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

[H Mak]

OK. To me that sounds a bit too much of a longshot to assign it much value.

You're right. Like investing your entire wealth in a start up electric car company and a start up rocket company

 

[Snerruc]

Karpathy was suggesting that there is a ton of potential gain in just improving computer architecture to be more suitable to the task.

So yes, I think Karpathy would agree with most of what you said.

But I don't think he would say the potential solutions lie only with material science. A vastly more efficient way to "do AI" in silicon could still get us a lot closer to the goal. The reason for this optimism is because the enormous delta between the efficiency of the brain and the (in)efficiency of a GPU cluster suggests that huge gains could still be made with architecture alone.

I guess you could also make the argument that Karpathy is not a computer architecture guy. So what does he know?

I seem to remember the University of London. Doing some work with those newfangled transistor things back in the early 60s. They were wiring them together in ways that mimicked the human brain and were surprised in how much a limited number could do and at a relatively low power for the time. Not sure what happened to the research, or if I remember it correctly. Perhaps we should revisit Asimov’s positronic brain.

 

[2daMoon]

Tom Nash brings an informative video focusing on the facts that support how long term investing is the most dependable way to guarantee profit in a system that is geared to fleece retail investors who allow themselves to be influenced emotionally.

Includes clips from interviews with Jeremy Siegel and with Larry Fink (Blackrock) which Tom discusses.

Also touches on the Fed, and on the advantages of Playing Dead.

This serves as a reminder of many things we already know, as well as makes a good primer for those who haven't heard anything like this, yet.

 

[bpjod]

I'm a bit confused, so maybe someone can help me out here.

I was under the impression that for a while now all Model Ys delivered in (western) Europe were being built in Berlin. My daily observation during my commute in The Netherlands of trailers full of Model Ys going from east (Germany) to west (The Netherlands, Belgium, UK) supported this thesis. However, the last few weeks I've been seeing the opposite: trailers full of Model Ys going from west (the harbors of Rotterdam or Zeebrugge) to east (Germany, Denmark, maybe even Scandinavia).

The only conclusion I can draw: Shanghai is still exporting Model Ys to western Europe. Is this a new development or was I mistaken and have they been doing this all along?

Still unloading ships from the re-routing after the Houthi attacks? Just a guess, but I imagine that takes some time and may still be in progress.

As an aside, I’m visiting your wonderful country. The number of Teslas here in Amsterdam is mind-blowing. So many EVs of all sorts here, it’s really encouraging to see. I think all taxis I’ve seen are EVs. People (North Americans) who say charging of EVs doesn’t work if you don’t have a garage are either intellectually lazy or simply repeating oil-fueled propaganda because there are plenty of curbside chargers here. Haven’t figured out exactly how they work (private vs public, payment, etc.), but clearly they are. If a densely packed city like this can make charging infrastructure work for neighborhoods with only streetside parking, then it can be made to work anywhere there’s electricity, as long as the demand is there.

 

[Usain]

OK. To me that sounds a bit too much of a longshot to assign it much value.

I think so too. It's just a justification that says Dojo is worth doing.

 

[CitizenKane]

You're right. Like investing your entire wealth in a start up electric car company and a start up rocket company

Not sure I follow the correlation. Because Musk very successfully built those two companies in the past, every opportunity within Tesla, however far from a concrete commercial application, should be valued at its highest potential?

 

[Usain]

Dojo is never going to beat Nvidia across a wide suite of applications. It lacks the software ecosystem to be a general purpose solution. It’s
not going to be the “undisputed leader in AI hardware”. If all it ever does is be more efficient at FSD training than Nvidia it would be a huge win for Tesla.

Never say never. A software ecosystem can be built out rather quickly if Dojo performance leapfrogs NVIDIA. One thing I do know is that when there is an opportunity, the software industry is really good at filling the void.

But I think you are right overall. As you said, "If all it ever does is be more efficient at FSD training than Nvidia it would be a huge win for Tesla."

 

[FreqFlyer]

I'm a bit confused, so maybe someone can help me out here.

I was under the impression that for a while now all Model Ys delivered in (western) Europe were being built in Berlin. My daily observation during my commute in The Netherlands of trailers full of Model Ys going from east (Germany) to west (The Netherlands, Belgium, UK) supported this thesis. However, the last few weeks I've been seeing the opposite: trailers full of Model Ys going from west (the harbors of Rotterdam or Zeebrugge) to east (Germany, Denmark, maybe even Scandinavia).

The only conclusion I can draw: Shanghai is still exporting Model Ys to western Europe. Is this a new development or was I mistaken and have they been doing this all along?

I was under the impression that China exports the Model 3 to Europe, Makes sense Tesla China would send some Model Ys too to soak up Shanghai production capacity. Import taxes from China are low now, though the EU is making noises about imposing higher import tariffs from China to defend against the Chinese automakers.

 

[Drax7]

The justification for lowering car prices was affordability.
Frankly $12,000 for FSD 12.3.2 beta is unaffordable to the masses.
Waiting to see how this gets implemented.

 

[Krugerrand]

Not sure I follow the correlation. Because Musk very successfully built those two companies in the past, every opportunity within Tesla, however far from a concrete commercial application, should be valued at its highest potential?

Did I miss where he said ‘highest potential’? No, I did not. So you added that to change the context. Now look who’s playing the facts vs FUD game after pointing fingers.

Let me help you understand the correlation. Ignore, downplay, laugh at a thought, an idea, a plan coming from Elon/Tesla at your own investment/trading/options risk.

No, not all of them come to fruition. That’s a fact. But most are realized eventually and are game changers whether people see/acknowledge/understand it in the moment or not. That’s a fact.

 

[H Mak]

Not sure I follow the correlation. Because Musk very successfully built those two companies in the past, every opportunity within Tesla, however far from a concrete commercial application, should be valued at its highest potential?

You called it a long shot. There's your correlation

 

[DarkandStormy]

Adam Jonas merely spoke publicly about the approx. delivery numbers that everyone from Troy to Vin counters, to weekly China insurance data watchers are acutely aware of. The delivery numbers are going to be much closer to 425K than 475K because the high number has been stale for months and not corrected as data came in. The importance of the Jonas note is that he's saying that even at this reduced number, TSLA is worth $320 price target, based upon expectations of future profits, adjusted downward to create a present value.

 

[Stretch2727]

Never say never. A software ecosystem can be built out rather quickly if Dojo performance leapfrogs NVIDIA. One thing I do know is that when there is an opportunity, the software industry is really good at filling the void.

But I think you are right overall. As you said, "If all it ever does is be more efficient at FSD training than Nvidia it would be a huge win for Tesla."

From a FSD perspective I don't worry about the training compute. Somehow this will get solved as the hardware is not fixed.

I worry about the compute in the car as this is fixed. I would guess with 300K less lines of code it is fine, but on the flip side can the NN can get large and inefficient as you add training? This would be interesting to understand better.

 

[Xd85]

Gee I wonder why they don’t share anymore.

 

[Krugerrand]

The justification for lowering car prices was affordability.
Frankly $12,000 for FSD 12.3.2 beta is unaffordable to the masses.
Waiting to see how this gets implemented.

That’s an excellent point of juxtaposition.

What if the car itself is super inexpensive (in the scheme of things), then adding $12,000 isn’t out of the question.

Thought exercise - if the vehicle is manufactured like a Mattel toy and sold for cost that would make it pretty darn inexpensive. If the vehicle came with one ‘you can’t buy it without this’ option and that option was FSD -

Do we understand that implication in terms of affordability, profit margins, and paradigm shift?

 

[MartinAustin]

Gee I wonder why they don’t share anymore. View attachment 1032981

When you're cutting & pasting a CNBC article into this forum and not admitting it, you know your strategy needs a tune-up

 

[Singuy]

Dojo is never going to beat Nvidia across a wide suite of applications. It lacks the software ecosystem to be a general purpose solution. It’s
not going to be the “undisputed leader in AI hardware”. If all it ever does is be more efficient at FSD training than Nvidia it would be a huge win for Tesla.

Tesla has just barely managed to cross the minimum compute threshold for FSD. They still need to increase data and compute by another order of magnitude. Musk’s instructions to offer free FSD trials has as much to do with data as increasing take rate.
Training for full autonomy will probably require on the order of a million chips. At $30k per H100, that is $30 billion in NVDA chips. Now figure in the power required at about 500w per chip.

Each D1 chip probably costs Tesla about $5k and consumes half the power. Dojo will save Tesla tens of billions.

Sounds like made up numbers when Elon said compute is no longer a constraint, but you are here saying they need 90% more chips.

Training technique and software optimization trumps number of chips. Give Tesla 300k H100 in 2016 and they wouldn't know what to do with it as lots of the end to end training techniques were not even developed until 2020.

 

[Xd85]

When you're cutting & pasting a CNBC article into this forum and not admitting it, you know your strategy needs a tune-up

ouch. was literally wondering what the "2 specs it will take 3-5 years to catch up on" Would you prefer a direct link?

 

[MartinAustin]

The justification for lowering car prices was affordability.
Frankly $12,000 for FSD 12.3.2 beta is unaffordable to the masses.
Waiting to see how this gets implemented.

Just go to any car loan calculator. You will see that, over a 5yr loan, the payments for an extra $12,000 borrowed are about $200/month if the interest rate is 6%. (figures are approximate)

If the interest rate falls, the payments fall, making the monthly loan cost for lifetime FSD less than $200/month (and those payments stop after 5yrs)

It is far from unaffordable at its current price of $12,000. Tesla are probably tracking this type of calculation to make aure they can price FSD "lifetime" and "monthly" accordingly

 

[2daMoon]

Did I miss where he said ‘highest potential’? No, I did not. So you added that to change the context. Now look who’s playing the facts vs FUD game after pointing fingers.

Let me help you understand the correlation. Ignore, downplay, laugh at a thought, an idea, a plan coming from Elon/Tesla at your own investment/trading/options risk.

No, not all of them come to fruition. That’s a fact. But most are realized eventually and are game changers whether people see/acknowledge/understand it in the moment or not. That’s a fact.

Some TMCers may on occasion be afflicted with the Debbie Downer malaise.

When things get a little rough people will have a tendency to post up interpretations of their experience in a way that is unbalanced. Likely, because this will support a confirmation bias that aligns with the level of fear, uncertainty, and doubt they feel in the moment.

Though this does seem to be a fairly normal human reaction to such conditions, it is likely more of an intuitive one, and originating from our lizard brains. It is possible to thwart this effect by focusing on verification and balance of the data in order to turn an automatic reaction into a calculated response.

By being honest about the emotions being experienced, and realizing their influence, a person can lean the other way just enough to hold a more balanced view and be more thorough in their analysis. Thus avoiding a tendency toward reacting to an urge to credit confirmation bias with having more accuracy than it may actually have, before all the data has been processed.

This alteration of intuitive reaction to thoughtful response CAN be accomplished if practiced repeatedly. This becomes a tool folks can employ to recognize the effect and apply effort and a keen personal awareness in order to overcome these basic survival instincts which may mislead us into accepting only part of the picture as being the whole picture.

Hopefully, some will find this perspective useful.