs pictures were Internet and sheet metal forming
• 33:17
  Speaker 1: data wasn't. And so that's that was an additional problem
• 33:20
  Speaker 1: we have to solve, as you pointed out, which is
• 33:21
  Speaker 1: a big problem in physical AI.
• 33:23
  Speaker 2: So I want to talk a little bit more about
• 33:26
  Speaker 2: AI and robotics. Jansen Wong has been talking about it,
• 33:31
  Speaker 2: as I'm sure you know in video and videos vc
• 33:35
  Speaker 2: ARM as an investor in your company, other people are
• 33:38
  Speaker 2: working on what you're working on. I mean, I'm curious
• 33:41
  Speaker 2: what does the sort of AI and robotics path look
• 33:43
  Speaker 2: like to you? For the next few years, and what like,
• 33:45
  Speaker 2: what do you understand about it now that you didn't
• 33:48
  Speaker 2: understand whatever five years ago? Like what what have you
• 33:51
  Speaker 2: really come to realize by working on it all the time?
• 33:54
  Speaker 1: I think the biggest problem for physical AI is data
• 33:57
  Speaker 1: generation for now to train models. So we need to
• 34:01
  Speaker 1: either there's two things need to happen. Either new types
• 34:04
  Speaker 1: of models needs to be created, new architectures, new new
• 34:07
  Speaker 1: algorithms basically, which I'm sure it's going to happen that
• 34:11
  Speaker 1: can learn more with less data basically and the same
• 34:15
  Speaker 1: way humans kind of learn more with less data. Right.
• 34:18
  Speaker 1: But at the same time, I think, you know, we
• 34:21
  Speaker 1: only exposed our models to categorically to ten percent of
• 34:25
  Speaker 1: type of data that humans receive. You know, you think
• 34:28
  Speaker 1: about you know, human intrictions. You and I are now talking,
• 34:33
  Speaker 1: if it was AI, AI is probably only listening to
• 34:35
  Speaker 1: the words we're saying, right, But that's only ten percent
• 34:38
  Speaker 1: of communication. I can see your lips moving, I can
• 34:41
  Speaker 1: see your eyebrows moving. I can see like maybe you're
• 34:43
  Speaker 1: folding your arms and okay, I know that like okay,
• 34:45
  Speaker 1: maybe there's all these ninety percent of the signals are
• 34:48
  Speaker 1: not captured. That that's used for learning. You know, you
• 34:52
  Speaker 1: look at if you ask chat GPT or Dolly or
• 34:56
  Speaker 1: you know any of the you know, even even you
• 34:58
  Speaker 1: know Grock say okay, draw me a clock that is
• 35:02
  Speaker 1: shows five thirty. It cannot show you draw you a clock.
• 35:06
  Speaker 1: It will draw you a clock, but it doesn't show
• 35:07
  Speaker 1: five thirty. Actually, most the time it shows ten ten.
• 35:10
  Speaker 2: Ten ten, because that's where watchhands, like analog watchhands look good.
• 35:14
  Speaker 1: Right, it's a nice little v because those are all
• 35:17
  Speaker 1: the images that they're seen on internet because they watch it.
• 35:19
  Speaker 2: It's almost always ten ten. It's the classic watch photo.
• 35:22
  Speaker 1: It's like five thirty is also ten ten, because.
• 35:24
  Speaker 2: It's always ten ten right to a generative AI, it's
• 35:28
  Speaker 2: always ten ten somewhere.
• 35:30
  Speaker 1: So I think, but that humans, you know, receive this
• 35:33
  Speaker 1: data of movement. When you grow up you look at
• 35:35
  Speaker 1: the clock on the wall as a kid, you're like, okay,
• 35:37
  Speaker 1: now I intuitively get it. I think I know what's
• 35:39
  Speaker 1: going on, so I can actually make it work. So
• 35:42
  Speaker 1: even though we train it a lot of data, I
• 35:44
  Speaker 1: don't think we trained it on the right categorically right
• 35:46
  Speaker 1: data yet right to get all the intuitive understanding that
• 35:50
  Speaker 1: we have today. So I think we have a data
• 35:53
  Speaker 1: problem and that exists the physical AI. So I think
• 35:56
  Speaker 1: the applications will win. There's a lot of people are
• 35:58
  Speaker 1: working in this. I think the applications will win who
• 36:01
  Speaker 1: can either synthetically generate that data or they can actually
• 36:05
  Speaker 1: scale in the physical world in a way where they
• 36:07
  Speaker 1: can actually generate the day for themselves. But the scaling
• 36:12
  Speaker 1: needs to happen with less data, and I think that
• 36:14
  Speaker 1: was That's why I'm like, for example, like very bullish
• 36:17
  Speaker 1: on manufacturing. So I think the data is going to
• 36:19
  Speaker 1: be the biggest challenge. And I think, you know, in
• 36:21
  Speaker 1: order for us to massively change this space, we need
• 36:25
  Speaker 1: to be able to get to the data. I don't
• 36:26
  Speaker 1: think algorithms is a bottleneck there yet. It's just a
• 36:30
  Speaker 1: data for us.
• 36:31
  Speaker 2: And is it just a matter of people doing what
• 36:36
  Speaker 2: you're doing and like finding little wedge places to start
• 36:39
  Speaker 2: and having people sort of hold the hand of the
• 36:41
  Speaker 2: model and training up the models. I mean that seems
• 36:46
  Speaker 2: slow on a certain level, like not you know, obviously
• 36:48
  Speaker 2: it's working for you, but like, is there some kind
• 36:52
  Speaker 2: of breakthrough move people can make? Can you put sensors
• 36:55
  Speaker 2: somewhere in the world to you know, train AI without
• 36:58
  Speaker 2: having to you know, have a human stand next to
• 37:01
  Speaker 2: it as it messes up one piece of sheet metal
• 37:04
  Speaker 2: after another.
• 37:05
  Speaker 1: Yeah, I think I think that there is there's another path,
• 37:09
  Speaker 1: which is simulation path. Make physics based simulations faster and
• 37:15
  Speaker 1: kind of learn. Let the robots just go play in
• 37:17
  Speaker 1: a digital playground as opposed to deploy it in real role,
• 37:20
  Speaker 1: and that becomes a computation problem. And then you know,
• 37:22
  Speaker 1: as long as you have enough computation, you can train
• 37:24
  Speaker 1: to robots. But I think, you know, I think you
• 37:26
  Speaker 1: know the good examples that we have had such success
• 37:28
  Speaker 1: so far as like autonomous cars, right, did the same
• 37:31
  Speaker 1: thing we were doing, but in the car like Okay, Tesla,
• 37:35
  Speaker 1: you know, deploy the fleet of robots that are capturing
• 37:37
  Speaker 1: data still be driven by humans, but the data can
• 37:40
  Speaker 1: be used later on to kind of automate it.
• 37:42
  Speaker 2: I mean, that's an interesting case because it has been
• 37:46
  Speaker 2: much harder clearly than many people thought. Maybe most people thought, right, Like,
• 37:51
  Speaker 2: I know, that's a particular instance where you're really worried
• 37:54
  Speaker 2: about edge cases. I don't know, is autonomous cars like
• 37:59
  Speaker 2: a good model or not. It seems complicated.
• 38:01
  Speaker 1: I think the model of capturing data is there, but
• 38:03
  Speaker 1: then the the task at hand is very hard. Yeah, right,
• 38:07
  Speaker 1: so I think that's the challenge, right so where it
• 38:10
  Speaker 1: says like with us, it's still much more structured environment,
• 38:13
  Speaker 1: And I think that's that was the thinking we're thinking.
• 38:15
  Speaker 1: I think the hardest problem right now in physical AI
• 38:17
  Speaker 1: is finding the business model of how do you scale
• 38:20
  Speaker 1: data capture without requiring billions of dollars in investment?
• 38:24
  Speaker 2: So what do you make in today?
• 38:26
  Speaker 1: I imagine you know, so last time I checked in
• 38:29
  Speaker 1: the facility, one four of the sales are working on
• 38:32
  Speaker 1: a defense application.
• 38:35
  Speaker 2: Is it secret? Can you tell me what it is?
• 38:38
  Speaker 1: It's a missile? And two of them were working on
• 38:41
  Speaker 1: an aerospace application. This is components of an aircraft or
• 38:45
  Speaker 1: a drawing. And one of them, as an interesting one,
• 38:48
  Speaker 1: was working on an architectural component, which is a roof
• 38:51
  Speaker 1: tile for a specific building that's used by the Department
• 38:56
  Speaker 1: of the by Bureau of Water Recognition.
• 38:58
  Speaker 2: Oh, I was I was going to say, what is it?
• 39:00
  Speaker 2: Something like Frank Gary, like nightmare weirdo metal park.
• 39:05
  Speaker 1: Oh, those we have had those in the past two
• 39:07
  Speaker 1: but this one is actually very practical. Well, it's this building.
• 39:10
  Speaker 1: It's actually very interesting. Exactly these buildings, these large industrial
• 39:14
  Speaker 1: buildings that built they built in the sixties or fifties,
• 39:17
  Speaker 1: and they use these type of roof tiles that the
• 39:19
  Speaker 1: manufacturer doesn't exist anymore. And anybody else who they went
• 39:23
  Speaker 1: to quoted them hundreds of thousand dollars to make those tiles,
• 39:26
  Speaker 1: and we're like, oh no, we can make it for you.
• 39:29
  Speaker 1: But also that show is kind of the diversity. I mean,
• 39:31
  Speaker 1: like like I say, in the morning, we have like
• 39:32
  Speaker 1: aerospace parts. In the afternoon, roof tiles for a industrial complex,
• 39:37
  Speaker 1: for you know, for a dam.
• 39:40
  Speaker 2: Now you're in the sheet metal business. I know you're large.
• 39:43
  Speaker 2: Dream is much larger than that, right, but like what
• 39:48
  Speaker 2: like that, tell me where you are now? Tell me
• 39:51
  Speaker 2: where you are now? Like what are you doing? What
• 39:53
  Speaker 2: are you selling? And then kind of what's the next
• 39:56
  Speaker 2: big step.
• 39:57
  Speaker 1: So some of our systems are now operating out in
• 39:59
  Speaker 1: the wild and working for the customers. And but I
• 40:03
  Speaker 1: think the next phase of growth for us is getting
• 40:07
  Speaker 1: into each of these applications and own more of the
• 40:10
  Speaker 1: process so we can teach the robocraftsmen the future processes
• 40:13
  Speaker 1: not just sheet for me, but also maybe how to
• 40:15
  Speaker 1: assemble it, how to weld it, how do you surface
• 40:17
  Speaker 1: finish it right. So what we are doing now in
• 40:21
  Speaker 1: the next phase is actually instead of selling parts or
• 40:24
  Speaker 1: components or systems, we're actually saying, Okay, can we get
• 40:29
  Speaker 1: this robocraftsman to actually build you a subassembly or a
• 40:33
  Speaker 1: full product, not just a component of it, but a
• 40:36
  Speaker 1: full product. So that's something we're describing with folks. Can
• 40:39
  Speaker 1: we have the robocraftsmen build the full drone for you?
• 40:43
  Speaker 1: Can we have the robocrafts and build you a full
• 40:45
  Speaker 1: missile as opposed to just build missile you know missile scans.
• 40:49
  Speaker 2: Is there that seems like a leap? Is there not
• 40:53
  Speaker 2: an intermediate step?
• 40:55
  Speaker 1: Like yes? Yes? So I mean how we're doing it
• 40:58
  Speaker 1: is we're gradually stepping into it right the same way
• 41:01
  Speaker 1: she metal was our first application. So we're putting a
• 41:03
  Speaker 1: facility that maybe makes drones, but the main component that
• 41:08
  Speaker 1: we automate today is sheet for me, which is the bottleneck.
• 41:12
  Speaker 1: And then we do the welding in a traditional way
• 41:14
  Speaker 1: on the same robots, but we actually instruct them to
• 41:17
  Speaker 1: do it.
• 41:18
  Speaker 2: So that way, the robot is kind of back where
• 41:21
  Speaker 2: it was on sheet metal five years ago, but it's
• 41:24
  Speaker 2: learning how to weld now exactly.
• 41:26
  Speaker 1: I used to work in a you know, a shop
• 41:29
  Speaker 1: that we will do custom cars, build the custom cars
• 41:31
  Speaker 1: with hand, and so it was also near near and
• 41:33
  Speaker 1: dear to my heart. So what we realize is that
• 41:37
  Speaker 1: with our technology, for the first time, we can actually
• 41:41
  Speaker 1: enable a product that didn't exist in automotive, meaning that
• 41:45
  Speaker 1: instead of buying a car that's mass produced and every
• 41:48
  Speaker 1: single one of them look the same, you can now
• 41:50
  Speaker 1: let the customer design a custom car for them. You know,
• 41:54
  Speaker 1: right now, if you go buy a car, you can
• 41:55
  Speaker 1: you have options of what the what the seat color
• 41:58
  Speaker 1: would be, or maybe the color of the car would be,
• 42:01
  Speaker 1: and what some trim options. But you can't really choose
• 42:04
  Speaker 1: the design of your car. You can't say, oh, I
• 42:06
  Speaker 1: want a different hood and I want a different fender
• 42:09
  Speaker 1: because going to the back same problem you have to
• 42:11
  Speaker 1: make tooling and mold for the vender of certain designs.
• 42:15
  Speaker 1: It cannot easily change it. So with our technology you can.
• 42:18
  Speaker 1: So what we started doing was like, okay, applying this
• 42:23
  Speaker 1: freedom that this technology provides to now automotive is the
• 42:26
  Speaker 1: ability of the customers to be able to go to
• 42:28
  Speaker 1: a website design a fully customized car for themselves. It
• 42:33
  Speaker 1: can be either from already design panels round car designer
• 42:36
  Speaker 1: or adding a specific customer customizations they want to do,
• 42:40
  Speaker 1: for example, logo of their company to their door of
• 42:42
  Speaker 1: the car or the hood of the car, and actually
• 42:44
  Speaker 1: get a completely unique car right manufactured for them. And
• 42:49
  Speaker 1: we're actually working with this with some of our automotive
• 42:52
  Speaker 1: partners Automotive Aims as well. Right, we actually showed some
• 42:55
  Speaker 1: of this work in the biggest aftermarket show in the
• 42:58
  Speaker 1: United States is called SEMA with our partner Toyota. So
• 43:02
  Speaker 1: I think this is going to be, in my opinion,
• 43:05
  Speaker 1: one of the new product categories in automotive. We have
• 43:08
  Speaker 1: had a time of his cars, we have had you know,
• 43:11
  Speaker 1: electric cars, and I think now for the first time,
• 43:13
  Speaker 1: with technologies like ours, you can have custom to order cars,
• 43:17
  Speaker 1: like cars that are like, you know, the same way
• 43:19
  Speaker 1: you choose what T shirt you wear and your T
• 43:21
  Speaker 1: shirt is different than mine. We also don't have to
• 43:23
  Speaker 1: drive the same you know model S or you know
• 43:26
  Speaker 1: Model three. We can actually have our own customized Model
• 43:29
  Speaker 1: three is and modelesses.
• 43:30
  Speaker 2: So what's the I mean is that the if you
• 43:33
  Speaker 2: think sort of long term for Makeina is like that
• 43:36
  Speaker 2: what you think about like give me the give me
• 43:38
  Speaker 2: the five year vision yeah, or ten year or whatever.
• 43:43
  Speaker 1: Yeah. So I think the long term motivation behind our
• 43:46
  Speaker 1: company is can you grant this democratization of ideas for
• 43:52
  Speaker 1: people who want to build anything? Right? Can I express
• 43:54
  Speaker 1: myself if I'm a builder, can I go build something
• 43:56
  Speaker 1: with not having to build a factory for So that's
• 43:58
  Speaker 1: really the long term goal. So I imagine in the
• 44:02
  Speaker 1: next five to ten years, you can as a designer,
• 44:07
  Speaker 1: somebody who has an idea, you can go to a website,
• 44:09
  Speaker 1: get guided through your ideas on how to make and
• 44:14
  Speaker 1: design a physical product, hit a button and say, okay,
• 44:17
  Speaker 1: I want twenty of these, and I want in Chatsworth, California,
• 44:22
  Speaker 1: and the right facility programs the right number of robots
• 44:25
  Speaker 1: to actually do those operations without any hardware or investment
• 44:29
  Speaker 1: that needs to be made for those of specific parts
• 44:31
  Speaker 1: and ship it to you two days later in the
• 44:33
  Speaker 1: right location. That is the future we're building towards cars
• 44:37
  Speaker 1: is just you know, one of the products that could
• 44:41
  Speaker 1: be built. But I imagine that you know this technology
• 44:43
  Speaker 1: or technology like these, technologies like these can be used
• 44:45
  Speaker 1: to do the myriad of designs. I think the moment
• 44:48
  Speaker 1: you open up this possibility of any designs could be
• 44:52
  Speaker 1: a reality. I think so many things will be created
• 44:54
  Speaker 1: that we're not even thinking of right now. You know
• 44:57
  Speaker 1: the fact that we have cars today and they all
• 45:00
  Speaker 1: look the same as limitation of technology. But the moment
• 45:03
  Speaker 1: you can open up this creativity of turning ideas into
• 45:07
  Speaker 1: physical reality without a without an initial investment or huge
• 45:11
  Speaker 1: barrier to entry, then I think we're going to have
• 45:13
  Speaker 1: all kinds of drones, all kinds of satellites, all kinds
• 45:15
  Speaker 1: of rockets, all kinds of cars that you're going to
• 45:18
  Speaker 1: be this like you know, Cambrian explosion of different designs
• 45:22
  Speaker 1: that's going to come into our world. And I think
• 45:23
  Speaker 1: that's what future is about. The future is about you
• 45:25
  Speaker 1: know what I call it, Like future is custom Like
• 45:27
  Speaker 1: future is about being able to make these all these
• 45:29
  Speaker 1: ideas in reality. We had this explosion happening in digital world. Yeah,
• 45:33
  Speaker 1: you know, now we have even models generating images and
• 45:37
  Speaker 1: videos and there's this you know, explosion of different ideas
• 45:41
  Speaker 1: and content being created using the technology. But the link
• 45:45
  Speaker 1: is broken to the physical world and the physical work
• 45:47
  Speaker 1: is still pretty uniform because it's very hard to make
• 45:49
  Speaker 1: things in the physical world. Can we bridge that gap?
• 45:52
  Speaker 1: Can we connect the digital world of creation to physical
• 45:56
  Speaker 1: world of creation and create the same variety in the
• 45:59
  Speaker 1: physical world as we have in the digital world. I
• 46:02
  Speaker 1: think that's the goal in our company.
• 46:06
  Speaker 2: We'll be back in a minute with the light Year Round.
• 46:17
  Speaker 2: Let's finish with the Lightning Round. Do you drive a
• 46:22
  Speaker 2: customized car?
• 46:24
  Speaker 1: I don't actually yet. Well, if I am, what have I.
• 46:27
  Speaker 2: Seen on your Instagram? What's that truck you keep posting
• 46:30
  Speaker 2: on your Instagram?
• 46:31
  Speaker 1: So I so I have a truck that's customized. I
• 46:33
  Speaker 1: don't drive it around as much, but maybe this year
• 46:36
  Speaker 1: I'll start taking it out. This year I've been you know,
• 46:38
  Speaker 1: we have been kind of stelf about it, talking about it,
• 46:41
  Speaker 1: but we haven't talked about it in a big way
• 46:43
  Speaker 1: because we have a big release coming soon.
• 46:45
  Speaker 2: I mean, you're literally posting it on Instagram. It's not
• 46:48
  Speaker 2: that stuff. Tell me about Tell me about that truck
• 46:51
  Speaker 2: you keep posting on Instagram? What's going on with that?
• 46:53
  Speaker 1: So it's a truck is fully the full body is
• 46:55
  Speaker 1: fully customized.
• 46:56
  Speaker 2: It says anvil in the back when you post it?
• 46:58
  Speaker 2: Is it called anvil?
• 46:59
  Speaker 1: Dumb question and call it and call it anvil? I
• 47:01
  Speaker 1: think it's the idea was actually the shape design of
• 47:04
  Speaker 1: it was inspired by amil. If you look at the
• 47:06
  Speaker 1: front fender, it actually looks like a the front bumper
• 47:09
  Speaker 1: looks like an anvil. But also the idea is that, like,
• 47:11
  Speaker 1: you know, we're actually forming shits on an anvil. So yeah,
• 47:14
  Speaker 1: it was very fitting.
• 47:15
  Speaker 2: Tell me about that truck, like, tell just tell me
• 47:18
  Speaker 2: what's it look like.
• 47:18
  Speaker 1: Yeah, so for example, like you know, we put a
• 47:20
  Speaker 1: lot of form and sharp edges in the in the hood. Right.
• 47:24
  Speaker 1: Most vehicles have a very hard time if you look
• 47:27
  Speaker 1: at most of the you know, hood of the vehicles,
• 47:30
  Speaker 1: they are very smooth because it's very hard to actually
• 47:32
  Speaker 1: put sharp angles in the hood. So if you look
• 47:35
  Speaker 1: at this truck, this truck has a lot of angles,
• 47:37
  Speaker 1: a lot of sharp detail right in the hood. Right,
• 47:41
  Speaker 1: And and and that's very expressive of the type of
• 47:43
  Speaker 1: person for example, that I am, right, I like things
• 47:46
  Speaker 1: that are edgy, and and that truck is certainly edgy. Right.
• 47:51
  Speaker 1: It's bare metal, right, you know, there is no blemishes
• 47:55
  Speaker 1: being hiden and hidden under the under the vehicle. You know,
• 47:59
  Speaker 1: a lot of people when cyber Truck came out, we
• 48:02
  Speaker 1: got very excited about you know, oh, it's bare metal.
• 48:05
  Speaker 1: It looks like a metal, but then there was no
• 48:06
  Speaker 1: form in it because it's actually very hard to make
• 48:08
  Speaker 1: it for metal look nice. And so that's one of
• 48:12
  Speaker 1: the things we wanted to show. We want to show that, Okay,
• 48:14
  Speaker 1: you can actually have a form metal with a lot
• 48:16
  Speaker 1: of detail in it and still keep it bare metal
• 48:18
  Speaker 1: because it will look nice. Right, So, yeah, a lot
• 48:22
  Speaker 1: of design features of it. For me kind of represents
• 48:24
  Speaker 1: the type of personality and character that I have. But
• 48:27
  Speaker 1: I think that's how every car should be. You know,
• 48:29
  Speaker 1: people should be able to have that freedom to choose
• 48:31
  Speaker 1: what their cars look like.
• 48:34
  Speaker 2: How many skull tattoos do you have?
• 48:38
  Speaker 1: I've got three? Why it's it's so, yeah, it's an
• 48:46
  Speaker 1: interesting thing. So a skull for me represents kind of
• 48:50
  Speaker 1: and it's an abstract for death of ego. So I
• 48:54
  Speaker 1: have a tattoo on my thumb which is a skull
• 48:57
  Speaker 1: that's holding a microphone to his ears. And this was
• 49:01
  Speaker 1: a time where you know, I felt like, you know,
• 49:03
  Speaker 1: I had a good platform and I could talk a
• 49:05
  Speaker 1: lot and people would listen. But then I realized I
• 49:08
  Speaker 1: should yes, that's right, but I should maybe keep them
• 49:10
  Speaker 1: keep the mic close to my ears and also listen
• 49:13
  Speaker 1: as opposed to talk all the time, right, So I think.
• 49:17
  Speaker 2: Skull microphones don't work that way for the record, but.
• 49:21
  Speaker 1: I like it as a metaphor exactly. But I think
• 49:24
  Speaker 1: the idea of is around, you know, kind of reminders
• 49:27
  Speaker 1: of you can see a lot of my tattoos on
• 49:29
  Speaker 1: my on my hands, so it's a really reminder for
• 49:31
  Speaker 1: myself to know that, you know, be present and make
• 49:36
  Speaker 1: sure that you know you're not involved with your ego
• 49:38
  Speaker 1: too much and you can see others people's perspective.
• 49:41
  Speaker 2: Is there any tension between ego death and custom cars?
• 49:46
  Speaker 1: Tension between ego death and custom cars? I don't know.
• 49:50
  Speaker 2: I'm just playing, but like you know, custom car kind
• 49:52
  Speaker 2: of seems like, hey, look at me, I'm special, and
• 49:55
  Speaker 2: ego death seems like, oh, don't look at me, I'm
• 49:58
  Speaker 2: not so special.
• 49:59
  Speaker 1: Yeah, no, I think I think the difference is I think, yeah,
• 50:01
  Speaker 1: if you have attachment to your custom card, then maybe
• 50:04
  Speaker 1: there's tension. But I more think of it in terms
• 50:07
  Speaker 1: of expression. Right. You know, you can be an artist,
• 50:10
  Speaker 1: you can. You can you can design your home the
• 50:13
  Speaker 1: way it expresses you. You can design the theme of
• 50:16
  Speaker 1: your podcast the way it expresses you. You can design
• 50:18
  Speaker 1: your car. Also, the way it expresses you. I think
• 50:21
  Speaker 1: it's leus so about oh look at me, I'm special.
• 50:23
  Speaker 1: It's more like, here's my expression to the world for
• 50:26
  Speaker 1: the people to see. But I think that expressiveness is
• 50:30
  Speaker 1: it is pretty amazing. I think that's uniquely one of
• 50:32
  Speaker 1: the unique things about humans that like, you know, we
• 50:34
  Speaker 1: we I think all we do when we come to
• 50:37
  Speaker 1: this world is expressing ourselves right, expressing uself through our work,
• 50:40
  Speaker 1: expressing through ourselves through our relationships. And if you can
• 50:45
  Speaker 1: enable people to express themselves better better, I think that's great.
• 50:49
  Speaker 1: But if you get attached to your expressions and your
• 50:51
  Speaker 1: ideas and your thoughts and think, oh, I'm better than
• 50:54
  Speaker 1: everybody else, and I think that that becomes that becomes
• 50:56
  Speaker 1: a little bit of an ego driven trip.
• 51:05
  Speaker 2: Edward Mayer is the co founder and CEO of Mocking Labs.
• 51:10
  Speaker 2: Today's show was produced by Gabriel Hunter Cheang. It was
• 51:13
  Speaker 2: edited by Lyddy jeene Kott and engineered by Sarah Bruginner.
• 51:17
  Speaker 2: You can email us at problem at Pushkin dot FM.
• 51:21
  Speaker 2: I'm Jacob Goldstein and we'll be back next week with
• 51:23
  Speaker 2: another episode of What's Your Problem.