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

[PlaidCPA]

I am so excited to see the margins on 3/Y since this Q they are not mixed up with S/X. Very rare opportunity this Q.

 

[Words of HABIT]

The market needs to see Robotaxi certification cleared by the government before they can factor in robotaxi revenues, before that they can start modeling FSD take rate and put that under margins.

This is a huge hurdle that it's up in the air if Tesla will ever get the certification or not even if they have a very competent system. Many shorts or wrong doers are willing to claim all sorts of lies against the system, clouding the government's judgement. Currently the government literally investigates every freaken major crash. And now we know some people are blaming autopilot for their accidents EVEN when their system is not active trying to get out of their liability. This is a uphill battle Google didn't have to deal with this because it was such an enclosed system since only employees get to run their autonomy program.

There is no one World government. Some Countries will adopt first and reap the benefits of FSD architecture. Tesla can activate in those Countries first. All other Contries will then follow once they understand the savings measured in lives saved. It is a no brainer.

 

[Singuy]

There is no one World government. Some Countries will adopt first and reap the benefits of FSD architecture. Tesla can activate in those Countries first. All other Contries will then follow once they understand the savings measured in lives saved. It is a no brainer.

We have seen autopilot FUD all over the world. No shortage of the Chinese and the Germans play into the FUD's hand. Tesla even lost the lawsuit calling it autopilot in Germany.

No one world government, but definitely one world full of FUD. I am bearish on FSD robotaxi, but bullish on FSD. It's too easy to scare people with lies about Tesla's system. Most government erred on the side of caution.

 

[CLK350]

I am so excited to see the margins on 3/Y since this Q they are not mixed up with S/X. Very rare opportunity this Q.

I recall seeing recently 3/Y margins in China reported to be in the 30-35% range; with Fremont in the 20-25% range.

Also curious, how to play/ trade the upcoming weeks w/ my 20% trading money - so figuring out how the HF's will play it - taking macros and Biden's Infra Plan. Not holding my breath there, given how agnostic political entities are to oil & gas donations.

Not expecting either this unique opty for the Lidar gang and the lagging ICE's (plus the 3 letter agencies) to knock down FSD a notch by starting a global V2X (Vehicle to everything, where VW is a pioneer BTW ) initiative in their road & infrastructure plans.

Or if going by the numbers, and if playing short term go with DiscoDucky's not advice

 

[Knightshade]

The market needs to see Robotaxi certification cleared by the government before they can factor in robotaxi revenues, before that they can start modeling FSD take rate and put that under margins.

This is a huge hurdle that it's up in the air if Tesla will ever get the certification or not even if they have a very competent system. Many shorts or wrong doers are willing to claim all sorts of lies against the system, clouding the government's judgement. Currently the government literally investigates every freaken major crash. And now we know some people are blaming autopilot for their accidents EVEN when their system is not active trying to get out of their liability. This is a uphill battle Google didn't have to deal with this because it was such an enclosed system since only employees get to run their autonomy program.

RTs are legal today in a bunch of US states.

Tesla does not need any "certification' to run them there. In many they simply need to say the car is autonomous and safe and can obey all existing traffic laws. That's it. The state doesn't even check- they take the MFGs word for it.

The "OMG REGULATORS" thing is 100% a red herring obstacle to introducing an L5 (or even L4) system once a company actually has one they trust enough to let the public use.

The only thing stopping someone today is nobody has one today they trust that much (outside of heavily geofenced L4 stuff).


Certainly there'll be other places it IS a real hurdle (the EU is the big elephant in that room) but if you've already GOT the system there's no "waiting for regulators" stopping you right now from making a bunch of RT revenue in the dozen or two states you can turn the system on right now while the rest catch their laws up.

And having it run safely in that bunch of US states is a pretty good way to signal to other places it's safe and they might wanna hurry with updating their guidelines.

It's also be a pretty good way to signal to the market it's here.

 

[Unpilot]

There is no one World government. Some Countries will adopt first and reap the benefits of FSD architecture. Tesla can activate in those Countries first. All other Contries will then follow once they understand the savings measured in lives saved. It is a no brainer.

No world government?...But illuminati...and lizard people...George Soros.

You must be a Sheeple...open your eyes man!

 

[Cosmacelf]

Biz Journal reporting SpaceX will be building across the toll road from Giga Austin. Does that make sense?

Yes. Almost certainly for Starlink. SpaceX needs to build millions of Starlink terminals.

 

[Singuy]

RTs are legal today in a bunch of US states.

The "OMG REGULATORS" thing is 100% a red herring obstacle to introducing an L5 (or even L4) system once a company actually has one they trust enough to let the public use.

The only thing stopping someone today is nobody has one today they trust that much (outside of heavily geofenced L4 stuff).


Certainly there'll be other places it IS a real hurdle (the EU is the big elephant in that room) but if you've already GOT the system there's no "waiting for regulators" stopping you right now from making a bunch of RT revenue in the dozen or two states you can turn the system on right now while the rest catch their laws up.

And having it run safely in that bunch of US states is a pretty good way to signal to other places it's safe and they might wanna hurry with updating their guidelines.

I thought the company needs to submit a bunch data like intervention free miles to the authorities first? Isn't that what Waymo did?

 

[goinfraftw]

Yes. Almost certainly for Starlink. SpaceX needs to build millions of Starlink terminals.

Probably billions, tbh. Don't forget IoT.

 

[woodisgood]

I like the idea of 'you kids get off my island' much better than 'you kids get off my lawn.'

"You kids" refers to one's own kids, right? Asking for a friend.

 

[PeterJA]

For Tesla to become the world’s largest company in the next few years, FSD is required.

But to drill down more specifically, what does the market need to see in order to start factoring FSD into TSLA's valuation?

IMHO, it's simply a clear and straight line to L4/5. When the market believes it is inevitable and only a function of time, you'll see the valuation increase. This next step change of the FSD beta is going to publicly show exactly how fast the system learns with 1,000 (or 2,000?) beta participants. I think folks are going to be able to extrapolate the rate of improvement to tens of thousands or hundreds of thousands of vehicles, and they are going to realize that there aren't going to be many (any?) edge cases that simply will be unsolvable for Tesla given enough time. Right now most of the market is still focused on questions like "Who has the best approach for self-driving?" or "Is self-driving really solvable?"

Yes, it will still be much time before there's enough data to prove safety, regulators get on board, we work through all the political and philosophical BS that will undoubtedly come, and robotaxis take whatever form Tesla decides and become a revenue-generating reality. But once the market sees this as inevitable, the stock will [probably] start to reflect it. That could actually start as soon as the next FSD beta release if the improvement is profound. That may be what Elon is suggesting. It's also the kind of inevitability, that, if I were a CEO of a competing self-driving outfit, might cause me to step down. I'm not saying it will be, but the next release of FSD beta could be, publicly, game over.

The completion of Dojo (Tesla's AI training supercomputer) should be a turning point in FSD progress, if not the market's reaction. Currently Tesla's FSD neural nets are getting pretty smart, but their rate of learning is limited by the speed of a staff of humans labeling camera data from the fleet. When Dojo takes over the labeling job, and sends the humans into "Operation Vacation," the rate of learning will shift from relatively slow to "exploding with unrelenting fury," in the words of futurist Ray Kurzweil.

I recall Elon said in an earnings call that Dojo should be completed this year. Maybe that time is near, or already arrived, enabling the general release of FSD Beta soon.

 

[Knightshade]

I thought the company needs to submit a bunch data like intervention free miles to the authorities first? Isn't that what Waymo did?

That's California specifically.

A bunch of other states you need submit anywhere from:

Literally nothing (but it has to be able to obey all traffic laws)
to
"I certify this car self drives, can obey all traffic laws, and is insured"

And that's about it.

If anybody has an L4 or L5 car they trusted the public to use today, they could have it on the road today in some states (and Monday in others once govt. offices open to submit their "I promise this is safe" doc).


Here's an example- US State of Florida-

Florida Will Allow Autonomous Cars with No Safety Drivers on Public Roads Starting July 1

The new law is intended to make Florida competitive with other states that have welcomed self-driving vehicles and testing.

www.caranddriver.com

L4/L5 cars legal. RIGHT NOW (for a couple years now in fact). No human driver needed.

If used for ridesharing they must have minimum insurance levels specified.

That's it.


or hell- here's Texas- home of GF Austin-

https://capitol.texas.gov/tlodocs/85R/billtext/pdf/SB02205F.pdf#navpanes=0

Passed in 2017.

Allows self-driving cars with no human in it required as long as it can obey all traffic laws, is insured, and has an onboard data recorder.

 

[Mo City]

The completion of Dojo (Tesla's AI training supercomputer) should be a turning point in FSD progress, if not the market's reaction. Currently Tesla's FSD neural nets are getting pretty smart, but their rate of learning is limited by the speed of a staff of humans labeling camera data from the fleet. When Dojo takes over the labeling job, and sends the humans into "Operation Vacation," the rate of learning will shift from relatively slow to "exploding with unrelenting fury," in the words of futurist Ray Kurzweil.

I recall Elon said in an earnings call that Dojo should be completed this year. Maybe that time is near, or already arrived, enabling the general release of FSD Beta soon.

According to Elon, Dojo isn't needed for L5.

 

[Singuy]

According to Elon, Dojo isn't needed for L5.

I think Tesla's Dojo is their AI division, a spin off from the FSD team. They have developed self supervised vision recognition algorithm (aka the holy grail) that will enable AI vision for general robotics and automation. This I believe will be put into their factory 3.0 model, and will also help other companies train for general AI vision applications.

Tesla's AI day will be all about Dojo and the revenue stream that comes with it (beyond FSD).

 

[Opus BC fan]

Is it Monday yet?

 

[HG Wells]

If Tesla keeps executing like this, we'll all be retired in a couple years and we can all lose track of days together....................(until we all get on each other nerves and then tell each other to get off each other's islands)

You need to think bigger. Much bigger.
Continent, yes I think it might be possible.
New Zealand or maybe Australia if I stretch it a bit.

 

[powertoold]

When Dojo takes over the labeling job, and sends the humans into "Operation Vacation," the rate of learning will shift from relatively slow to "exploding with unrelenting fury," in the words of futurist Ray Kurzweil.

This isn't the correct interpretation of what Dojo will be used for, as far as Elon and Karpathy has explained.

For now, Dojo is simply a NN-optimized training cluster with a custom CPU and cooling design. Dojo would allow Tesla to create larger NN models since the model size is limited by how long it takes their cluster to train / process the data. After Tesla transitions to video labeling (which will still be done manually by humans), they'll need a lot more training flops.

Operation Vacation refers to automated data curation from the fleet. Humans will still need to label the data, but the fleet will help collect some of it.

 

[JustMe]

Is it Monday yet?

Nope - three more sleeps!

 

[Knightshade]

This isn't the correct interpretation of what Dojo will be used for, as far as Elon and Karpathy has explained.

For now, Dojo is simply a NN-optimized training cluster with a custom CPU and cooling design. Dojo would allow Tesla to create larger NN models since the model size is limited by how long it takes their cluster to train / process the data. After Tesla transitions to video labeling (which will still be done manually by humans), they'll need a lot more training flops.

Operation Vacation refers to automated data curation from the fleet. Humans will still need to label the data, but the fleet will help collect some of it.

Yup... this confusion always seemed weird to me.

The reason you need human labelers is they're labeling objects the NNs can't yet ID well enough.

So you label examples, then use them to train the NN.

If the computer could already reliably ID the thing well enough to label it on its own, then it wouldn't need to be trained to recognize it.


The still to video transition benefit is you get a system that understands the thing labeled "Truck" in frame 1 of the video is the SAME object in frame 2....so you don't need to label every frame like you did before....and then another big gain is getting it to understand the object labeled "truck" in frame 27 on ONE camera is the same object in frame 39 from another camera.

Now you can label a ton more data in the same amount of time because you're not having to label the same thing over and over for a given set of data... and thus need a much more robust back end to train against the much larger data set.

That's what Dojo will do.


Operation Vacation is the engineering team setting up the infrastructure so new tasks can be created and trained without THEM being around. It still needs the data labelers and PMs who will use that infrastructure to pick a task type, gather and label the training data, then run the training cycles through the system.

 

[buttershrimp]

Nope - three more sleeps!

Premarket buy via schwab put in Sunday night? I need to get in quick before stock price goes up by 2 full Gordons.