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

[Buckminster]

And a drive by wire hint from employee Bill Wright (take with a grain of salt)

Surely CT rear wheels at least will be drive by wire. Might as well make them all.

They are just practising for gen 3.

 

[Drumheller]

TL;DR of long post:
It is ~100% certain that Robotaxi FSD can be solved with Tesla's V12 approach. This fact alone should make you shake the couches.

Look at Karpathy's talk above and pay special attention to the slide at about 25:45 (see below set at proper time).

He presents the LLM scaling law which says performance of an LLM is a smooth, well-behaved, predictable function of N, the number of parameters and D, the amount of training data. He also mentions that there is no known limit to this scaling law.

I see no reason the same law would not apply to Tesla's FSD models. This is something that has been talked about a lot.

Conclusion #1: There is no reason to believe that FSD V12 will reach a local maximum. V12 is the kind of model that Tesla wanted all along. It will just keep getting better and better without any new approach or technology breakthrough. (Or as Karpathy puts it, algorithmic progress is a bonus, but not necessary).

Conclusion #2: It is ~100% certain that Robotaxi FSD can be solved with Tesla's V12 approach.

Conclusion #3: It is extremely likely that Tesla will be the first to market with a Robotaxi FSD. Tesla is currently the only company that can satisfy the "D" variable. Tesla is the only one with enough quality training data. (Yes, Kathy Wood has been right about this)

Conclusion #4: It is possible that Hardware 3 will not be able to handle a model with a large enough "N", number of parameters, to do Robotaxi FSD. Hardware 4 has a better shot at it. But I think we can safely say that even Hardware 3 will eventually get a version of FSD that is very, very good at driving. This is why Elon is saying that it looks like Hardware 3 will be better than a human. Hardware 4 will be even better and Hardware 5 will be better still. This is all about the ability to deal with a larger "N".

If you have been following the progress of FSD closely, none of the above should be new. But it is nice to see how the narrative in the Tesla community comports perfectly with the LLM scaling law Karpathy is discussing.

According to the scaling law, it's not a matter of if Robotaxi FSD will get here. It's when. "When" is anyone's guess. But if you can grok what I've said above, you should gain a lot of confidence that FSD V12 is key because it promises almost infinite improvement over time. It is the most important FSD release ever. The system is now in place to solve this. It will happen.

So shake those couches folks. It's very likely that FSD is fixin' to get really, really good.

(Disclaimer: While I do have a degree in computer science, I am not an AI expert. I'd love to get feedback from those who know this stuff better than I do. Am I right?)

My background includes two computer science degrees but I know little of modern A.I. techniques. I'm mostly in agreement with what you've posted. Tesla's approach has always been the most correct approach, in my opinion. I think the cameras in hardware v3 are sufficient but I'm less certain about the FSD computer in v3. I think it could be enough but I don't have enough information to offer a stronger opinion.

I won't agree v12 is 100% enough because I wouldn't say anything is 100% certain until it happens but I'd comfortably agree with a 90% certainty.

Interesting times ahead, hopefully a bunch of positivity.

 

[SOULPEDL]

TL;DR of long post:
It is ~100% certain that Robotaxi FSD can be solved with Tesla's V12 approach. This fact alone should make you shake the couches.

Look at Karpathy's talk above and pay special attention to the slide at about 25:45 (see below set at proper time).

He presents the LLM scaling law which says performance of an LLM is a smooth, well-behaved, predictable function of N, the number of parameters and D, the amount of training data. He also mentions that there is no known limit to this scaling law.

I see no reason the same law would not apply to Tesla's FSD models. This is something that has been talked about a lot.

Conclusion #1: There is no reason to believe that FSD V12 will reach a local maximum. V12 is the kind of model that Tesla wanted all along. It will just keep getting better and better without any new approach or technology breakthrough. (Or as Karpathy puts it, algorithmic progress is a bonus, but not necessary).

Conclusion #2: It is ~100% certain that Robotaxi FSD can be solved with Tesla's V12 approach.

Conclusion #3: It is extremely likely that Tesla will be the first to market with a Robotaxi FSD. Tesla is currently the only company that can satisfy the "D" variable. Tesla is the only one with enough quality training data. (Yes, Kathy Wood has been right about this)

Conclusion #4: It is possible that Hardware 3 will not be able to handle a model with a large enough "N", number of parameters, to do Robotaxi FSD. Hardware 4 has a better shot at it. But I think we can safely say that even Hardware 3 will eventually get a version of FSD that is very, very good at driving. This is why Elon is saying that it looks like Hardware 3 will be better than a human. Hardware 4 will be even better and Hardware 5 will be better still. This is all about the ability to deal with a larger "N".

If you have been following the progress of FSD closely, none of the above should be new. But it is nice to see how the narrative in the Tesla community comports perfectly with the LLM scaling law Karpathy is discussing.

According to the scaling law, it's not a matter of if Robotaxi FSD will get here. It's when. "When" is anyone's guess. But if you can grok what I've said above, you should gain a lot of confidence that FSD V12 is key because it promises almost infinite improvement over time. It is the most important FSD release ever. The system is now in place to solve this. It will happen.

So shake those couches folks. It's very likely that FSD is fixin' to get really, really good.

(Disclaimer: While I do have a degree in computer science, I am not an AI expert. I'd love to get feedback from those who know this stuff better than I do. Am I right?)

It's all about timing, but I don't believe Tesla is using all the data it could up to what we have installed anyway. This is simply because it keeps making the same mistakes in my area... although some are gone, not all. Not sure I can draw this conclusion, but it appears a compute limitation only, especially since mine are not "edge" cases as far as I can tell. Maybe they intentionally only get it to 90% then iterate to advance faster overall, IDK.

It does boil down to just a matter of time, and pretty soon. Heck if we can have a conversation with just 70GB of "kernel" AI (and even less), who knows how compact and efficient this can become. Can we then surpass human reaction time? That's where hardware 4 and 5 come in I bet. It's very exciting to see and experience this unfolding!

 

[Tony73]

I don't think you read my post!
- *wait* until investors give up on Rivian's money-losing ways
- buy Rivian in a hostile take-over at *pennies on the dollar*
- *drastically improve* the efficiencies of their R1-S and R1-T [and the delivery truck]
- fire 80% of the people
- there's a greater than 0 chance this could happen one day

So... yeah... pay $4B (number pulled out of my...), get rid of inefficient hardware, software, people... I'd now add the obvious: improve the efficiencies of the factories while they're at it. Saves them from locating/siting/permitting/building new factory(ies). Again, not likely, but plausible... if Rivian is still losing money (and out of cash) in a couple years, and all of Legacy Auto is trying to stay alive (and can't be going on a buying spree), and Tesla dows not know what to do with all their FCF... some get the point.

Even you are struggling to cite the benefits from all the work and aggravation of a hostile takeover.
Saving “locating/siting/permitting” is not what worth everything you just described. This is a company that built a factory in Germany in record time. Tesla has plenty of space in GigaTx. They are adding more space in GigaNevada and will soon start building out GigaMexico. I think they are pretty good at locating/siting/permitting factories. That is not Tesla’s bottleneck.

Improving the efficiencies of existing designs is often as much work as starting from scratch.

Tesla has done one big acquisition in its history and it did not exactly pan out. They really should stay away from those and play to their strengths which is clean sheet, first principles design.
Rivian has nothing Tesla needs even if they were giving it away for free. Not factories, vehicles models or brand.

I am a fan of Rivian by the way. My position is not a knock on Rivian. It’s just that Tesla can do so much better internally for so much less than anything they could acquire. They are a cost leader and a technology leader.

Someone else can buy Rivian. Maybe a Chinese company seeking North American exposure.

 

[Stretch2727]

Saw the CT in NYC today.

Build quality looked perfect. No uneven gaps that I could see.
Panels looked flat with no waviness.
It was about the size I expected, but my expectation was for large.
Lots of traffic in the showroom. There was actually a short wait as I think they were limiting the number of people in the store.

No Tesla badge or script anywhere outside.

Can wait to see on the road.

 

[Usain]

It is a patent application; it doesn’t mean you will get it granted as broad as claimed or at all.
Based on a few past Tesla applications I did look at, the patent agency Tesla hired did not appear to be particularly good, which wouldn’t improve the chance Of getting stuff granted.

Is there any evidence that Tesla is having trouble getting their patents granted? From my personal experience, it's not that hard to get the patent office to go along with at least some of the claims in the patent application.

 

[jabloomf1230]

No Tesla badge or script anywhere outside.

Another money saving innovation by Tesla!

 

[TesLeif]

Even you are struggling to cite the benefits from all the work and aggravation of a hostile takeover.
Saving “locating/siting/permitting” is not what worth everything you just described. This is a company that built a factory in Germany in record time. Tesla has plenty of space in GigaTx. They are adding more space in GigaNevada and will soon start building out GigaMexico. I think they are pretty good at locating/siting/permitting factories. That is not Tesla’s bottleneck.

Improving the efficiencies of existing designs is often as much work as starting from scratch.

Tesla has done one big acquisition in its history and it did not exactly pan out. They really should stay away from those and play to their strengths which is clean sheet, first principles design.
Rivian has nothing Tesla needs even if they were giving it away for free. Not factories, vehicles models or brand.

I am a fan of Rivian by the way. My position is not a knock on Rivian. It’s just that Tesla can do so much better internally for so much less than anything they could acquire. They are a cost leader and a technology leader.

Someone else can buy Rivian. Maybe a Chinese company seeking North American exposure.

Again, it would have to be a really good deal... "pennies on the dollar", and that dollar would be on the *future* price of Rivian once investors give up on them. And I'm not saying they *should* do this... just that there is a less-than-zero chance they could.

You say you are "a fan of Rivian", but then say Tesla does not need Rivian, even if they were free. I agree they don't *need* Rivian, but there is value to be purchased: *some* of their engineers/staff; their 3 vehicles are very well received and could be improved upon (Tesla is constantly improving their own vehicles, so Tesla never gets their vehicles 100% right the first time either); plenty of neat design features/touches/patents/IP; the factories are likely usable to a large degree; their BRAND has value (most people consider it a good brand, and it could be retained as a sub-brand); the Adventure Network is small but appears to work well and is located in remote places where Tesla is not; *some* show rooms; *some* service centers; etc.

Also, note that it would be a defensive move: prevent BYD (or someone else) from moving into that void and succeeding at it.

Things that would be changed right away: *drastic* drop in options for exterior colours, interior trims, battery/power train options, etc.; new Tesla-made battery pack; Tesla Hardware 5; Tesla software.

But sure... wOrSt iDeA eVeR!!

 

[loquitur]

I don't think you read my post!
- *wait* until investors give up on Rivian's money-losing ways
- buy Rivian in a hostile take-over at *pennies on the dollar*
- *drastically improve* the efficiencies of their R1-S and R1-T [and the delivery truck]
- fire 80% of the people
- there's a greater than 0 chance this could happen one day

So... yeah... pay $4B (number pulled out of my...), get rid of inefficient hardware, software, people... I'd now add the obvious: improve the efficiencies of the factories while they're at it. Saves them from locating/siting/permitting/building new factory(ies). Again, not likely, but plausible... if Rivian is still losing money (and out of cash) in a couple years, and all of Legacy Auto is trying to stay alive (and can't be going on a buying spree), and Tesla dows not know what to do with all their FCF... some get the point

Musk is on record that if the angular Cybrtrk doesn't fly they can always redo (p. 318 of the Isaacson bio "[....] We're not doing a traditional boring truck. We can always do that later. [....]", I'll posit the following. Sometimes it's easier to start from scratch on a concept using existing facilities + knowledge worker assets rather than to improve on efforts by others.

As an imperfect analogy, consider vacancies in commercial real estate high rises in places like San Francisco.
Some suggest repurposing some buildings to be residential. This has happened in Manhattan to some extant.
However, in some cases it was observed that the cost to retrofit exceeded that of just tearing down a building
entirely to rebuild.

A BK Rivian offering super cheap factory space is one thing. If the brand proves valuable that's another.
But von Holzhausen has gone thru trad (softer-looking) pickup designs before (early mockups resembled the El Camino)
so that might again obtain. But perhaps van design is worth a better look.

 

[UkNorthampton]

Another UK oil refinery down, 5 more to go. Electrification mentioned

(closedown completing 2025, reworking as oil terminal/storage starting operation in 2027 - multiple fuel products)

Grangemouth Refinery is an oil refinery complex located on the Firth of Forth in Grangemouth, Scotland, currently operated by Petroineos.

It is the only operating crude oil refinery in Scotland (following the cessation in 2014 of Bitumen refining activities at the Nynas Dundee Refinery[1]), and currently one of the six remaining UK Refineries.

The business cited the infeasibility of continuing to operate as a profitable going concern, under the multiple pressures of over-capacity in the North-West Europe market, the ongoing demand destruction for road fuel from the electrification of road vehicles, compounded by the high cost of manufacturing in the UK.

Grangemouth Refinery - Wikipedia

en.wikipedia.org

Grangemouth oil refinery could cease operations by 2025

Petroineos says it aims to turn the Grangemouth site into a fuels import terminal by spring 2025.

www.bbc.co.uk

 

[jerry33]

I can't bring myself to use FSD much anymore (still occasionally). Knowing v11 is a dead branch, I no longer feel like using it provides useful data to Tesla to "solve FSD". It's not good enough for me to use around town regularly vs me driving. I'll be happy to start using again, as a beta-tester, once v12 rolls out. I'm excited to see if it's as good as the hype.

It may not provide additional useful data to Tesla, but it still provides safety.

 

[MP3Mike]

So... yeah... pay $4B (number pulled out of my...), get rid of inefficient hardware, software, people...

Rivian is already doing all of that, they have announced gen 2 of the R1 products coming in 2024 will eliminate a lot of the inefficiencies, and lower costs. So, a lot of that will be done before they ever have a chance to get to the point of failing,

I just can't see Tesla buying any failed EV manufacturer, there just isn't enough up side. (It would make a lot more sense for some other legacy OEM to buy them.)

 

[EVWatcher]

Rivian is already doing all of that, they have announced gen 2 of the R1 products coming in 2024 will eliminate a lot of the inefficiencies, and lower costs. So, a lot of that will be done before they ever have a chance to get to the point of failing,

I just can't see Tesla buying any failed EV manufacturer, there just isn't enough up side. (It would make a lot more sense for some other legacy OEM to buy them.)

The only value Tesla could buy from rivian is the brand. If the price was right there’s value in Tesla just buying the name and making a line of vehicles under that brand so the consumers who would never buy a Tesla can buy a Rivian.

Like it or not those people exist and they always will. Tesla having a portfolio of names someday does make sense.

 

[mars_or_bust]

One way or another I fail to see how these 200 doldrums can last much longer. Possible, but unlikely. Just too *many* ways it can break out. Cheers

Especially if you compare it to EVERY other car company - they have NOTHING exciting to look forward to. No AI, no robots, no energy projects, no exciting new models, no own batteries (except BYD), no new factories - nothing. Tesla has it all, and more - a breakout WILL come.

 

[Cosmacelf]

RJ Scaringe would never let that happen and he has veto power.

He does? Another effed up corporate governance structure?

 

[Artful Dodger]

The schadenfreude is especially crisp this Thanksgiving: (v12 incoming)

Sam Abuelsamid Has Been Hatin' Hard on Tesla For Over 13 Years! | James Stephenson (15 min ago)

BULLSEYE!

 

[Knightshade]

He does? Another effed up corporate governance structure?

He owns 100% of the class B shares, which have 10 votes each rather than 1 vote each for normal shares... so his nearly 1% of all shares gives him almost 10% voting power... more relevantly though in order to make any major changes at the board level or in the company’s bylaws, the holders of at least 80% of Class B shares need to approve it. As a reminder he holds 100% of them . It's a bit like how the Ford family retains insanely outsized voting power to this day, but moreso.

Especially if you compare it to EVERY other car company - they have NOTHING exciting to look forward to. No AI, no robots, no energy projects, no exciting new models, no own batteries (except BYD), no new factories - nothing. Tesla has it all, and more - a breakout WILL come.

To be fair, they also have single-digit P/Es (GMs is like 4, Fords is 6-something, even Toyota is only ~9) so they don't need those things to justify their current stock prices. Teslas P/E is in the high 70s as of close today.

That's been a major Q argument- you can't justify the stock price if they're "just" a car company-- all those other things you mention, at least one has to turn into a real, large, profitable LOB to do so. That there's as many potential ones as there are is the other side of that argument.

 

[Cosmacelf]

He owns 100% of the class B shares, which have 10 votes each rather than 1 vote each for normal shares... so his nearly 1% of all shares gives him almost 10% voting power... more relevantly though in order to make any major changes at the board level or in the company’s bylaws, the holders of at least 80% of Class B shares need to approve it. As a reminder he holds 100% of them . It's a bit like how the Ford family retains insanely outsized voting power to this day, but moreso.

Wow, now I know how smug and carefree he can be while lighting fire to all his investor's cash. Must be nice to have a job for life. Well, until he burns through all the cash at least.

 

[Electroman]

No Tesla badge or script anywhere outside.

This is one vehicle that would scream Tesla from a mile, badge or not.

 

[Skryll]

This is one vehicle that would scream Tesla from a mile, badge or not.

Kind of like the X in 2016