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

[stopcrazypp]

I was thinking, if every company is HDmapping their own solution and no one is sharing or merging these data because they are competitors, then how will any of these companies make any money when the operating cost is so high?

Mobileye is claiming their method can scale because they are crowdsourcing to vehicles with their chips and cameras installed. There are already millions of cars using their system.
Mobileye REM™ - Road Experience Management

Say cruise AND waymo are all HD mapping, geofencing SF right now. So now they have to share the same amount of customer base but HD mapped the same place twice, and has 2x the safety operators and stuff. Just because you HDmapped 2x doesn't mean you get 2x the customer base. This entire business model is equivalent to putting money in a furnace.

Those players (specifically Waymo) is claiming mapping will be easy and inexpensive. There's a tweet that claims with 100 vehicles, it will only take 66 days to map the entire 4 million miles of roads in the US and only cost a few million.
Autonomous Car Progress
I questioned that, given GM is doing things much slower for a map that is only used for L2 and mainly on highways (where presumably there are much less features per mile and you can travel much faster so would presumably take much less resources to map per mile than local city roads for example).

 

[Buckminster]

https://twitter.com/x/status/1568810453362409472

 

[dhanson865]

https://twitter.com/x/status/1567872016832774145

Is this the first L4 on public roads in the US?

For the people that marked this funny, that is pretty much how I mean it (a bit of sarcasm)

For the people that didn't read it as funny. Oh well. Maybe my softball wasn't over the plate.

 

[Singuy]

Mobileye is claiming their method can scale because they are crowdsourcing to vehicles with their chips and cameras installed. There are already millions of cars using their system.
Mobileye REM™ - Road Experience Management

Those players (specifically Waymo) is claiming mapping will be easy and inexpensive. There's a tweet that claims with 100 vehicles, it will only take 66 days to map the entire 4 million miles of roads in the US and only cost a few million.
Autonomous Car Progress
I questioned that, given GM is doing things much slower for a map that is only used for L2 and mainly on highways (where presumably there are much less features per mile and you can travel much faster so would presumably take much less resources to map per mile than local city roads for example).

Aren't Lidar and HDmapping used for localization? Do we have mobile eye demonstrating FSD type drives without those two things and just chips plus cameras?

 

[AudubonB]

Couldn't edit post Chernobyl disaster as we were traveling on I5 north by the mountain fire but seeing Teslas being shipped north.

Chernobyl disaster - Wikipedia

en.wikipedia.org

No worries. It seems none caught the Jean-Claude Killy reference, anyway.
We had guests this season from Grenoble and the husband was beside himself when I pulled out the name of the city’s most famous (recent, anyway) son.

 

[Artful Dodger]

Japan Fukushima nuclear disaster, Grenoble and 3 mile island)?

I was with you until you said Grenoble. Personally, I favour the Olympic movement...

Cheers!

 

[heltok]

As someone who has worked with Mobileye, they have a tendency to come across as very confident and promise a lot. Imo they seem a bit to phd:y for my taste, presentations showing too many equations etc. One time they claimed to have mathematical proof that their solution was needed when there are clearly examples of competitors having different implementations up and running. A bit humbleness wouldn't hurt.

I recently watched a video about some state of the art localization. I used to work with doing localization and maps, so it was fun to see what people are up to today. To be honest I am not impressed, results were not much better than my old results and the system felt like too much feature engineering and clever classical algorithms rather than the modern approach of lots of data data, tons of compute and let a model learn how to do it.

But anyway, it validated how I view the problem. If you are gonna use HD maps, you need to localize yourself in the HD map. Either you add another layer of localization data(landmarks) or you try to use the basic road geometery to position yourself directly into that or a combination of both. And it seems to work pretty well with not too much extra data. But then you run into all those corner cases where it breaks for some strange reason. Then you either feature engineer a better solution and see where that breaks etc or you are back to doing what Elon calls "solving real world AI". And once you have done that you don't need localization or HD maps anyway.

It seems that Tesla have started to feed some kind of non-HD map into the neural network. Probably this will be enough after a few cranks of the data engine. Think of the neural network as being a very intelligent person who gets to see the last 30 seconds of video, odometry, GPS, compass etc and also gets to see a very basic high level map of the intersection. The intelligent person can figure out the rest. Maybe not right now, but with enough training, optimizations and bug fixes I believe they can.

The race for Robotaxi and ADAS will be interesting. We have Tesla running ahead being willing to break their system and iterating quickly. We have Waymo taking the slow and safe approach. We have Mobileye and their customers adding some features here and there to different models and demoing LVL4 capable systems. And then we have the Chinese, who don't get enough attention. They do really cool stuff, but are not super trust worthy and their culture is so different that it can be hard to understand where they are at. But clearly the Chinese are very serious, look at Waymo competions, every category is won, second and third place by some Chinese group.

(scroll through the video for every slide with that light green table)

Lots of companies there doing some cool stuff:

If you want to learn more about AI in China this book is a must read:

Amazon.com

Anyway, it's a fun world to observe, so many teams making lots of progress. It seems likely that someone will succeed at something in the next few years. Then the question will be how far behind the others will be.

 

[stopcrazypp]

Aren't Lidar and HDmapping used for localization? Do we have mobile eye demonstrating FSD type drives without those two things and just chips plus cameras?

To clarify, Mobileye is doing the mapping using only chips and cameras. Their L4 vehicles will have lidar though.

They have a bunch of videos on their YouTube channel.

https://youtube.com/c/Mobileye

The longest recent video is here for their L4 system:

 

[Surfer of Life]

Don't you think the $5B Cap Raise that Tesla announced on Sep 1st, 2020 had something to do with TSLA action?

Tesla to sell up to $5 billion in stock amid its incredible rally - CNBC (2020-09-01)​

Here's the Pre-market chart from Sep 1st, 2020. Any guess what time Tesla's announced the Cap Raise?

View attachment 851281

So sure, TSLA Closed near $475 on Sep 1st, but it was down -11.4% intraday from $536 post-split (also an all-time high) in the Pre-market just before Tesla announced the Cap Raise. That's a significant move, and it bottomed out at $325 just 4 days later. That's a 40% drop!

Tesla did not recover to $533/share until after Nov 16 at 500 p.m. (when S&P announced that they would add TSLA to the S&P 500):

Tesla Set to Join S&P 500 - Nov 16, 2020 - press.spglobal.com

View attachment 851285

So the evidence suggests that TSLA was heavily manipulated, and that only forced buying by large players (MMs, brokers, Institutional Investors) leads to significant step-changes in TSLA SP.

We've seen this play out again and again over the years (indeed, each H1 in the past 6 years), like some kind of staged, scripted drama, performed in an ancient, drawn-out, and ritualistic manner... it's like...

Kabuki.

We've seen this play out again and again, but hasn’t the situation matured to the point this order of magnitude of forced buying and corresponding upward TSLA price rocket - many of the shorts having learned their lesson - is unlikely ever again?

 

[lascavarian]

Don't you think the $5B Cap Raise that Tesla announced on Sep 1st, 2020 had something to do with TSLA action?

Absolutely, what a wild ride it was! And with lots of Kabuki. Looking over my notes from 2020 brought back a lot of memories as your contributions confirm. It was a wild time and I only capture closing prices in my notes.

2020 vs 2022 are different but my observation is that price pressures have some similarities nevertheless. 2022 seems to show a stronger stock in the way those pressures are shrugged off.

 

[lascavarian]

As Semi is coming in a few months, there will be an infrastructure required.

They will break and need fixed. Will rangers add this to their duties or will it be something new. EMs recent tweets seem to reflect vehicle service is getting a fresh look in North America.

 

[Sancho]

It totally makes sense to me that Tesla is working on a VR/AR headset. Elon calls the shots on Tesla investments in special projects, and Elon is a gamer.

Although Elon talks a good game about transitioning the world to sustainable energy, and mankind becoming a multiplanetary species, sometimes I think he’d ditch it all to create the ultimate video game experience.

Elon is the world’s greatest entrepreneurial engineer, and a perpetual 12 year old boy. That unique combination screams VR/AR headset.

 

[SN_8]

We've seen this play out again and again, but hasn’t the situation matured to the point this order of magnitude of forced buying and corresponding upward TSLA price rocket - many of the shorts having learned their lesson - is unlikely ever again?

You'd think so, wouldn't you? But @The Accountant 's chart shows that it is still a possibility that history repeats itself.

 

[CaliBear1]

It totally makes sense to me that Tesla is working on a VR/AR headset. Elon calls the shots on Tesla investments in special projects, and Elon is a gamer.

Although Elon talks a good game about transitioning the world to sustainable energy, and mankind becoming a multiplanetary species, sometimes I think he’d ditch it all to create the ultimate video game experience.

Elon is the world’s greatest entrepreneurial engineer, and a perpetual 12 year old boy. That unique combination screams VR/AR headset.

A VR/AR headset will be a critical component for long transits to Mars and beyond. Think surgeries and complex repairs. And, perhaps, a gateway for neural networking. So this is well within the Mission parameters.

 

[jerry33]

You'd think so, wouldn't you? But @The Accountant 's chart shows that it is still a possibility that history repeats itself.

Short selling is similar to the LeBrea tar pits. It appears there is a free meal by jumping on the animals stuck in the tar, but all to often the predator (short seller) gets stuck too.

 

[The Accountant]

Looks like Tesla may have stopped doing "the Wave", meaning Q3 deliveries could be lower than normal quarterly production as more of the Q3 production will be delivered in Q4.

https://twitter.com/x/status/1568639215663030272

Mr Miserable who tracks ships to Europe believes that this late departing ship does not have Teslas on board.

He has a few posts on this; here is one HERE

excerpt from one of his posts:
"GRAND VENUS is alongside Shanghai South. She is not Tesla related as @Tentomushi found a schedule that has her heading for Bristol (probably MGs). Congestion at the main Shanghai terminal at Haitong may be causing the uptick in Shanghai South usage. There are 12 RoRos waiting to dock in Shanghai at the moment".

 

[winfield100]

Virtual grid works and pays $2/kW. The fake news that EVs would stress the grid are the biggest bull stink I've heard. Having Power walls might pay for themselves. This is just the beginning.
Another item that I think I disagree with Elon about is nuclear power plants. San Onofre was going to be shut down and people cried that San Diego would have blackouts but NOT TRUE because many people installed roof top solar which worked brilliantly. The Utilities are really not public IMHO. BTW the Nuclear Reactor that was to be shut down is on an earthquake fault. Does anyone realize the long term damages to the area and costs(Japan Fukushima nuclear disaster , Grenoble and 3 mile island)?
The Climate crisis is getting worse and people still think they are entitled to burn up the world. Now would be the time to stop this suicidal progress.

Experts: Electric vehicles could help, not hurt, California's energy grid

“EVs not only pull from the grid but have a humungous potential to put energy back on the grid. It’s something we don’t actively talk about because it’s in a piloting stage,” the vice chair of the California Energy Commission said. “But when we get to that point, there could be a net benefit...

www.kcra.com

@MSMike
if you extrapolate out to 2035 or so, using published data , nukes have a growth rate of roughly 0.5%/year, PV has a growth rate of around a bit over 22%
nukes have been pretty flat the last 20 years.
(University of Virginia had a small reactor on campus, at least in the early 1960's and if there was an "accident" you could dive thru a pool of water to escape the containment building, (yeah)) (what is that blue light? Cherenkov radiation)
plus about 9 years to build a nuclear plant, plus fuel
plus Diablo Canyon Nuke in California is on a major fault, plus South Anna nuke in Virginia is, surprise, near a fault that was discovered when it shook a few years back, etc, etc
edit:
plus as i vaguely recall Chernobyl generated over 100 gigawatts in under 2 weeks when it melted, and one of the reasons it "broke" were bored technicians fooling around in the middle of the night
edit: (for the Americans)
"what melts in the ground and not in your hand?
"Hershey, Pennsylvania" (3 Mile Island nuke reference)
(down the river is Baltimore, Washington DC and Norfolk/Newport News, but it _should_ be diluted by then)

just nuclear and solar in terawatt hours actual generation

Downloads | Energy economics | Home

Use this section to download the PDF edition of the Statistical Review of World Energy, the Excel data – and more

www.bp.com

 

[dm28997]

@MSMike
if you extrapolate out to 2035 or so, using published data , nukes have a growth rate of roughly 0.5%/year, PV has a growth rate of around a bit over 22%
nukes have been pretty flat the last 20 years.
(University of Virginia had a small reactor on campus, at least in the early 1960's and if there was an "accident" you could dive thru a pool of water to escape the containment building, (yeah)) (what is that blue light? Cherenkov radiation)
plus about 9 years to build a nuclear plant, plus fuel
plus Diablo Canyon Nuke in California is on a major fault, plus South Anna nuke in Virginia is, surprise, near a fault that was discovered when it shook a few years back, etc, etc
edit:
plus as i vaguely recall Chernobyl generated over 100 gigawatts in under 2 weeks when it melted, and one of the reasons it "broke" were bored technicians fooling around in the middle of the night

just nuclear and solar in terawatt hours actual generation
View attachment 851338

Downloads | Energy economics | Home

Use this section to download the PDF edition of the Statistical Review of World Energy, the Excel data – and more

www.bp.com

Great points. Love the extrapolation comparison and details.
Just one other issue is disposal and costs of spent nuclear fuel besides the environmental issues. I got a good reminder to use blue light blocking glasses

The steep costs of nuclear waste in the U.S.

earth.stanford.edu

 

[Usain]

John Mc Elroy speaks on how legacy OEMs are on "the Eve of Disruption" due to their archaic organizational strategies.

This would be a good video to share with anyone curious about discovering more of what it is that sets Tesla apart from the rest of the pack.

I was just about to post this video myself. It may be the best, most succinct, explanation of Tesla's advantage over legacy auto. It's well worth 7 minutes of your time.

But rather than focus on the organizational structure, I was going to point out what McElroy calls the "software-defined" vehicle. A lot of people have described a Tesla as a "computer on wheels". I think they say this, half jokingly, because of the big screen. But the "computer on wheels" concept is more true than most people realize. See the video for a really good explanation of the "software-defined" vehicle.

This is the basic reason a Tesla is superior to anything legacy auto has created. A Tesla is a computer first and a vehicle second. Correspondingly, a smart phone is a computer first and a phone second. Legacy auto gets it backward, which leads to difficulty in solving simple problems like over the air updates.

A "software-defined" vehicle is just a fundamentally better car.

 

[ZachF]

It totally makes sense to me that Tesla is working on a VR/AR headset. Elon calls the shots on Tesla investments in special projects, and Elon is a gamer.

Although Elon talks a good game about transitioning the world to sustainable energy, and mankind becoming a multiplanetary species, sometimes I think he’d ditch it all to create the ultimate video game experience.

Elon is the world’s greatest entrepreneurial engineer, and a perpetual 12 year old boy. That unique combination screams VR/AR headset.

VR will be helpful for teleoperation of both cars and bots.