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

[lklundin]

I would love to believe the propane theory, but a propane thank (even a small camping type one) is a difficult evidence to get rid of once the crime is done...

If arson was indeed committed then that impacted the SP much more strongly than subsequent discovery of evidence of such a crime would be able to make it rise again...

Some months ago a Model S started burning spontaneously while driving slowly on a street in some city in California. While superficially similar, this recent one was actually very different.

The CA fire had smoke being vented from specific openings, so appearing as jets that changed from smoke to flames as the battery heated up.

This recent one had smoke coming more or less uniformly from the sides and front (where there is no battery pack) for just a few seconds, after which the fire turned explosive.

I sincerely hope that there will be a proper investigation of this fire - and I guess with three cars burned and likely lots of smoke damage in the building, anything else would be surprising.

 

[ZachF]

So I am 2 hours in and they still haven't mentioned pothole avoidance!!

They're avoiding that topical pothole...

 

[Ad Astra]

It was quite clear from his demeanor during the investor presentation that Elon appeared absolutely maddened by the fact that Tesla’s lead on multiple fronts still isn’t obvious to most people. As was pointed out during the event, autonomy and electrification are not mutually exclusive. Even the best autonomous ICE vehicle would be uncompetitive against the worst autonomous EV, therefore, the faster that autonomy is realized the faster ICE vehicles are pushed towards obsolescence. If we agree that electrification of the vehicle fleet is an inevitability on the path towards full autonomy (or vice versa), then Tesla’s other “moat” becomes all the more important.

The other moat:

I’ve often marveled at the fact that many can’t seem to grasp that Tesla acts in accordance with whatever increases demand for batteries. It’s quite clear that in order for a true solar electric economy to blossom the problem of the intermittency of renewable energy sources has to be solved and, thus far, managed static storage systems deployed at scale appear to be the best solution. Now, in order to realize the cost efficiencies necessary to drive adoption of these storage systems, batteries must be produced at incredible scale. It then follows that in order to achieve such scale there must be an equally high volume application and the requisite demand for such an application to merit going down the rabbit hole of complex large scale manufacturing. As it happens, vehicles require a lot of batteries in a setting that is incredibly demanding (due in large part to the immense variance of the loads placed upon batteries in a vehicle application) and this leads to dramatic improvements in the core technology (i.e thermal management), safety and takes advantage of economies of scale. Vehicles also make for a much more compelling consumer narrative* as the public has a tangible product that can illicit an unforgettable limbic response and organically initiate the supply/demand cycle. From here, it’s quite trivial (relatively speaking) to manufacture and deploy static storage systems at the scale necessary to disrupt the behemoth that is the energy industry. All the while, continuing to satisfy demand for the vehicles leads to an interesting infrastructure dilemma because our current energy generation and distribution system is ill equipped to handle a majority population of EV’s. This, in turn, will inevitably generate demand for Tesla’s energy products as it will likely represent the most cost effective solution to transition to decentralized micro grids because, remember, Tesla produces more batteries out of one factory than the rest of the auto industry (and soon the world) combined...IMHO, not an advice, boiler plate, boiler plate.


*Personally, I find battery tech incredibly compelling but I’ll concede that I am likely in the minority)

 

[SOULPEDL]

a simple answer might be like
There are well over 400 different types of vehicles sold in the US listed on Kelly Blue Book
they all drive differently, sensors would be randomly placed, a great way to get total gibberish

Tesla has around 450,000 fairly identical vehicles that act fairly the same, with a few dozen sensors each. all fairly identically placed reporting with not a lot of +/- on placements of readouts
( when target shooting, you were very precise, all close together, but the target was _over there_)
this removes a lot of variables

There u r. Where ya been hiding? We need more stock talk here! Bring it on. What will tomorrow look like?

 

[jerry33]

no, you simply put all the laws into handwritten rules. detecting which state you are in is trivial. Deer jumping in front of the car has nothing to do with this cause it's not part of the law.

You need to distinguish several parts of the system, the perception engine, that understand the surroundings (recognize a deer and noted its movement), and the prediction engine, which predicts the possible movement of moving agents identified by the perception engine (noted the deer maybe in your way), and the planning engine, which create multiple action plans and pick one. The first two are pure NN based, the third one is a mixed bag. knowing the deer maybe in your way and thus stop is most probably handled by NN.

The manually specified rules (coming from road laws) goes into the third part, after multiple actions are produced, right before they are choosen, the rule change the risk factor of different actions.

Sorry, we're talking past each other. I was only saying that an entirely rule based system doesn't work because it's not possible to code a rule for every situation.

 

[SOULPEDL]

They're avoiding that topical pothole...

Ya and summon? HW3 upgade schedule... Lots in the dark still.

 

[jerry33]

It's a decaying exponential
Growing fleet * diminishing returns = linear improvement?

It would be except that in the presentation it was said they could filter out the items which already had sufficient data. I believe it was filtered out automatically, but I could have misremembered that part.

 

[scaesare]

Because Tesla hasn't really started working on full self-driving. I followed the tech presentation in detail. They're nowhere close.

So you've said.

The issue is, I think by everyone's definition but yours, they have.

You've discussed your take on their approach (expert driver knowledge distillation) and why they must not be working on it (no job postings for expert drivers).

However, they may just be leapfrogging past the "human trained" bit, and even if they aren't, you (we) don't know who they've hired on a direct basis.

Not to mention that the presentation clearly demonstrates technology that must be part of a self driving solution (i.e. object recognition, categorization, environment mapping, path predition, etc...), so by extension those aspects are directly working on self driving. This is true even if they aren't where you would want them to be (either in timeline or approach) when it comes to other aspects such as policy, etc...

 

[lklundin]

"Sexual harassment" allegations have been shown to be mostly false in this day of age; if anything, they are a signal of someone worthy of being attacked, and probably signifies they are a good person.

Before this day and age, how often would a young woman even attempt to allege sexual harassment from a powerful man?

 

[neroden]

Eroden, what is you position on Tesla?

I have probably half my personal wealth in Tesla. (Depending on how you calculate some of the options.)

I think EVs are going to take over the world as quickly as they can be manufactured.
I think Tesla can manufacture EVs substantially more cheaply and in higher volume than anyone else.
I think they can also manufacture stationary batteries substantially more cheaply and in higher volume than anyone else.
I believe that demand is not an issue, because nobody else is mass producing enough EVs
I believe the Supercharger network is a huge moat.
I believe they have the best driver assist features on the market and will for the forseeable future.
I think they have a huge premium value in their brand.

I think they are in an economies-of-scale business, where the key is to produce enough units to cover the high fixed costs. They have not yet managed to do this, and this is the critical difference between doing poorly financially and doing extremely well financially.

I am very worried about the fact that they have not managed to reach 10,000 Model 3s per week.
I am worried about the appalling service and sales communications, which seem almost baked into corporate culture, and are extremely alienating.
I am worried about the inability to provide geographically distributed service centers.
I am worried about their blase attitude toward basic software issues which customers care about, like getting the media player working. (There's also a set of complaints about Powerwall: people just want to be able to set simple manual instructions for charging and discharging, and the "smart" Powerwall modes are dumb and don't work right. It's not getting fixed)
I think the robotaxi nonsense will not amount to anything, financially speaking.

What would you change?

Apart from the list already mentioned above, I'd get Musk to stop claiming that he's going to have some miraculous thing in two years. Just stop making timeline claims. Timeline claims are OK for things you've already DONE ("We will release that next month", "Our prototype is in testing now") but not for things in R&D.

 

[SOULPEDL]

a simple answer might be like
There are well over 400 different types of vehicles sold in the US listed on Kelly Blue Book
they all drive differently, sensors would be randomly placed, a great way to get total gibberish

Tesla has around 450,000 fairly identical vehicles that act fairly the same, with a few dozen sensors each. all fairly identically placed reporting with not a lot of +/- on placements of readouts
( when target shooting, you were very precise, all close together, but the target was _over there_)
this removes a lot of variables

So partner with a high volume mfr and start gathering. Otherwise how will they ever catch up? Just playing Dev Adv here. Are there seriously no options at all? Hard to believe, but if true, I'm for sure going to be rich!

 

[neroden]

Not to mention that the presentation clearly demonstrates technology that must be part of a self driving solution (i.e. object recognition, categorization, environment mapping, path predition, etc...), so by extension those aspects are directly working on self driving.

I concede the linguistic point. I also concede that Henry Ford was working on self-driving when he developed the Model T, as obviously it was necessary to develop a vehicle in order to have it do self-driving.

 

[MarcusMaximus]

Currently, from what I can get from the presentation, they're using a combination of manual path planning (doesn't scale) and "copying the average driver" (who isn't good enough at it). They'll have to do at least one more iteration on their entire path planning scheme (toss out what they've got and try again with a slightly different approach). Oh, they will do that, but it'll take time...

I don’t think this is true... Karpathy at one point highlighted a video showing path planning on a curving road, with the path extending to occluded segments of road not on-screen. I believe the statement was that the network has learned how to path-plan taking into account what the road will likely do ahead, rather than just what’s visible/labeled. And that this is what is enabling cloverleafs in current firmware.

 

[neroden]

I don’t think this is true... Karpathy at one point highlighted a video showing path planning on a curving road, with the path extending to occluded segments of road not on-screen. I believe the statement was that the network has learned how to path-plan taking into account what the road will likely do ahead, rather than just what’s visible/labeled. And that this is what is enabling cloverleafs in current firmware.

I believe this was by copying drivers.

 

[BioSehnsucht]

This is incorrect. Again I'm only pointing out things that are clearly incorrect. I need every Tesla fan who have the absolute truth to their disposal. Nvidia performance at batch size of 1 has exponentially increased compared to Drive Px 2. Again Tesla is comparing a 3 year old, 3 gen old chip to a chip from 2019 rather than comparing a chip from 2019 to a chip from 2019. They also compared their chip to the TPU1 when talking about "batch size of 1". But notice that its TPU1 which came out 4-5 years ago.

Again you have to look at every Tesla comparison in close context. The devil is in the detail.

I can't find anything about NVidia tensor core performance at batch size of one. All I can find even from recent stuff still recommends using larger batch sizes to improve GPU utilization (which implies that small batch sizes will perform poorly, while larger batch sizes allow more hiding of the stalls and latencies via pipelining of the rest of the batch). So while NVidia might be able to get some number of TOPS it is unlikely it can do so at a low latency.

 

[BioSehnsucht]

Very informative. It's having serious trouble getting out of the merge lane. Just one of a million problems to solve. They aren't anywhere near the corner cases yet.

That's not FSD, that's currently available NoA. That's why there's video - it's not from one of the FSD demo cars. Of course it's having trouble with indecision on when to try and get in, because that's how the cars currently are on public firmware. So it's not very informative unless you haven't been driving a Tesla equipped with NoA in such traffic before.

 

[scaesare]

I concede the linguistic point. I also concede that Henry Ford was working on self-driving when he developed the Model T, as obviously it was necessary to develop a vehicle in order to have it do self-driving.

Now you're just being (more) difficult (than usual).

The difference, of course, is that path planning and environment mapping clearly are required for FSD, but not needed for a general manual automobile.

So yes, let's thanks Henry for axles and brakes... clearly foundational, even if his solution was a bit light on the required TOPS.

 

[neroden]

Sure, but human drivers do that same thing routinely. At least it actually GOT in the lane. You have to look at this from a progress POV. Within the last 4-5 months, AP has gained a TON of new perceptive abilities. Being able to anticipate other cars merging in and doing on-off ramps and interchanges has been a huge leap for AP2 in the last few months.

And if they were advertising it as "best driver assist on the market, and nobody's going to be better, ever", that would be great.

The trouble is they're making wild sleep-in-your-car claims. Again. I mean, if they said "We are still working towards the goal of being able to sleep in your car", that would be fine. But acting like it's going to happen in 2022, nationwide, is delusional.

 

[jerry33]

I am worried about the appalling service and sales communications

Don't know about sales because it's been so long, but what is wrong with the service communications? They always answer promptly (and I'm in a State where I'm not allowed to phone the service centre directly, but have to go through corporate). If it's an SC issue, I get text messages showing the progress, if it's mobile service, they call before they arrive to make sure I'm there. I really don't see how it could be better. Because this has taken place for over six years, I don't believe I'm an anomaly.

 

[BioSehnsucht]

The fundamental aspect of SRAM is that its a memory chip that is ALOT faster and uses way less power than DRAM. Its used for low latency purposes. Everything about DRAM vs SRAM performance wise is night and day. We are talking ~10 GB/s for DRAM and ~200 GB/s for SRAM. and yes Nvidia also has local on-die memory

Nvidia has small unaddressable SRAM caches, not addressable SRAM memory that functions as a massive register bank. Just because they have SRAM doesn't make it comparable to Tesla's architecture.