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

[mongo]

This is the first time we have an official confirmation that FSD is actually shadowing the driver. I am pretty confident it wasn't the case until very recently. Great stuff!
View attachment 1036245

Your conclusion does not follow from the data presented.
Tesla FSD is already known to be trained at Tesla's server farm with video clips uploaded from cars driven by humans. That doesn't mean the NN is trained in the car (which the hardware does not support).
Inference compute != training compute.

 

[unk45]

The idea comes from a photo of a car model in the Issacson book.

For trips for more than 2 passengers existing Models 3/Y,S/X can do the job. As there is no need for a driver, there is an extra seat available.

Statistics probably show that most taxi trips are for 1-2 people. What is needed is comfortable seating for 1-2 adult passengers and adequate luggage space.

As there are no rear passengers and no need to drive the vehicle, the seats don't need to move backwards and forwards...

There is no need for a centre console so the Robotaxi can have narrow frontal area and hence reduced drag, on roads and in boring tunnels a smaller size allow more vehicles per mile of roadway.

I think it is likely that the 25K compact car is a stretch/ 4 seat version of the Robotaxi. It probably has less luggage space but it seats more people.

When booking via the app the number of passengers and the amount of luggage can easily be specified in advance via the app, so the right type of vehicle can turn up.

In future there is probably a van coming and we may see versions of Model 3/Y S/X that are better adapted to Robtoaxi use.

Seriously, I do remember that. However, two seater taxi is successful almost entirely in highly congested Asian cities, primarily in Parts of Southeast and South Asia. In numbers India is the largest by far, and many of those three wheelers are already electric. The idea that a two seater is a good solution because the typical load is one to two people seems rational, but…human behavior and market needs are quite different. For example, Robotaxi seems likely to be used by people doing shopping. What two seater will have luggage space. Of course it could happen. Were it to happen I would be confident it would be a commercial failure.

 

[nativewolf]

Seriously, I do remember that. However, two seater taxi is successful almost entirely in highly congested Asian cities, primarily in Parts of Southeast and South Asia. In numbers India is the largest by far, and many of those three wheelers are already electric. The idea that a two seater is a good solution because the typical load is one to two people seems rational, but…human behavior and market needs are quite different. For example, Robotaxi seems likely to be used by people doing shopping. What two seater will have luggage space. Of course it could happen. Were it to happen I would be confident it would be a commercial failure.

True and good points, the 3 wheeler has space for 5-6, I know, I've done it many many times from Thailand to Cambodia to India. It's cheap. They should make a true 3 wheel electric for developing counties. USAID funded such an effort, fyi. Using in wheel electric motors.

 

[Usain]

Surveys show that a huge number of Americans don't buy an EV because they are worried about an expensive battery replacement. And they think that it costs more to maintain an EV.

Solution: Tesla should offer a 200,000 mile battery warranty and free oil changes for life!

 

[Usain]

I imagine in 3 years we might have optimistically 5-6 cities running robotaxis. Peak optimism. Doubt that materially moves our earnings especially with a bunch of operational cost to observe and mcontrol the fleet. “Oh just you wait 2030 is the year of robotaxi.” You can keep doing this same *sugar* over and over again.

One year ago I heard the word “occupancy network” here every week I swear. You don’t have a clue. You have misplaced belief. It’s basically religion.

No one is doubting the potential of robotaxi. Where the **** did you see me say that? I’m telling you that you’re blind and it’s not happening anytime soon. And if it does happen there’s a whole bunch of *sugar* that comes with it which I outlined in my previous posts which no one here has articulated a reasonable rebuttal (because everything I say is true) and Tesla might start to rocket from $50 to $500 rather than $300 to $1500.

So you do think Tesla's technology is good enough that robotaxis could be in 5-6 cities in 3 years. I think so too.

So if Tesla can run robotaxis in 5-6 cities then there is no reason it can't run robotaxis in 50-60 cities or 500-600 cities. Unlike Waymo, Tesla's cost is rock bottom. So once FSD is good enough for a few cities, it will be ready for deployment in many cities.

You say it can happen in 3 years but it's not happening soon. Strange contradiction. 3 years seems pretty soon to me.

If robotaxi is solved in 3 years then Tesla stock goes to the moon, unlike Rocket Labs which is the stock you were touting 3 years ago.

 

[unk45]

True and good points, the 3 wheeler has space for 5-6, I know, I've done it many many times from Thailand to Cambodia to India. It's cheap. They should make a true 3 wheel electric for developing counties. USAID funded such an effort, fyi. Using in wheel electric motors.

They do make them in large numbers:
Thailand:สามล้อไฟฟ้า | BIZ NEX Motor | Thailand #1 Electric TukTuk
India: ://www.godigit.com/motor-insurance/commercial-vehicle-insurance/find/best-electric-rickshaw-in-india
From huge companies such as Mahindra to many small ones electric Tuk-Tuk /Rickshaw are ubiquitous.
Several countries/companies have full service with multiple options from buying outright to leasing with charging or battery exchange.
If that is not enough this trend has even reached Brazil, with emphasis on local urban delivery. (FWIW several of my local delivery services for Supermarkets now use these, as do some other service providers.)

Tuktuks elétricos - Cicloway - #DescubraOMovimento

Tuk Tuk elétrico: ESG para a sua empresa! Conquiste clientes com a sua frota elétrica e 100% sustentável através dos tuk tuks elétricos Cicloway.

www.cicloway.com.br

France: ://www.goupil-ev.com/en/
-Then there is the Citroën Ami and many more in Italy too:
These are commercial providers, several of which are thriving. There are many more in other countries. Perhaps unsurprisingly , few North Americans know about these. Nearly all urban centers in Italy, France etc have postal delivery, service providers and often passengers in these tiny vehicles.
Of course we cannot forget one of the most charming (I have one reserved so I am biased:
This one even comes in two models, one does not even need a driving license!:

Home

The Microlino is a space-saving, electric bubble car with: Two seats, range of up to 230 km, 12.5 kW power and top speed of 90 km/h.

microlino-car.com

None of these, thus far, has anything remotely like a simple cruise control, much less auto navigation. It is entirely plausible to equip this category with localized auto driving. Given the very tight navigation areas for these there, even were Tesla to help, some sort of geo-fencing would be necessary because they often navigate in areas normally precluded for other vehicles.

Lastly, even in normally restricted areas there are places where Tesla can enter such areas. For example I drove a Model X in La Città Alta of Bergamo. Robotaxi in such locations worldwide can have enormous potential, even small ones.

Those two seater Robotaxi, if small enough and with some luggage space could be potentially popular, but also limited due to the impact on pedestrians and other vehicles.

Understanding the Elon advocacy for Robotaxi, the reality is that such a market is highly variable in type and context depending on characteristics of each such area. Because of those practical and legal issues, I strongly believe Robotaxi will not be a generalized application. Even if it were possible, there are local rules governing transportation for hire in nearly every urban area in the world.

 

[nativewolf]

I am completely out of touch unk45! Thanks, fun times!

 

[dhrivnak]

This is what I wanted

Electric motors sound awesome when you don’t quiet them into oblivion

Yes my Roadster was similar, To me it sounded like the Enterprise launching into Warp drive, and it felt like it.

 

[Zaddy Daddy]

How, exactly, will you measure disengagements when there is no human in the car, nor remotely controlling it?

Who will be making the "critical" judgement to "disengage" and how will this be accomplished?

No one is arguing you can't use accidents as a metric after there is no human in the car!

But you can't put the cart before the horse. Tesla won't be allowing unsupervised FSD until after they've passed the accuracy / reliability gauntlet. So who cares?

We care about it's progress now. These are last 5 minutes of a game, the most intense. It's all happening while a human is in the car. The humans aren't going to let an accident happen - accident metrics are not that helpful.

Critical disengagements are very helpful. Is there bias in them? Yes? Are their issues with self-reporting? Yes.

Despite those issues, will you still be able to see signficant changes in the disengagement metrics if the software improves reliability 10x? Undoubtedly yes. We are already seeing it. 3x increase in miles per critical disengagement from V11 to V12 so far.

Despite the biases and variance, you will not have software capable of driving around unsupervised until mile / critical disengagements gets to around 30,000 (now at 300).

We can sit back and enjoy watching the metric ride.

If you believe that is an easy thing to happen within the next year or two, I suggest you buy a lot of shares.

If you believe it's never going to happen, I wouldn't

As for me, I'm playing "wait and see".

 

[Singuy]

No one is arguing you can't use accidents as a metric after there is no human in the car!

But you can't put the cart before the horse. Tesla won't be allowing unsupervised FSD until after they've passed the accuracy / reliability gauntlet. So who cares?

We care about it's progress now. These are last 5 minutes of a game, the most intense. It's all happening while a human is in the car. The humans aren't going to let an accident happen - accident metrics are not that helpful.

Critical disengagements are very helpful. Is there bias in them? Yes? Are their issues with self-reporting? Yes.

Despite those issues, will you still be able to see signficant changes in the disengagement metrics if the software improves reliability 10x? Undoubtedly yes. We are already seeing it. 3x increase in miles per critical disengagement from V11 to V12 so far.

Despite the biases and variance, you will not have software capable of driving around unsupervised until mile / critical disengagements gets to around 30,000 (now at 300).

We can sit back and enjoy watching the metric ride.

If you believe that is an easy thing to happen within the next year or two, I suggest you buy a lot of shares.

If you believe it's never going to happen, I wouldn't

As for me, I'm playing "wait and see".

Miles to disengagement is an exponential growth metric. Lots of errors in the beginning and then lots of miles prior to disengagement in the end.

If you plot the miles to disengagement, it shows a very well plotted exponential curve from v1 to v12. The reason is there are thousands upon thousands of problems to fix in the beginning. Fixing 10 problems probably gives you 1 additional mile before disengagement because the system is so unstable. However in the tail end where it already drives 300 miles perfectly, fixing just 1 or 2 problems can raise 300 miles to 3000 miles.

 

[Krugerrand]

Understanding the Elon advocacy for Robotaxi, the reality is that such a market is highly variable in type and context depending on characteristics of each such area. Because of those practical and legal issues, I strongly believe Robotaxi will not be a generalized application. Even if it were possible, there are local rules governing transportation for hire in nearly every urban area in the world.

So, what if the world just became more synchronized? Just throwing it out there and not looking to debate it on the forum.

It took a while but eventually even the slower amongst our species finally realized that Tesla’s charging standard and network was better than the rest by a mile. No contest.

I see a truly different world ahead for mankind if we hope to survive. Won’t happen in our (you and I at least) lifetime, but I believe it needs to happen. I’m not suggesting one culture, but I am suggesting getting on the same page about a lot of things to make lives for all more streamlined and as a byproduct, more rewarding. UBI, road design/road markings (tunnels and FSD), access to high speed Internet for all, renewable energy for all and done such a way that it’s essentially free, bots - Can people not see what he sees, so wrapped up in all the ways they think and believe it can’t work, instead of the ways it could work?

FYI, just picked your post from the hat in which to respond. And I truly am not wanting to debate all the ways in which it can’t possibly happen. Everyone is already doing a bang up job of that.

 

[ItsNotAboutTheMoney]

Despite the biases and variance, you will not have software capable of driving around unsupervised until mile / critical disengagements gets to around 30,000 (now at 300).

You think 30,000 is OK? No wonder insurance premiums are rising!

How Common Are Car Accidents? - Friedman & Simon, L.L.P.

After experiencing an accident that someone else caused, you may be wondering, “How common are car accidents?” The rate of accidents depends in part on the age and experience level of the driver. The vast majority of Americans will be involved in at least one crash in their lives. Click to learn...

www.friedmansimon.com

(Referenced Rates of Motor Vehicle Crashes, Injuries and Deaths in Relation to Driver Age, United States, 2014-2015 - AAA Foundation for Traffic Safety
Old, but it's harder to find collision v fatality)

Ages 16-17: 1,432 collisions per 100 million miles driven.
Ages 18-19: 730 collisions per 100 million miles driven.
Ages 20-29: 549 collisions per 100 million miles driven.
Ages 80-plus: 432 collisions per 100 million miles driven.

Collisions for all other age groups amounted to less than 330 per 100 million miles driven.

Even in the worst group that's 69832.4 miles per collision, and 303030 miles for ages 30-80.

Fatality statistics are easier to find:

Fatality Facts 2021: State by state

A summary of fatality statistics about state by state compiled by IIHS from 2021 Fatality Analysis Reporting System (FARS) data.

www.iihs.org

According to the IIHS, the average vehicles miles per death in the USA is well over 1 million.

You really need a _much_ higher rate per critical disengagement for unsupervised driving.

 

[2daMoon]

No one is arguing you can't use accidents as a metric after there is no human in the car!

But you can't put the cart before the horse. Tesla won't be allowing unsupervised FSD until after they've passed the accuracy / reliability gauntlet. So who cares?

We care about it's progress now. These are last 5 minutes of a game, the most intense. It's all happening while a human is in the car. The humans aren't going to let an accident happen - accident metrics are not that helpful.

Right now there is data being generated during the time the car is operating autonomously. The entire period up to a disengagement provides significantly more data that can/will be used to demonstrate accuracy and reliability.

Critical disengagements are very helpful. Is there bias in them? Yes? Are their issues with self-reporting? Yes.

Critical disengagements are helpful for training the neural net, dealing with edge cases, and showing the Tesla engineers information on what needs to be focused upon.

These events do not provide any useful data for determining accuracy and reliability in the sense you describe.

Despite those issues, will you still be able to see signficant changes in the disengagement metrics if the software improves reliability 10x? Undoubtedly yes. We are already seeing it. 3x increase in miles per critical disengagement from V11 to V12 so far.

Despite the biases and variance, you will not have software capable of driving around unsupervised until mile / critical disengagements gets to around 30,000 (now at 300).

Again, this metric might be useful for Tesla engineers. It would be tainted for use in determining FSD's accuracy and reliability for a third-party regulatory purpose because it does not reflect FSD operation. A disengagement is a human operation.

We can sit back and enjoy watching the metric ride.

If you believe that is an easy thing to happen within the next year or two, I suggest you buy a lot of shares.

If you believe it's never going to happen, I wouldn't

As for me, I'm playing "wait and see".

Adoption as a system that operates without supervision will happen when it happens. It is already significantly more adept than humans in the lion's share of operation. The edge cases are being added constantly, and greater abilities to manage them are accumulating daily. This is the march of 9s that Tesla's engineers are striving toward, and it is measurable.

How many digits to the right of the decimal point will it be considered enough to allow unsupervised operation? This is something I'd rather Tesla decide, than some regulatory agency without the background, knowledge, and understanding of the system. And certainly not with their using data based entirely on human behavior to evaluate an autonomous driving platform with a goal of taking the dangerous human out of the operation entirely.

As a driver assist device it can be used now. Over time people will become more and more comfortable with FSD in that role. That use by customers is what will demonstrate the accuracy and reliability which will lead to acceptance. This requires no belief. It will either continue to improve, or it won't.

Performance up until now indicates that constant improvement is the norm for Tesla.

There is no need for third parties to use human disengagement in the evaluation of FSD.

 

[2daMoon]

You think 30,000 is OK? No wonder insurance premiums are rising!

How Common Are Car Accidents? - Friedman & Simon, L.L.P.

After experiencing an accident that someone else caused, you may be wondering, “How common are car accidents?” The rate of accidents depends in part on the age and experience level of the driver. The vast majority of Americans will be involved in at least one crash in their lives. Click to learn...

www.friedmansimon.com

(Referenced Rates of Motor Vehicle Crashes, Injuries and Deaths in Relation to Driver Age, United States, 2014-2015 - AAA Foundation for Traffic Safety
Old, but it's harder to find collision v fatality)

Ages 16-17: 1,432 collisions per 100 million miles driven.
Ages 18-19: 730 collisions per 100 million miles driven.
Ages 20-29: 549 collisions per 100 million miles driven.
Ages 80-plus: 432 collisions per 100 million miles driven.

Collisions for all other age groups amounted to less than 330 per 100 million miles driven.

Even in the worst group that's 69832.4 miles per collision, and 303030 miles for ages 30-80.

Fatality statistics are easier to find:

Fatality Facts 2021: State by state

A summary of fatality statistics about state by state compiled by IIHS from 2021 Fatality Analysis Reporting System (FARS) data.

www.iihs.org

According to the IIHS, the average vehicles miles per death in the USA is well over 1 million.

You really need a _much_ higher rate per critical disengagement for unsupervised driving.

I think you are making a presumption that each critical disengagement represents an avoided collision or death.

What if that presumption is false and critical disengagements only represent the level of comfort with a maneuver of the one choosing to disengage being exceeded?

Not every person given the ability to disengage FSD will behave identically in a given circumstance. This is the root of the problem with many human drivers that leads to collisions and deaths.

 

[dhanson865]

positive spin a Tesla involved in an accident for once

Tesla driver miraculously survives Florida crane collapse - Interesting Engineering

Incredibly, the driver of a Tesla Model Y, which was hit by the falling wreckage, walked away from the incident unscathed.

interestingengineering.com

 

[bpjod]

Surveys show that a huge number of Americans don't buy an EV because they are worried about an expensive battery replacement. And they think that it costs more to maintain an EV.

Solution: Tesla should offer a 200,000 mile battery warranty and free oil changes for life!

Figure out how to do the oil changes OTA so they’re no longer a time-sucking inconvenience and I’m in!

 

[GhostSkater]

My last Supercharger before returning it happened to be a V4 dispenser one, and I found some interesting stuff

There are space for two more cabinets, but I looked around and it doesn’t seem to be any pre run conduits or wires for more stalls, meaning that higher power true V4 ones that will be upgraded in the future might have a cabinet for 2 stalls instead of 4 right now

Either that or it’s just for dropping in two more cabinets and linking them to the current ones for higher site power, right now maximum simultaneous power a V3/V4 dispenser can do when full is 97 kW per plug, not a problem right now because Teslas drop in power quickly and it’s rare to have full sites, but as charge curves get fatter the problem will become more and more apparent

 

[bpjod]

After a week of living with the lastest FSD a couple thoughts have come to mind:

1. When purchasing a new vehicle, would you rather buy the one that can chauffeur you and your loved ones around more safely than a human driver or a vehicle that will never have that capability? Not saying it's there yet, but I'm more optimistic than ever.

2. Hypothetically, let's grant that the CEO's "politics" alienate 40% of the population (as some like to argue, but I don't believe):
2a. Has any car company owned 60% of the world-wide market share? Anything even close in the last 100 years? There's still a lot of potential sales to be had even with 40% ideologically opposed.
2b. Tesla's product mix, especially the parts they're focussing most intently on, increasingly look to me to be Business to Business (B2B) goods rather than Business to Consumer (B2C). Think about it: Megapacks vs Powerwalls, Tesla Semis, Robotaxis vs SEXY, Optimus, relatively stagnant solar division. If they decided to hell with B2C and focussed solely on Megapacks, Semis, selling Robotaxis to fleet operators (or operating a world-wide fleet themselves) and building an army of hundreds of millions (billions?) of robots to replace human workers, they'd still have a great chance to become the largest company the world has ever seen. Who needs to chase virtue-signalling consumers? Who knows what other skunkworks teams they've got working on the next big thing?

 

[Krugerrand]

I think you are making a presumption that each critical disengagement represents an avoided collision or death.

What if that presumption is false and critical disengagements only represent the level of comfort with a maneuver of the one choosing to disengage being exceeded?

Not every person given the ability to disengage FSD will behave identically in a given circumstance. This is the root of the problem with many human drivers that leads to collisions and deaths.

That’s a fact. I recently disengaged because I wasn’t interested in seeing if the car would continue entering into a deep, blind, paved ditch while executing a turn (yeah, the design of the turn must have been by a first year engineering student - it’s so dumb - indeed, that whole street is dumb and until the new FSD the car either failed to make the turn onto the street entirely or decided to turn at the next more conventional street), possibly ripping off its facia, or if it would get confused on its own and disengage, or eventually see the ditch and correct. I’ll let FSD give that corner a go when I have a FSD functioning CT.

I also disengage when I get to a destination where it wants to stop ‘at the front door’ and parking is around the side on a different street.

 

[DarkandStormy]

Ummmm, you say prioritize, but the Reuters article literally says "Tesla has canceled the long-promised inexpensive car that investors have been counting on"

We have a supposed robotaxis unveil coming. Nothing for Model 2.

So...deduce from what what you will.