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

[StarFoxisDown!]

Yes - I feel frustrated when people start talking about S/X "refresh". Its as if these people don't want S/X sales to recover. They are hoping personally to get a refreshed S/X and don't realize such rumors have an impact on sales.

I definitely feel it would have been better for Elon to tweet that they're taking the recently updated Raven Model S. Shouldn't be necessary to clarify about something so silly but it would have nipped the speculation in the butt

 

[sparcs]

SoylentBrown on Twitter
​

another interesting twitter exchange is about BMW vehicle fires. SoylentBrown on Twitter

"We have been working with two insurance clients on BMW fires - tons of those happening while parked at home inside the owners garage."

Didn't realize that was a thing until I googled and found this ABC news article:

As mysterious BMW fires continue, calls for investigation into possible causes grow

 

[KarenRei]

BTW, on a bit of an unrelated topic: recent discussions about Model S and overheating, the different motor types, etc has led to a realization: the new Roadster with the SpaceX options package should be able to regen directly to the COPV.

The power levels needed to refill the COPV in any remotely reasonable period of time are huge . It's going to need the power output of one or more of the main traction motors. This means a direct physical connection to the motors. Ideally in the rear, since the COPV is in the rear, and there's two motors back there that could be utilized rather than the one in the front. But if you have a direct physical linkage with the motors, then the regen torque coming in from the wheels can also go to the compressor. So you can regen directly to recompressing the COPV. Brake into a turn, rocket out of it

One thing of note is that Roadster with the SpaceX options package is going to want to be as PMSRM-focused as possible, rather than induction. Namely, because when you're pushing the vehicle hard, it's never going to "get a break"; the motors will be running at ~100% at all the times. If they're not delivering power to the wheels, then they're delivering it to the compressor, and vice versa. Induction motors - unless Tesla has significantly advanced the state of the art - would suffer from rotor overheating if constantly run at a 100% duty cycle. So any motors that the compressor is connected to, you're going to want to be PMSRMs.

 

[Beltsbear]

If you make an arbitrarily short vehicle life, you can come up with any negative number you like.

Half the problem is assuming after 150k miles that the battery has zero value. Even if the vehicle is done, the battery is still worth $4000 or so.

 

[Right_Said_Fred]

My inclination is towards a software update which includes "track mode". I don't think Tesla needs to do a refresh in order to bring MS back to it's former luster. If the MS can win against the Taycan on the Ring, I think that'd be enough advertisement to increase MS sales.

And they should call it Not-A-Turbo-Package.

 

[KarenRei]

Prep for the Iceland store/service centre launch is well underway.

Just got a post on our Tesla Facebook group from someone with connections to Tesla. Translating:

-------------
Tesla opens 9/9 at 9:00
Tesla Reykjavík opens on Monday at 9 AM. Everyone is welcome but there will not be any particular schedule. 8 Tesla employees will be present, two of them being employees of the Icelandic branch.

Show cars will not be arriving in the country in a timely manner. A Model 3 has been borrowed to have as an example in the store.

Further, I was asked to request that owners of 5-10 cars could be present on the first day. Owners will get some kind of signage to identify themselves as owners rather than guests. This would be a good opportunity to plug your referral codes.

The parking lot in front (east of the building) will be allocated for Tesla on this day. At 5 PM it's planned to drive a convoy to show the cars and it would be fun if as many people as possible would be with.

Follow Elon Musk closely on Twitter on Monday.

I'll perhaps get more information tomorrow, Saturday.
-------------

He then conducted a poll asking people when they'll be able to be present with their car and whether they'll take part on the convoy.

Also: Iceland is showing up on the Tesla list for some people but not others... they're clearly working on the site:

You can see our flag there (Ísland). You know, some people complain about how long, cold, dark and windy Iceland's winters are - but on the other hand, our flag is a big plus...

Thank you, I'll show myself out....

 

[humbaba]

Here is something interesting from SoylentBrown on twitter a while ago. Didn't get much attention.

"We observed one vehicle improving in ability when driven over the same roads over time"

SoylentBrown on Twitter

@BrownSoylent Aug 18

One final thought for everyone: We observed one vehicle improving in ability when driven over the same roads over time. But a 2nd identical Tesla did not improve... Until we repeatedly drove the 2nd car over the same course. Would indicate learning isn't universal/centralized...​

Here is my take on how that could happen.

Locally the car keeps track of where disengagements happen and where they don't. We know NN produces results with % probability. Using that probability the procedural code figures out what to do. For eg. at a particular place, NN might produce lanes that are straight with 70% probability and curved with 30% probability. If AP took the 70% route with some hesitation - and there was no disengagement, and this happens a few times, the procedural code would "learn" to take the straight path with little hesitation. Ofcourse the learning is happening outside the NN. Also, if there was a disengagement, that 70% probability will be discarded for 30% curved path. Basically each car can increase or decrease various probabilities that NN gives it over time. This can be easily done (from programming complexity perspective) and won't even take up too much space.

While I've not noticed the AP get better over time on my regular roads, I definitely see the AP being worse in unfamiliar roads with similar road conditions. This would point to the same kind of "learning".

Ofcourse, all this observation (all of us, including Soylent's team) could be just interpretation of something else that is happening which is not learning.​

This is wishful thinking. People don't understand neural nets and are constantly misinformed about them. For example, claims that they "learn" -- which they certainly do not (by any useful meaning of the word -- trade specific usages not withstanding). While it is more efficient to run a neural net program on an appropriately designed chip (like Tesla is doing) they can be run on regular hardware. The only difference is efficiency, that's it. It is just a program. The main difference (to an end user) between traditional programming and neural nets is, roughly, that traditional programming is deterministic and neural nets are probabilistic.

I wrote quite a bit, but I think I can shorten all of that to simply say this: no one (other than Tesla) has enough statistically meaningful samples to allow them to draw any conclusions about how autopilot performs. Due to random variation you (anyone), by chance, may encounter driving variation that appears to be learning. But there is no learning, you are just experiencing different parts of the set of possible outcomes. Others will have experiences, again strictly by chance, that would lead them to believe that autopilot is getting worse, even if it has improved overall from 99.99% to 99.993% reliability.

The long version follows:

Back in the day some programs used self modifying code (its been too long so I don't recall for certain, but I believe the game Doom was one example), but they didn't learn either. It was used because dynamically modifying the code paths improved execution efficiency. At the cost of complexity, but that is a longer story. In any case, these days a CPU will disallow modification of code in memory, but there are details and caveats beyond the scope here. My point is that -- in practice -- programs use static code that does not modify itself. Execution changes based on parameters changing, but not the code itself changing. And, yes, in some cases it can be difficult to separate the two.

But Tesla's code for autopilot is not self modifying. It is a static binary that only changes when a new version is compiled. That new version is taking into account improved data sets ("training") and users only see an actual change when they download a new version.

So why do people see things that they interpret as the car "learning"? Well, aside from the widespread misinformation about neural networks there is the fact of how neural networks work. When I write a program it is, essentially, deterministic. If I'm writing a game I can use pseudo random numbers to vary things up, but the code itself will always give the same results for the same inputs.

But neural networks don't quite work like that. For starters, you are incredibly unlikely to actually be feeding in the same inputs. Driving down the same road you will have slightly different speeds, the sun will be in a different position, there will be different cars, numbers of cars, pedestrians, etc., all conspiring to give a different input from the cameras despite it being the same road.

Which takes us to the core of neural network programming: rather than being deterministic it is probabilistic. Thats a high level view, not an implementation level injection of randomness. The idea is that, given a particular situation (say, identifying a set of pixels as representing a car) you want the neural net to arrive at the correct result 99.999% of the time. You take your data set and you "train" your model. I think it is more accurate to call it "iterative compilation" but I didn't make the terms up. For starters, because the process is done repeatedly until a goal is reached. In unbiased (or "physical") rendering this is done until you are satisfied with the image quality. For neural networks you do it until you hit the accuracy level you want. (Of course, sometimes your model and data set aren't up to achieving it so you have to collect more data or improve your labeling or improve your model or ..., but that's a detail.)

The point of all of this is that at the end of the day it there is a probability that the car will perform as desired with autopilot in any given situation. If it goes from 60% to 90% you are likely to notice an improvement (about 75% fewer undesired outcomes), but you might not. If it goes from 99% to 99.1% it has improved, but you may very well not notice (about 10% fewer undesired outcomes, and already scarce to start with). And, if you experience something in the remaining 0.9%, you will have a worse outcome.

To get an independent, black box, sense for how autopilot performs you not only need to use the same stretch of road, but you need to ensure the same time of day, the same environment (other cars, etc.), that you engage it with your car in the same position and speed, and then run it thousands of time to get a statistically meaningful sample. At the end of the day Tesla is the only one who could even remotely do this kind of real world validation of their accuracy targets.

 

[EVNow]

You can see our flag there (Ísland). You know, some people complain about how long, cold, dark and windy Iceland's winters are - but on the other hand, our flag is a big plus...

Plus ? More like a cross knocked down by the wind

 

[EVNow]

This is wishful thinking. People don't understand neural nets and are constantly misinformed about them.

I would say the same thing, if this was not SoylentBrown. He (apparently) runs a team of hackers/investigators that learn in depth about various models for OEMs / large investors.

 

[humbaba]

I would say the same thing, if this was not SoylentBrown. He (apparently) runs a team of hackers/investigators that learn in depth about various models for OEMs / large investors.

It doesn't mean he understands neural networks. In fact, my lengthy discussion covers why he is wrong [edit: and how people fall into the trap]. Don't believe me? Fair enough. Is Karpathy a good enough expert for you? Watch his presentations on how autopilot works, on what he calls "programming 2.0", etc.

[edit: there isn't even a mechanism to enable any sort of self-modifying/local customization. Your car definitely does not have Tesla's labeled data, nor does it have the facility to "train" (iteratively compile) to produce new code. It is just foolishness.]

 

[EVNow]

It doesn't mean he understands neural networks. In fact, my lengthy discussion covers why he is wrong. Don't believe me? Fair enough. Is Karpathy a good enough expert for you? Watch his presentations on how autopilot works, on what he calls "programming 2.0", etc.

[edit: there isn't even a mechanism to enable any sort of self-modifying/local customization. Your car definitely does not have Tesla's labeled data, nor does it have the facility to "train" (iteratively compile) to produce new code. It is just foolishness.]

That's why the question. He understands NN quite well, from my reading.

I think local "learning" without NN modification is quite possible. See how I think it could be done in my post above.

 

[humbaba]

That's why the question. He understands NN quite well, from my reading.

I think local "learning" without NN modification is quite possible. See how I think it could be done in my post above.

I didn't say that local learning couldn't be done. But it is a plain and known fact that Tesla's autopilot is not doing it. People fall for this all of the time and it simply isn't happening. I used generalizations about neural networks because... well, they are generally true and because they are specifically true for what Tesla is doing.

Maybe in the future, but definitely not now, and not on the FSD chip.

@Artful Dodger I think is the one who said their next FSD chip is being designed to enable local learning. Maybe -- but that would just be for efficiency purposes. It requires a different approach than what Tesla is doing and, to achieve performance goals, would certainly require new hardware.

edit: as to your proposal? I started to write a response, but basically I'd be trying to teach you how neural networks work (at a high level) and I'm not that person. If you are interested, I strongly recommend you watch Karpathy's presentations. I found it to be invaluable in advancing my understanding.

 

[JRP3]

+1
Precool the pack for max continuous power (track mode) instead of preheat the pack for max instantaneous power (launch mode). Plus raise the power limits based on Raven fleet data.

So this makes me think about how Tesla can certainly beat the Taycan time if they want to. Cooling is mainly a concern for a vehicle that needs to achieve the longevity of hundreds of thousands of miles. Take the same vehicle with the same components and boost performance so that they only have to last for 50 miles max and you can push the power levels to plaid and beyond. Doesn't matter if the pack is completely dead after the circuit, run every thing as a dead short and remove the software limitations!

 

[KarenRei]

So this makes me think about how Tesla can certainly beat the Taycan time if they want to. Cooling is mainly a concern for a vehicle that needs to achieve the longevity of hundreds of thousands of miles. Take the same vehicle with the same components and boost performance so that they only have to last for 50 miles max and you can push the power levels to plaid and beyond. Doesn't matter if the pack is completely dead after the circuit, run every thing as a dead short and remove the software limitations!

Tesla has no need to cheat. The basic stats of the Raven S are enough.

 

[humbaba]

Tesla has no need to cheat. The basic stats of the Raven S are enough.

I agree, but I'm not all that convinced that we are going to get a lap time next week. If so, then its all a moot point anyway.

1. Elon doesn't actually say they are going to do a record lap time, he just lets everyone assume that
2. We know that a Model S is in fact at Nürburgring
3. Reportedly, Tesla hasn't reserved a slot. As amusing as it is to contemplate it being reserved under Farty McFartyFartPants, I somehow doubt that the management doesn't know who is paying for a slot. So it seems unlikely that Tesla actually has a reserved slot.
4. But, Tesla doesn't need a slot, they can use the track the hour before and hour after it is open (manufacturer pooling). However, this isn't exclusive and so while it can be used to learn the track and collect data it isn't good for posting a record time

So what I think is going to happen is that Tesla will run the Model S around the track and collect data. They'll encourage all of the media hoopla (free advertising). They may well post video of the Model S on the track. They may announce/start selling/otherwise release a track mode for the Model S. But unless they get cooperation from other manufacturers (which, for some reason I can't quite put my finger on, I find highly doubtful) they will be hampered in achieving a record time.

Now, if -- despite the constraints -- Tesla does set a record lap? Sure, Elon will tweet/retweet that and there'll be a media frenzy. I'm just not very sanguine about Tesla doing so under the constraints.

 

[Intl Professor]

Plus ? More like a cross knocked down by the wind

Don't tread on others' flag!

 

[Prunesquallor]

"Tesla did not send us a record request and did not rent an exclusive time slot." The representative further said that the Nordschleife is "fully booked through the season," and added, "I think that there is no availability for Tesla to set up a record attempt in the next days."

come on Elon.

Wow. This smells like “pay me”.

 

[MP3Mike]

Half the problem is assuming after 150k miles that the battery has zero value. Even if the vehicle is done, the battery is still worth $4000 or so.

The other 75% of the problem is that Elon/Tesla says that the current battery packs are good for 500k miles. (With the next generation packs being good for 1 million miles.)

We have certainly seen Model S/X packs go ~400k miles already.

 

[SO16]

I agree, but I'm not all that convinced that we are going to get a lap time next week. If so, then its all a moot point anyway.

1. Elon doesn't actually say they are going to do a record lap time, he just lets everyone assume that
2. We know that a Model S is in fact at Nürburgring
3. Reportedly, Tesla hasn't reserved a slot. As amusing as it is to contemplate it being reserved under Farty McFartyFartPants, I somehow doubt that the management doesn't know who is paying for a slot. So it seems unlikely that Tesla actually has a reserved slot.
4. But, Tesla doesn't need a slot, they can use the track the hour before and hour after it is open (manufacturer pooling). However, this isn't exclusive and so while it can be used to learn the track and collect data it isn't good for posting a record time

So what I think is going to happen is that Tesla will run the Model S around the track and collect data. They'll encourage all of the media hoopla (free advertising). They may well post video of the Model S on the track. They may announce/start selling/otherwise release a track mode for the Model S. But unless they get cooperation from other manufacturers (which, for some reason I can't quite put my finger on, I find highly doubtful) they will be hampered in achieving a record time.

Now, if -- despite the constraints -- Tesla does set a record lap? Sure, Elon will tweet/retweet that and there'll be a media frenzy. I'm just not very sanguine about Tesla doing so under the constraints.

Interpret this tweet how you want...

Twitter

 

[JRP3]

Tesla has no need to cheat. The basic stats of the Raven S are enough.

Probably, but if I were Tesla I'd tweak the software to allow higher current draw, greater voltage swings, and higher temperature limits for that test. Why not?