Tesla, TSLA & the Investment World: the Perpetual Investors' Roundtable

[SunCatcher]

You're welcome.

https://twitter.com/x/status/1428765860173336582

 

[woodisgood]

The thought that Tesla might acquire Boston Dynamics misses the fact that Tesla's AI architecture is completely different from BD. The latter is essentially algorithmic. it's not based on the deep-learning neural net approach Tesla has. BD have taught their dogs to do amazing specific tricks in specific contexts. But they don't remotely have the kind of general use that Tesla has in mind. It wd be a complete distraction and confusion for Tesla to make this acquisition.

Agree. Where BD has the head-start is the mechanics of organic movement, mimicking muscle groups with pneumatics/hydraulics/servos to give the robot the dexterity to handle its tasks. This is the area in which Tesla will be playing a bit of catch-up, and hopefully a talent pipeline from the RoMeLa orbit will be established to get the right people. Of course, I’m sure we’ll also be poaching from other robotics companies like BD.

In terms of the software/AI/perception, Tesla will be by far ahead of anyone.

 

[heltok]

Humans are not energy intensive compared to robots. In general, biological systems are very energy efficient. A human consumes around 100 Watt. I doubt Tesla will build an entire robot within that power envelope.

Human labor is expensive, energy is dirt cheap: Yoda

There’s a great SMBC comic exploring the geopolitical consequences of having Superman turn a crank to provide an unlimited source of energy. We could imagine Yoda using the Force to run a similar generator. But how much power could he really supply?

Yoda’s greatest display of raw power in the original trilogy came when he lifted Luke’s X-Wing from the swamp. As far as physically moving objects around goes, this was easily the biggest expenditure of energy through the Force we saw from anyone in the trilogy.

The energy it takes to lift an object to height h is equal to the object’s mass times the force of gravity times the height it’s lifted. The X-Wing scene lets us use this to put a lower limit on Yoda's peak power output.

First we need to know how heavy the ship was. The X-Wing’s mass has never been canonically established, but its length has—12.5 meters. An F-22 is 19 meters long and weighs 19,700 kg, so scaling down from this gives an estimate for the X-Wing of about 12,000 lbs (5 metric tons).

mx=mf22×(12.519)3≈5,600kg
Next, we need to know how fast it was rising. I went over footage of the scene and timed the X-Wing's rate of ascent as it was emerging from the water.

The front landing strut rises out of the water in about three and a half seconds, and I estimated the strut to be 1.4 meters long (based on a scene in A New Hope where a crew member squeezes past it), which tells us the X-Wing was rising at 0.39 m/s.

Lastly, we need to know the strength of gravity on Dagobah. Here, I figure I’m stuck, because while sci-fi fans are obsessive, it’s not like there’s gonna be a catalog of minor geophysical characteristics for every planet visited in Star Wars. Right?

Nope. I’ve underestimated the fandom. Wookieepeedia has just such a catalog, and informs us that the surface gravity on Dagobah is 0.9g. Combining this with the X-Wing mass and lift rate gives us our peak power output:



5,600kg×0.9g×1.4 meters3.6 seconds=19.2kW
That’s enough to power a block of suburban homes. It’s also equal to about 25 horsepower, which is about the power of the motor in the electric-model Smart Car.

At current electricity prices, Yoda would be worth about $2/hour.

 

[CLK350]

The thought that Tesla might acquire Boston Dynamics misses the fact that Tesla's AI architecture is completely different from BD. The latter is essentially algorithmic. it's not based on the deep-learning neural net approach Tesla has. BD have taught their dogs to do amazing specific tricks in specific contexts. But they don't remotely have the kind of general use that Tesla has in mind. It wd be a complete distraction and confusion for Tesla to make this acquisition.

Precisely why it would be a good fit, they have complementary (not duplicative) strengths: BD has the mechanical part, Tesla has the sensors and brains.

 

[Knightshade]

For those asking why a Tesla Robot.

What does a Robot do?: A Robot takes in Visual data and clues, and audible data and clues, understand the environment and react to it. So there are two parts to it:

a) Ability to process images and sounds and understand at a fast rate. Needs to be context aware
b) Perform the physical action.

Tesla has already solved (or will be solving) the former through the FSD project. A car driving at 70 mph is in a more challenging environment than a servant robot, but essentially they are performing a very similar function.

The latter part of actually doing the physical job, is not easy but solvable problem. Tesla can get that done relative easily and it depends on the specific physical job.

If I tell my robot, 'put the red screwdriver in its place'.

a) it should know what a screwdriver is, where it is, its orientation, know where the top drawer is, how to open it, and lastly the path to get to the various end points. All of that FSD has solved.
b) The physical part of picking up, walking, opening the door and putting it in, still needs to be solved.

So Tesla is more than half way there in getting a Robot ready.

Yeah there's actually one interesting aspect in which the "car is harder" thing doesn't entirely apply.

Distance estimation.

They actually addressed this in the Q&A, where Elon said they were aiming for centimeter accuracy getting distance from vision-- because that's PLENTY sufficient in a car.

When you're braking to not hit something, you don't need mm accuracy because you should be leaving a lot more of a safety buffer than 1 mm anyway.

Even for more precise stuff like parking, 1cm is plenty close enough.

Makes sense, right?


But if I'm wanting something to do human tasks, while all of them are happening at much lower speeds than driving- I'll often want better than 1 cm precision.

If I'm off my a little under half an inch parking- who cares? Even in a Cybertruck there's more than a foot of clearance to either side of me in a standard parking spot.


But if I want my RoboButler to pour some expensive champagne into a flute, I'd prefer he not miss the glass.



There's solutions for this... For example since the bot isn't needing to make split second decisions- and can move its head (cameras)- it ought to be capable of a lot more precision than the car is in the long run... but it's not QUITE as simple as "just use FSD"

 

[StealthP3D]

The thought that Tesla might acquire Boston Dynamics misses the fact that Tesla's AI architecture is completely different from BD. The latter is essentially algorithmic. it's not based on the deep-learning neural net approach Tesla has. BD have taught their dogs to do amazing specific tricks in specific contexts. But they don't remotely have the kind of general use that Tesla has in mind. It wd be a complete distraction and confusion for Tesla to make this acquisition.

I actually think the impressive achievements of Boston Dynamics might have been the impetus for Elon to go into humanoid robotics. Because the deeper he looked, the less impressed he was. First principles thinking can be applied to any endeavor to great effect.

 

[LiveLong&Profit]

All these elonflops are over my head. I need a reference point. So this exagigacabinetflop chip thing. It’s like really good right? So is Nvda our new competitor or is it Amazon?

Well, in the extreme version of TSLA-optimisn, no one really is anymore...

• What is the TAM of steady, stable labour needing a large component of physical work?
• Who can mass produce sturdy non-dangerous, versatile, sort-of-cheap - and cool robots?
• What is a good chunk of the world market for human labour with a major physical component worth in 20 years?

 

[Cosmacelf]

As predicted, the street considers any kind of android announcement to be a distraction for the company.

Yeah, a 70,000+ employee company can't possibly devote a small team to advanced R&D! Yikes...

 

[Electroman]

Since I can barely spell AI, I have been looking for some expert reactions on twitter. Here are a few. Do feel free to add any others you might come across, good or bad. Hopefully this can add more signal and less noise to the conversation.

First is from Lex Fridman:

https://twitter.com/x/status/1428540411807666178

A few from James Douma, who Dave lee interviewed a few times. I suppose this means his mind is blown, in a good way.

https://twitter.com/x/status/1428554155077951490

https://twitter.com/x/status/1428540794026090500

This guy green who i thought would have some comment on the new information, just has a long rant on tesla dropping radar w/ factchecking. Sad.

https://twitter.com/greentheonly

The autonomous FSD Forum has been weirdly silent. Not much on reddit machine learning or on the hacker news forums yet. Any other hot takes you folks are seeing?

As I have said many times, that green guy is a drama queen. He seeks attention. Take a grain of truth and blow it up. He is an half-empty-glass guy. Knowledgeable, but he thinks he knows better. Guys like him are looking for opportunities to say, 'ha ha see, I told ya' even if that stuff is trivial in the end.

 

[LiveLong&Profit]

When/if there is a mars colony, it'll probably be >10 robots for every one person.

This.
The first couple of Starships will be loaded up with digging equipment - and Tesla Bots.

 

[TheTalkingMule]

Options premiums are starting to drop a bit finally. In for some 8/27 calls, will double down if we open down Monday. I think there was some actual buying demand today that will need to be unwound by MM's in the next couple sessions.

 

[Cosmacelf]

Tesla AI Training as a Service

Yes, this is like Amazon offering AWS. Tesla's AI cloud offering is much better than anyone right now, so it should grab cloud AI market share. Nice side business that can grow quite large as AI grows.

 

[22522]

Looks to me you're applying for a job at Tesla Robotics. Go for it!

Thanks.

It is an interesting problem and it is good practice to think along with Elon.

I am really motivated make everything easy about this robot (even controls) so it gets done on time and costs almost nothing to make, with 100% yields.

I think the hiring process is specialist silos - save Elon. Pretty sure Elon could crank out a good robot design in 2 weeks. And I proposed architectural decisions, rightly Elon's decisions. (the position is filled!)

... Does anyone at Tesla even know who Johnny Quest is?

Thanks again.

 

[Todd Burch]

Did I hear right in the presentation that CT will debut with HW4 in a year? I thought we are closer to CT production than that...

You heard that CT will debut WITH HW4 in about a year. Not that CT will debut in about a year.

 

[Unpilot]

The US military might be interested in them having US ownership. [Remember that Tesla is a US company...]

Somebody should tell team Biden.

 

[LiveLong&Profit]

my thoughts for what it is worth
• The self driving portion was the best. The level and quality of simulation is impressive
• Dojo has amazing specs, but it isn't as far along as I thought it was
• The dancing "robot" was terrible and I can't believe anyone thought it was a good idea to do that
• I don't get the point of Tesla Bot and how that furthers the mission of sustainable energy and transportation

I liked the dancing fake-bot.
It is like the Cybertruck broken window.
It gives detractors a fig leaf to not take this seriously.

Prediction
Next year: The prototype robot does simple dance moves - not impressive, but still - not limited to walking and standing.
Year after: Tesla Bot v2 does dancing fairly well - not too clumsy.
...
Year N: Tesla bot does a carbon copy of 2021 dancing sequence *exactly*. Down to the millisecond. Frame by frame.

 

[Krugerrand]

Amazing that wallstreet doesnt think that a massive company cant handle multiple projects at once. How about ICE manufacturers being distracted with ICE business with company like Tesla taking away market share of cars.

This what I think; Wallstreet has never seen anything like Tesla. Nobody on Wallstreet ever had a professor who’s seen anything like Tesla. Wallstreet has always tried to pigeon hole Tesla into ‘car mfging co’. Clearly Tesla hasn’t been another GM, Ford, VW, etc… since ever. But because they mostly make cars (for the time being) the thought process is, ‘they’re a car co. and so they are evaluated by that metric.

Excuse the stereotyping, but it exists because there is much truth in it. Wallstreet is made up of a certain type of person with certain tendencies, personality traits etc…. They all went to the same schools, they all studied the same courses, they are all programmed to think a particular way.

Don’t argue with me about outliers. I’ll concede the point that there’s always an example that doesn’t fit the stereotype. But they can be ignored because it’s not the norm. Show me an accountant who is comfortable ball-parking numbers and I’ll show you a Wallstreet analyst who doesn’t get their eyebrows waxed.

So, no. There is zero chance a car company can also produce humanoid robots. No chance in pumpkin pie.

 

[Prunesquallor]

"Boring, repetitive, and dangerous"

Yeah?

That is too contrived. There is one aspect of work that is worse than all of those..even the "dangerous" aspect.
DEMEANING/Degrading.
Only one time have I seen a person whose job was to pick up litter not be in a demeaning position..well that and every street walker. (Luke I think you just pointed out why demeaning was not stated....)

It may be worth (re)watching Star Trek TNG “The Measure of a Man”.

 

[printf42]

Not sure I agree with this -- Elon made sure to emphasize that, even with HW3 + FSD computer, he is confident that Tesla cars will be able to run FSD, at up to 300% better than a human (He said HW4 with the upgraded cameras would be like 10x better than a human). I believe he is making sure that people understand that FSD is deliverable with HW3, just at a safety rate lower than HW4. This implies to me that cars with HW3 will have to pay if they want HW4 installed.

I think so too, HW4 retrofit would come with a cost, uncertain whether camera replacement would be needed too.

But I am quite sure if you delay car purchases to wait for HW4, FSD package price would rise much more than the cost of retrofits.

 

[Cosmacelf]

BTW, don't mean to throw cold water on Tesla's AI potential, but there is a whole class of AI that Tesla (and everyone else for that matter) is NOT doing, and that is real time learning.

Tesla's AI is the "standard" system that trains a huge Neural net for perception and maybe (eventually) end to end control (meaning the network both perceives the world and then interacts with it - right now today, the car is not driven by a neural net, it is driven by complex C++ algorithms. They have a plan towards getting to end to end control). What this means is that the car/robot isn't very flexible for truly novel situations. You have to train the robot for whatever you think it'll see in the world before it can handle it.

Not only that, but it can't learn from novel situations. So the robot/car will only "learn" via monthly or whatever software updates. It can't learn on the spot through generalization.

Does this matter? Well, people will run up against the robot's limitations pretty fast. Kinda like Alexa is fine, but there a real limit to what it can do.

Real time learning is a completely different neural network architecture and one that researchers are only barely exploring right now. It'll be interesting to see how far Tesla and the rest of the industry goes before/if hitting a brick wall of diminishing returns before having to actually tackle it.