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

[CyberShy]

Some things far enough out of my knowledge base that, while I grasped their essentials when first discussed, I realize a usable understanding of them and their important distinctions once again is elusive. So -

Someone who is confident s/he can clearly explain Neural Net vs End-to-End please step up to the plate and assist me and I rather suspect others.

Thank you.

No. Please don’t. Not in this thread.

 

[Krugerrand]

This I find astonishing. Most of Isaacson's biographies are written after the death of the subjects, Henry Kissinger and now Musk being the two exceptions. I would have expected Musk and close associates to have early copies of the book to weed-out such inaccuracies, seems like a very obvious thing to do

Now if Musk said "I don't want to read it, just go", then that's very sloppy as you can be sure the usual detractors will be cherry-picking every little morsel they can to throw shade at Musk and his ventures

Shouldn’t be shocking. Everyone incorporates their own biases/realities even in the face of non-questionable facts. People literally can’t be or do without interjecting their own ‘interpretation’.

The very first excerpt of this book required ‘clarification’ immediately. Now we’re on excerpt ?three? and there’s been continued ‘clarification’.

I doubt even Elon could write his own biography without having to clarify, explain or outright contradict himself.

 

[Singuy]

Shouldn’t be shocking. Everyone incorporates their own biases/realities even in the face of non-questionable facts. People literally can’t be or do without interjecting their own ‘interpretation’.

The very first excerpt of this book required ‘clarification’ immediately. Now we’re on excerpt ?three? and there’s been continued ‘clarification’.

I doubt even Elon could write his own biography without having to clarify, explain or outright contradict himself.

Haha considering the number of tweets from Elon people misinterpreted which required a clarification in which people also misinterpreted, I would say this is true.

 

[mongo]

Some things far enough out of my knowledge base that, while I grasped their essentials when first discussed, I realize a usable understanding of them and their important distinctions once again is elusive. So -

Someone who is confident s/he can clearly explain Neural Net vs End-to-End please step up to the plate and assist me and I rather suspect others.

Thank you.

Spoiler: Rough explanation

Neural Net is orthogonal to End-to-End
E2E is a method to train a NN in one pass versus breaking the task into discreet steps

Train a NN to cook
Non-E2E:
Train a NN to recognize letters
Train a NN to recognize words
Train a NN (or C code) to handle grammar, semantics, actions, order of operations, sub operations (crack an egg)...
Link them together

E2E:
Train one NN to make food (output end) based only on images of recipes (input end)

 

[FastEddieB]

As far as advertising goes…

Lots of “normal” folks are not even aware of the CyberTruck’s existence. I’ll mention our order for one and people ask, “What’s that?”. Also that report of police officers having no idea what they were seeing on the car hauler. Advertising will enhance awareness for those not already Tesla fans, which I think will be a net positive.

 

[Singuy]

Spoiler: Rough explanation

Neural Net is orthogonal to End-to-End
E2E is a method to train a NN in one pass versus breaking the task into discreet steps

Train a NN to cook
Non-E2E:
Train a NN to recognize letters
Train a NN to recognize words
Train a NN (or C code) to handle grammar, semantics, actions, order of operations, sub operations (crack an egg)...
Link them together

E2E:
Train one NN to make food (output end) based only on images of recipes (input end)

Without C code guidance which is a hard limit, people runs into the problem of subtlety.

For example.

Red light is stuck on red. People drivers after 10 mins decided to label the red light as a stop sign or else you block traffic.

Training this will result in the Tesla be more human like, however this can lead to teslas running red lights. Hard C code will stop the car, end to end training may have the car proceed like how a human would. However would fixing this edge case introduce massive regression? This is now the needle Tesla has to thread with end to end.

 

[2daMoon]

Might try that yourself, pot and kettle stuff. All states have legislatures with members that own dealerships. Very weak rejoinder IMO. Try harder.

My tendency is to acknowledge correction and learn from it. Generally, I'll show appreciation for those who point it out.

Thanks for the feedback.

As for legislators owning dealerships everywhere, that was the point being made. This isn't unique to any particular state.

The Auto Dealers have formed associations in many states to protect their cartel from being swept aside by direct sales from manufacturers. They now exist primarily to keep these laws on the books. Laws that were originally designed to protect them and now have been exposed by Tesla's direct sales in such a way that those avoiding the traditional experience rave about the alternative.

The customer has already been provided a choice to own a Tesla in these states, regardless of these laws. Thankfully, the dealership cartels have no such foothold in federal lawmaking and cannot prohibit sales through interstate commerce.

It seems clear that the days are numbered for the power of these state statutes, and for the traditional dealership experience itself. This won't change tomorrow, but it will change once the old model wears thin enough that it can no longer support itself.

At best these statutes will be abolished by the legislature. At worse, they will become Blue Laws because there won't be anyone left who cares enough to enforce them. Time will wound all heels who use this tactic to perpetrate this elaborate dance designed to lever profit from their victims, the customers.

Those archaic laws aren't enough of a deterrent to represent a real problem that prohibits Tesla's sales from keeping pace with production in those states where these laws exist. The Dealership Cartels create FUD by bringing up the topic and publishing misinformation that leads people to the idea that Tesla cars can't be bought in certain states.

Whining about this over and over in a forum where many/most participants already know this probably won't make much difference to the timeline of change. However, doing so can further propagate the FUD the cartels want to spread when posted repeatedly, with such unnecessary drama. Anyone who does this has clearly bought into the FUD and is doing the work for the cartel to help spread misinformation.

This is the fact of the matter, these laws have no significant effect on the growth of Tesla and of TSLA, regardless of the color of either the pot or the kettle.

Some say this is bullish.

 

[kbM3]

Indeed. There’s another reason too. The intervention we saw on Elon’s FSD 12 build was the car stopped at a red light. Left turn signal turned green, and the car made a mistake and started moving forward on the still red light. NN that do not have a rules overlay will make mistakes like this and it isn’t clear if the car would have ever stopped in time because it had NO training data of good human drivers proceeding from a stop on a red light.

There’s only three ways to fix this FSD 12 mistake.

1. Train with more good driver videos so that it doesn’t make that mistake again. But there’s no certainty you’ve got it nailed. There’s always a chance it will will screw up and then the system doesn’t know what to do once it is moving in an intersection on a red since it’s never seen that situation before. Absent cross traffic, it is likely to just keep running the red (and risk getting t boned by a fast inbound car).

2. Train (probably in a simulation) situations where the system initially made a mistake like proceeding on a red, and show it what to do (like stop suddenly). But you need lots of intervention data to find out where the system screwed up.

3. Personally, I don’t think either 1 or 2 will work well enough to be perfect. You need to do what humans do. Experienced drivers are usually driving on a human equivalent of auto-pilot, where you can be deeply thinking of something, or on a complex phone call, and your unconscious NN just drives the car. Indeed, I’ve been in that state, and guess what, sometimes my unconscious auto pilot NN will make rules of the road mistakes like starting on a red light, or the one that 95% of all drivers do, not follow rules on a 4 way stop sign when all directions are busy. Anyways, what happens in the really bad situations, like starting on a red light, my conscious part of the brain, the part that knows road rules, intervenes and takes over from my unconscious auto pilot.

Similarly, I think Tesla will have to have C code looking over the shoulder of the NN and take over when the NN makes bad mistakes. The C code can then give control back over to NN when it has handled the situation. The nice thing about this emergency supervisor is that Tesla pretty much already has that code in the form of v11. You just have to add take over and give back control code based on bad NN decisions.

4. OR program a more complex NN like an LLM level that understands the rules of the road and that part looks over the shoulder of the driving NN.

The big mistake was NOT running the red light. The far larger problem was attempting to drive into the cars which were turning. The car should never do that whether the light was red, yellow or green.

 

[Unpilot]

I think that a degree of advertising makes some buyers feel more confident about their purchases. Could actually help

Driving the car makes me feel more confident about my purchase.

 

[kbM3]

George Hotz, to his credit, always insisted that end-to-end AI was the way to go since the very beginning of Comma AI. Thankfully, he has too much integrity to sell out to other car companies and much to the advantage of Tesla, didn’t have much capital to accelerate the process.

I thing EVERYONE knew the end goal was video in and controls out. In fact, Tesla stated that at the first Autonomy Day in 2019.

where Hotz was wrong, was his belief the best path was to go straight there (just like Waymo & Cruise going straight to L4). Tesla knew you needed the incremental approach and learning and improving and generating positive cash flow as you go.

 

[mongo]

The big mistake was NOT running the red light. The far larger problem was attempting to drive into the cars which were turning. The car should never do that whether the light was red, yellow or green.

It didn't attempt to drive into other cars. Oncoming traffic started at about the same time as the Tesla (3 seconds after green transition), and the car started decelerating before Elon intervened (if the UI is reliable).
Going, blue tentacle:

Crossing traffic, tentacle is gray

Tentacle is gone, decelerating

Tentacle is gone, decelerating, still blue wheel (FSD)

 

[SOULPEDL]

Anybody notice the red "Toyota" sign on the outfield wall? I didn't think so...

Toyota who?
But that Labatts beer commercial at the end - that I did notice.

 

[unk45]

Apologies, it appears I have made an ASS of myself. Finance is riddled with acronyms.

Here's the definition:

• Residual Value ("RV"): A notional future expected resale value of the vehicle set by the lessor at the time the lease is entered into - It's their estimate of what they reckon the vehicle will be worth at the time the lease ends.
  • It's important to the lessee as the monthly payments have to cover the difference in value between the purchase price and the RV + interest on the total value
  • It's even more important to the lessor because once they get the vehicle back at the end of the lease they have to sell it. If they set the RV too high they lose money on the sale, too low and they will be uncompetitive against other lessors when calculating monthly payments and not get the business.
  • It's called a balloon payment if you have a loan rather than a lease - but on a loan the borrower is usually on the hook to pay it regardless of it's accuracy.

Most people, including people in the industry accept your definition because it is logical. Logic, though, has nothing to do with Residual Value. Residual Value is the contractual defined future value of lease collateral. That is it! Nothing else!
why is both precision and accuracy important in this definition? Because Residual Value can be and frequently is, subvened, so is sometimes higher than any expected future resale value.
Next, the basis of payments includes amortization of the purchase price including all fees over the tenor of the lease.
A device called Money Factor is used to establish actual excess payment beyond the amortization. It is not technically an interest rate, although it absolutely is effectively equivalent. Why is that distinction important? Tgreen reasons: first this avoids usery restrictions; second, OEM money factory subvention is common; third, this allows front loading of total effective interest.
The net effect of these arcane and pedantic definitions is to make early terminations seriously profitable. Since early terminations typically range between 30-50% that can be hugely profitable because these arcane definitions make all the risk on the lessee. Whether by collision damage, early trade or something else these are often the major lessor profit centers.

Tesla, as many of us know, plays none of these games, so often seems not-competitive.
Finally, this is US retail leased. US commercial leases, both finance and operating, are not atball like this. Leases eligible for tax treatments are usually, but not,always, better than this. Company vehicles are not like this.

One anecdote: some years ago my company was commissioned to develop lease products for some new models. A few months later one senior member of my team leased a new car using our product. He signed for extended warranty (100% dealer markup plus captive 55% markup), plus accepting an effective 6% increase in effective finance costs. Even people with technical competence can get scammed on these products.

Moral: understand that Tesla is not part of such practices. Understand also that it is very, very difficult to understand lease risks. Understand that dealer F&I people rarely understand these topics, but they live on selling them. The very fact that many of us will insist I’m wrong, that they do know leases have a name in F&I circles. They’re called ‘marks’.

 

[UkNorthampton]

I enjoyed a video on USA new/used car prices, availability, days of inventory, a bit of Tesla (way ahead of others, threat to legacy).

I'm not from USA, so no idea how truthful, but interesting to me. Main point is that new prices have increased so much over the last few years. My instant thought was how a new lower-cost Tesla might affect the market.

Seems that legacy (including distributors & dealers) just keep setting up open goals for Tesla to score

1) Cancel chip orders
2) Unable to make cars
3) Raise prices sky high - MSRP, dealers etc
4) Restrict supply of lower cost models & trims.
5) Use excuses of chips/components (even now) to restrict supply or play bargaining games with UAW
6) Vacate the lower-cost market to Tesla
7) Really annoy previously brand-loyal / low-switching customers to the point where they cross-shop & are open minded to alternatives

Also - they claim Toyota margins are 3% in North America vs ~ 13% elsewhere - hence restricted supply.

Toyota, Ford, & GM SCREWED Buyers, Now They're Paying The Price

ð CarEdgeâ® https://caredge.comð Join the CarEdge Community: â® https://caredge.com/community ð Search for new and used cars: â® https://my.caredge.com/buy ?...

www.youtube.com

 

[SOULPEDL]

I thing EVERYONE knew the end goal was video in and controls out. In fact, Tesla stated that at the first Autonomy Day in 2019.

where Hotz was wrong, was his belief the best path was to go straight there (just like Waymo & Cruise going straight to L4). Tesla knew you needed the incremental approach and learning and improving and generating positive cash flow as you go.

OT

Spoiler: Human Learning.

I like to compare NN to humans in every way imaginable, both good and bad. I don't know the various process details on the software side, but "Visual Learning" was my thing - for humans, in Factories anyway.

I see a lot of analogies to "Spiral Learning" (which builds on existing new tasks through application of tasks just learned). Chunking is also an obvious learning term to break down the task into more manageable "baby steps" so to speak. Our brains can manage and process smaller volumes of data just like Dojo. (We saw this with the incremental nature of the FSD features over the past few years.) As the task complexity grows, the sub-tasks are practiced repeatedly and it's this building block nature of learning that is the proven method for humans.

Because there's an obvious break in knowledge required between Control and everything else, it makes sense that Control is the last "lesson" for this NN apprentice. Practice and experience comes next, building on the existing mental model that will change subtly over time. It's so human like that when these are driving themselves around, I would even go as far as to believe they will be sentinel.

 

[M|S|M MYP]

Driving the car makes me feel more confident about my purchase.

Hanging out at TMC makes me feel more confident about my purchase (Only on certain days of course)

 

[Max Plaid]

Shouldn’t be shocking. Everyone incorporates their own biases/realities even in the face of non-questionable facts. People literally can’t be or do without interjecting their own ‘interpretation’.

The very first excerpt of this book required ‘clarification’ immediately. Now we’re on excerpt ?three? and there’s been continued ‘clarification’.

I doubt even Elon could write his own biography without having to clarify, explain or outright contradict himself.

Sure, but there's a vast difference between "withdrawing a service" to "not activating a service", especially for something so monumental and sensitive

Anyway, well all know that Musk could cure cancer, eradicate world hunger, solve the population crisis single handed* and turn water into wine/IPA, and he'd still be "Spaceman Bad" for the more simian-minded out there

(*this one he's actually trying his best to deliver)

 

[aubreymcfato]

Lex would be a poor choice. Urban would be great IMO.

Lex is not a writer, if I'm not mistaken, and Urban wanted to write a long blogpost and it took him 6+ years to write a fairly short book.
As someone who dabbles in writing myself don't underestimate the absolute distance there is between these two and Isaacson. He is a seasoned biographer, and has done it several times. He's a master of his craft, he can write 600+ pages book in a year and probably has a good writing/researching process with a lot of people who help him. You remember the story about the prototype and the manufacture? It's the same but with written pages. I only read The Innovators and at the time I was in awe about the bunch of stories he could weave together, the structure of the book and the underlying motives of competition VS collaboration. It's hard stuff to write a very good book.
He also is an international established name in publishing, and has very good reputation. Very good choice, IMHO.

 

[SOULPEDL]

Headed to see some new family in Florida and reserved a Model Y rental... or Equivalent... is that possible? It's in Tampa Florida from Budget, so Hertz isn't the only one playing in this space I see.

I doubt it has FSD so this will be difficult... you think I'm kidding? There are MANY times I much prefer to let the car drive. Especially if a good song comes on and I need to drum while driving (DWD).

Anyway, I'm about to introduce Tesla to a part of Fla that likely doesn't see many around. Superchargers thin out north of Tampa I'm noticing. Wish me luck! Maybe I should grab a Gators sticker and put it on the rental.

 

[lascavarian]

It seems clear that the days are numbered for the power of these state statutes, and for the traditional dealership experience itself. This won't change tomorrow, but it will change once the old model wears thin enough that it can no longer support itself.

Thanks for your detailed comments. Generally, I don't see these statutes going away unless there is a price to keeping them. I would be happy to be wrong. Some states have liquor control boards even though there are breweries on every corner. There is no cost to keeping the antiquated LCB laws in place.

I think Tesla can extract a cost to the dealership protective laws but only if they are willing to do it. Probably not. It would be nice if some of the others directly affected kept the issue active focusing on dealerships as opposed to states. YMMV