Category
🤖
TechTranscript
00:00So, Tesla just announced a whole human-sized, human-shaped, sentient robot.
00:11This is not a drill, this is a thing that really just happened, and a lot of things
00:16come to mind when hearing that, so let's get into it.
00:20Now, the two things that might have come to your mind when hearing that, like me, are
00:23number one, why Tesla, of all companies, and number two, why a robot human, of all things.
00:32But if you frame it a certain way, it actually can make a lot of sense, but you gotta break
00:36a few things down first to get there.
00:38So let's get there.
00:39So, Tesla.
00:40Tesla is a car company, mostly.
00:42You know, I've reviewed their cars, we've seen their announcements, their factories,
00:47maybe we'll call it a vehicle company, they have a truck, they have a bunch of other stuff,
00:50but they have a factory that makes cars.
00:52So they're a car company, but only by that specific definition.
00:57It's actually pretty common to look at Tesla as a software company, just because their
01:01software advantage and all the things that they do with the software in their cars is
01:06so important to them.
01:07It's like they're a software company that happens to make electric cars.
01:11And so, when you compare to the Mercedes of the world, or the Audis of the world, or BMWs,
01:17I would never describe those companies as software companies, but for Tesla, that's
01:23where their main advantage is.
01:24And so part of this software thing that they're doing is this self-driving project, okay?
01:30Now aside from all the massive missed time windows and over-optimistic promises and deadlines
01:36and things like that, their self-driving is currently, as it is today, some of the most
01:41advanced in the world.
01:42And it's not because of high-resolution mapping or predetermined routes.
01:47It's because Teslas are using their sensors, or mostly cameras at this point, and constantly
01:53scanning their environments and making thousands of little on-the-fly decisions in real time.
01:59And then all of the miles driven with this autopilot in action is used to upload the
02:04data back to their servers and train the whole rest of the fleet.
02:07So this whole system is a constantly self-improving thing.
02:12The bigger it gets, the faster it can continue to get better.
02:15So this requires a lot of computing, a lot of horsepower, to be frank, and a lot of smart
02:20code, and that's what Tesla's been really good at.
02:23So Tesla had their AI day a couple days ago.
02:26You might not have heard about it, but they do this every year.
02:28They live stream a bunch of talks from Tesla engineers about computer vision and how their
02:33neural networks have evolved over time, and explaining their focus on making autonomous
02:38cars that rely on computer vision and their synthetic visual cortex.
02:42It's really great.
02:43It's really informative, complex stuff.
02:45And at this year's, they unveiled their new custom D1 chip, which is this chip that's
02:49designed and built entirely in-house by a car company, which is specifically made to
02:55train Tesla's full self-driving algorithms based on driving footage even faster.
03:00And they're basically going to be replacing all of their GPUs they're currently using
03:04to make this supercomputer designed to do computer vision machine learning.
03:08It's big stuff.
03:09At some point during all of this, Elon's on stage talking, and he says something about
03:15to the effect of Tesla is basically a robotics company.
03:18Tesla is arguably the world's biggest robotics company, because our cars are like semi-sentient
03:24robots on wheels.
03:25And if you think about it, it's like, okay, yeah, I could see that.
03:29I mean, obviously Tesla builds cars, so we'll call them a car company still.
03:34But like, if it's Tesla's AI day, and they're giving a presentation, and they're trying
03:38to make as good of a show as possible for recruiting purposes, they're going to call
03:42it robots.
03:44So they happen to make robots that are on wheels.
03:48But yeah, that's how you can get it to make sense mentally.
03:50It's like, all right, yeah, self-driving cars is just one application of AI and computer
03:58vision and robotics, just to solve that problem.
04:02So that's what's happening.
04:03It's a bunch of sensors scanning its environment and making a bunch of decisions based on that.
04:08And then training future versions of it based on what happens.
04:12That's just what's happening in self-driving cars.
04:14So that I can see making sense.
04:16Then Elon, right afterward, said this.
04:20It kind of makes sense to put that onto a humanoid form.
04:23He says it makes sense to make these robots in a humanoid form.
04:29And that is where I disagree.
04:31So this is the newly unveiled Tesla robot.
04:35It's called TeslaBot.
04:37It's a five foot eight, 125 pound, sleek looking, black and white, human shaped robot with this
04:45face mask with a screen on it.
04:47It's got hands with full 10 fingers.
04:50It's got feet, but with no toes.
04:52It's pretty slow.
04:53It has a max speed of five miles per hour so that you can outrun it.
04:56And then it can lift things up, up to 10 pounds with the arms extended, or it can deadlift
05:00150.
05:01Lots of cameras and actuators, and of course, the full self-driving computer inside.
05:06So they're saying it's going to be easy to overpower or run away from.
05:10So that should never be a concern.
05:13But the main purpose, the purpose of this TeslaBot, the thing that they kept saying
05:17over and over again on stage during the presentation and during Q&As when they're asked about it,
05:22they kept defaulting to, it will be able to do tasks that are repetitive, dangerous, or
05:28boring so that humans don't have to.
05:30And so if you buy Tesla as a robotics company, then yeah, this makes sense.
05:37It's just another robot that will use the sensors to scan its environment and make a
05:43bunch of decisions and navigate around this world that it's in.
05:47But the thing that I have to say, and this is going to come with the caveat that I'm
05:52of course not a robotics engineer, I'm not an expert at this, but here's my take.
05:58The human form is very inefficient.
06:01Now don't get me wrong, humans, we're pretty great, right?
06:04We're a very accomplished species and we're very complex and we've done a lot of things
06:09well, but they work as a body so well because number one, our brains are amazing, but number
06:15two, because we've built this world around us that we live in mostly with ergonomics
06:21designed for humans.
06:23But when it's not designed for humans, like this upright bipedal thing that's balancing
06:28all the time, it's not very efficient.
06:30We just did a retro tech episode about this, about humanoid robots.
06:35So I'll link it below if you haven't already watched it.
06:38It's a super fun one, but we did a lot of talking with people and experts about this
06:43and it turns out if you want to make a robot to take over a task, you typically don't want
06:49to build it like a human.
06:51You want to build it for a single task and make it as efficient as possible at that one
06:56task.
06:57So pop culture forever has been obsessed with humanoid robots.
07:01Like I can't even count all the instances of humanoid shaped robots in movies and TV
07:06shows over the years that we found.
07:09It's something we've been fascinated with since the beginning of time.
07:11There's endless versions of it, but it turns out the ideal way to make a robot vacuum the
07:15floor for you is not to build a human sized shaped robot that pushes around a vacuum.
07:21It's to make the robot the vacuum.
07:24The vacuum is the robot.
07:26You don't want a humanoid robot standing at the sink, washing dishes for you.
07:30The dishwasher is the robot, like that's the single purpose built robot.
07:35Tesla knows this.
07:36You don't have a humanoid shaped robot sitting in your car, driving the car for you.
07:40The car is the robot.
07:42The self-driving car that drives itself has already been done.
07:46So the list of things you would want a human shaped robot for is pretty small because labor
07:53is typically not designed around the human form.
07:57The one example they kept giving over and over in their AI day presentation was grocery
08:03shopping.
08:04So they, you know, you walk to the grocery store, you grab a cart or a basket and you
08:09pick out some items and put them in the basket.
08:12I guess that's, you know, you need a bunch of different peripherals and things to pull
08:16that off.
08:17And maybe a human shaped robot would be fine for that.
08:20But there's not a whole lot else I can think of.
08:22Like even when I picture the most simple, boring tasks, like making a bed or building
08:29a table, like when I picture a robot that would be especially good at that, I don't
08:34picture a human shaped one.
08:35I picture one that's super focused and super efficient.
08:38Maybe I'm just short-sighted, but I remember when I talked with Elon at the factory a couple
08:42years ago about all the things that they do with robots versus the couple of things left
08:47that they still do with humans, he had a really interesting answer.
08:51So some parts of it are like 80 to 90% automated.
08:56And then some parts of it are like only 10 to 20% automated.
09:00What are those parts that humans do better than?
09:03Humans are really good at adaptation and rapid evolution and like doing finicky things.
09:13It's like trying to connect a hose that's sort of dangling around and the robot's got
09:21to find the hose, grab it, then connect it to another hose.
09:25At that point.
09:26It's like really hard.
09:27Like a person can just go, oh, done.
09:29Gotcha.
09:30That makes a lot of sense.
09:31Yeah.
09:32And then when you see it, it's like, wow, it's super, super obvious.
09:35And then we try to have robots do this, and it's like robots grabbing the wrong thing
09:38and like trying to stick it over here, and it's like, oh, the hose was here when the
09:42robot thought it was here, and so now it tries to grab air and then smashes into the car.
09:48You don't want that.
09:49Yeah.
09:50It's a comedy of errors.
09:51A tragedy of errors.
09:52So yeah, all that really made me think.
09:54With as good as Tesla is with self-driving, why would you want a humanoid, a human-shaped
10:00robot to try to do things that humans are already good at?
10:03How long is it going to take for that robot to actually get good at things that humans
10:08are good at?
10:09And you know what?
10:10The best way to answer that would probably be to look at the most advanced robotics company
10:14that I can think of at the moment, which is Boston Dynamics.
10:17So there's like, there's two very different types of humanoid robots that I've seen exist
10:22in this world.
10:23On one end, it's like the dumb single purpose thing you see at CES.
10:28Like these are the barely functional robots you see that can really do only a few things
10:33decently well.
10:34They're kind of cheesy looking.
10:36They're human shaped for no reason, but they exist.
10:39And then on the other end of the spectrum, there's Boston Dynamics Atlas.
10:44And man, I wish when I was out there, I could have seen Atlas in person because just the
10:48dog spot is already impressive enough.
10:51Like it's got the sensors.
10:53It has a basic level of understanding, scanning its environment, and it can be programmed
10:57to do dangerous or boring or repetitive things and go places that humans wouldn't go.
11:02And we say dog because they've called it spot.
11:04And it's kind of looks like a dog if you make it behave the right way sometimes.
11:08But really, it's just a quadruped any four legged animal would do.
11:12But Atlas, on the other hand, well, well, I haven't necessarily seen it do anything
11:18specifically useful yet.
11:21But seeing that five foot human shaped robot navigate the world in crazy ways is something
11:28else.
11:29I mean, I did share a new video of it doing parkour and Elon seemed very impressed.
11:33But even after all these years of development and the history of how much we've seen it
11:37get better, Atlas is not a robot that's going to be on sale for people to buy anytime soon.
11:44And if you ask them about it, they'll tell you it's just them experimenting with their
11:48computing in a new form factor.
11:49And so at the end of the day, my take Tesla bot probably won't exist anytime soon either.
11:55You know, I know they said on stage there would be a prototype maybe next year.
11:59I I kind of doubt that they're going to leapfrog Boston Dynamics in the next two years.
12:06But then here's the other thing, even if they do come out with it, Tesla right now is training
12:11their self driving computers with billions and billions of miles of self driving happening
12:16on the streets today.
12:18Right.
12:19And the best way to get as much data about that as possible is to have it go out in the
12:23real world and do it.
12:25It's kind of genius, actually, that they're making so many cars and people are really
12:28using it for them.
12:29But now they're going to be asking for a humanoid robot to go out and navigate the entire rest
12:35of the real world where there's an order of magnitude, more stuff to identify and react
12:41to.
12:42Like, I mean, on the streets, there's obviously other cars and trucks and buses, and it can
12:46identify motorcycles and bicycles and pedestrians even.
12:50And then there's traffic cones and there's lights and all the rest of the traffic control
12:54devices, construction zones, arrows on the road is kind of starting to run out.
13:00That's almost it.
13:01But how well is the Tesla bot going to do when you ask it to go get bananas from the
13:05grocery store?
13:07Like really?
13:08Like how good could it get?
13:09Or if you ask it, hey, Tesla bot, go grab my AirPods Max from downstairs.
13:15Is it going to be able to know that that's headphones or is it going to have to have
13:18a model of every new pair of headphones that ever comes out to be able to identify which
13:23one you're talking about?
13:24And all of this training, that's you can see where I'm coming from.
13:27All this training has to come from real world data, which I guess would have to come from
13:33people using Tesla bots in their homes and starting to get this going, which seems a
13:37little less likely to me than people getting a self-driving car.
13:41And then there's just a whole bunch of other unknowns about this robot, too.
13:45What other colors will it come in?
13:46Will it come in matte black?
13:48How does it charge?
13:50Will it just sit on a USB port in the corner of a room every day?
13:53I don't know.
13:54Will this ever even go on sale?
13:56And I don't know if it's really even about this robot going on sale as much as it is
14:02just about, honestly, it could really just be Tesla exploring their computer vision in
14:09another form factor, just like Boston Dynamics.
14:11It's just about Tesla becoming the best AI and computer vision company that they possibly
14:17can.
14:18And this being just one more avenue to get better at it.
14:21So there's no good reason I can think of for a humanoid robot to exist out there in people's
14:28hands anyway.
14:30But hey, it makes a lot of good headlines.
14:33It's obviously something we've been fascinated with forever and will continue to be for a
14:36long time.
14:38And it could just be one more form factor that Tesla's exploring.
14:43And also, when have they ever said no to a good free PR opportunity?
14:45So anyway, that's about my thoughts on it.
14:47I'm super curious what you guys think.
14:49If you have any thoughts on, would you want something like this in your own home?
14:52Do you think the robot will exist?
14:54I'm sure the comment section will be fun below this one.
14:57Either way, thanks for watching.
14:59Catch you guys in the next one.
15:01Peace.