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

[StealthP3D]

I guess anything about Elon gets you lots of clicks these days:




View attachment 652795
get

How come I never heard they were supposed to go 155 mph in a tunnel only 0.4 miles long?

I miss all the good stuff!

My Performance Model 3 with one person in it could hit perhaps about 125 mph before I had to slam on full antilock braking so I could slow it down to 20 mph to exit the tunnel. I don't think the passengers would like that. Oh, wait, with a car full of passengers I could only hit about 110 mph before I had to shut her down. It might make sense to hit perhaps 50 mph in the first 1/10th mile before coasting down over the last 3/10ths of a mile. And i suspect that's where this is going. It will feel like being softly "shot" through the tunnel. Any faster would waste excessive energy for very little benefit. Longer tunnels will have higher speeds.

 

[gabeincal]

Interesting to see the scale of the planned tunnel network in Vegas. Light blue is the currently built and released tunnel that the news outlets have been talking about...

 

[gabeincal]

I'm assuming the vehicle staging area was empty? Thanks for posting these. I should probably become a patron.

No. I screwed up the final video file and didn't do checks before uploading. I was rushing. Fixing now and the staging area is FULL of cars. All 3&Y as far as I can see....

 

[ReddyLeaf]

I was surprised at the ratio of passenger cars to pickups/SUV's there were in the construction worker parking lots. Even more surprised how few were Tesla (less than 1%). This shows just how early we are in the transition.

Apparently you haven’t been to TX.

 

[mrmage]

I guess anything about Elon gets you lots of clicks these days:




View attachment 652795
get

Alternate headline:
TBC begins training Teslas to FSD through the public transit Loop.

 

[gabeincal]

The full and now fixed version of today's flyover. Apologies for the mixup.

 

[PeterJA]

Alternate headline:
TBC begins training Teslas to FSD through the public transit Loop.

Alternate headline:
We're a scum-sucking tabloid that published this because we couldn't get photos of Elon in his underwear

 

[2Pearls]

The price of the M3P sure as hell went down since I bought mine (Dec. 2018).

EXACTLY, Tesla has lowered the price on many of their vehicles. We had a holder on our Model 3 the morning they opened the sales process over 2 years before and we wanted the Performance model. Not sure if we donated to the cause, but when we finally Took delivery of our Performance Model 3 with track software in September 2018, our cost was $76k. I think you can buy that exact same new car now for the high 50’s/low60’s. We still love it!

 

[Cherry Wine]

No. I screwed up the final video file and didn't do checks before uploading. I was rushing. Fixing now and the staging area is FULL of cars. All 3&Y as far as I can see....

I saw a white Model X driving through the staging area in the full video. (1:29) Can't tell if it's a refresh though.

 

[StealthP3D]

Apparently you haven’t been to TX.

Not since 2011 or so, I expected more trucks and SUV's and fewer cars, especially for construction workers.

 

[MC3OZ]

The full and now fixed version of today's flyover. Apologies for the mixup.

Around 4:40 the crane is moving something...

Are they swapping a mold, or servicing the machine?

It will be hard to guess.

 

[Chenkers]

Around 4:40 the crane is moving something...

Are they swapping a mold, or servicing the machine?

It will be hard to guess.

It appears to be a set of molds. They've been there for the last couple of fly-overs with welding work being done previously. I thought it may be repairing fire damage but the gigapress that caught fire can be seen to be operating in this video so it may just be maintenance.

 

[CorneliusXX]

It appears to be a set of molds. They've been there for the last couple of fly-overs with welding work being done previously. I thought it may be repairing fire damage but the gigapress that caught fire can be seen to be operating in this video so it may just be maintenance.

These machines seem like they will need quite a bit of maintenance - an understandable outcome for something dealing with high heat and pressure, and with quite a few moving parts. Tesla had the presses up and running before they built the bridge crane - They must have thought that the cost of building the crane will reduce the servicing/time costs of the presses overall.

 

[transpondster]

I think there is a possibility that the UK & Ireland might see Model Y shipments from Shanghai or US, depending on capacity. While Berlin is ramping they probably dont want the extra complexity of switching to RHD models (such as those needed for UK & Ireland), whereas Shanghai & Fremont will be much further along in production ramp by the time the first Y rolls out of Berlin. Maybe would make sense for Shanghai to handle RHD models long term given the location.

Happy for someone more knowledgeable on RHD supply to correct me on this though - and possibly handling RHD production is much more trivial than I assume with minimal downtime required to switch?

not gonna happen 10% import tax kills logic of importing if Berlin is running. Norway/Swiss could be option as they don't have import taxes on cars (EVs?)

 

[lklundin]

I was shocked to come home and read that the ER is on a Monday (never happened in the 8 years I've been holding TSLA). This will make options on the 23rd very interesting. I was going to roll my 800 strike covered calls from the 16th to the 23rd on Monday to pocket more time value, because I was assuming the ER was going to be on the 28th. Now I will just sit tight and let them expire worthless on the 16th and wait until after the ER to do anything else. FOMO might cause a nice spike on the 23rd....

I am unsure if anyone should try this, but here is one way this could be interesting for a very aggressive (or very foolish) TSLA bull, who is already leveraged to the point that their broker will not let them buy more shares on margin:
Create a combo-options trade with expiry Friday before ER. One leg should be deleveraging (i.e. improve margin), this could be selling two or a few way OTM covered calls (which would be expected to not be exercised). The other leg needs to require less margin and could be the selling of a single put - which if exercised will be done with the shares delivered on additional margin.

The advantage is that the very leveraged position is held only during one trading day (assuming the ER is not somehow postponed) - after which the investor hopes to deleverage by taking profits.

Overdoing this or doing this without a good understanding of exactly how the broker calculates the margin requirements can in the worst case of a deep dive on the opening on the ER-day literally destroy the _entire_ portfolio. This can in principle also happen before expiry, but at least during the trading day the investor can deleverage by selling shares - at a poor price, otherwise there would be no need to deleverage. Also, the investor should hold no other short OTM puts that are near expiry (e.g. Friday after earnings) - otherwise the consequences of a dive on ER-day could bring these puts ITM, which can easily make the account margin deficit.

Basically, the idea is that at some leverage the broker will set the buying power to zero - but there is typically still some buffer of margin to reduce the risk that price fluctuations (i.e. drops) will make the account margin deficit (i.e. with the long positions unable to cover for the short positions, e.g. negative cash balance and the sold puts). So for a broker where an actual margin call will only occur if this buffer actually goes to (or below) zero, there is some extra margin (and some extra risk) to exploit.

Let's say for example that the margin buffer is 100k$ - and that a single put with strike 680$ is sold as part of the combo order. Then 68k$ is needed in case of exercise, this comes out of the 100k$ buffer. The shares delivered will improve the margin on ER-day by their current value (initially 68k$) times a fraction set by the broker - if the fraction is e.g. 50% then the margin buffer has been reduced to 100k$ - 68k$ + 68k$ / 2 = 66k$. With this reduced margin, the investor can then again apply the rules from the their broker to see how much the SP can drop on ER-day before the long position drops to a point where the account becomes margin deficit, and determine if they are willing to try this - as a YOLO example.

Having now outlined this idea, I think it serves more as an advice on how _not_ to (ab)use the margin buffers set by brokers.

 

[AudubonB]

Or put another way, that final “O” of your YOLO could be very, very curtailed.

 

[petit_bateau]

A discussion about the benefits of radar should account for how modern AI based on neural networks makes decisions. Understanding this clarifies Elon's comment about not needing radar and thinking probabilistically.

The NN AIs in Teslas are basically probability calculators. The decision making in NNs is probability based, not binary or rule based logic (e.g. not a rule like check under the car in front and see if another car is slowing down). NNs have weights (coefficients) applied to each input (pixel) of each sensor (camera, radar, etc) which generate outputs that are fed into more layers that have weights that effectively use input to predict what the proper objects are or behavior should be.

Determining the coefficients and NN models is hard, but it's easy to see that Tesla can assess the contribution of radar to the system. Try one pass without the radar, try it again with the radar, then compare the accuracy of predictions.

Arguing that radar handles certain cases is binary thinking because it assumes that a particular "case" is recognizable (aka TRUE/FALSE) and would handle it a certain way. However, NNs assign probabilities to predictions (e.g. 82.35% chance of being this situation), so it's not black or white.

There's also the matter of ROI when adding sensors. Using completely made up numbers, let say radar ($100) improves NN predictions by 5%, but adding two more cameras ($20) improves NN predictions by 10%. Clearly, Tesla is better off installing two more cameras instead of one expensive radar.

If two cameras improve NN predictions 10%, is that worth $20? After all, maybe accidents would drop from 10,000 to 9,000. That sounds like a lot, but autonomy will require 5 or more orders of magnitude improvement. In other words, reducing errors from 10,000 -> 1,000, then 1,000 -> 100, etc until it's .1 error. With such large reductions, 10% would not make a difference, because the 100,000% improvement needed from better AI models and training would dwarf a 10% improvement through additional sensors.

My numbers are made up, but Tesla knows what the real numbers are. If Tesla drops radar, it's because they have done the math that shows it does not add a material improvement, relative to improvements in FSD software.

It is not quite as simple as that. One would need to run the first test with a NN fully (but optimally) trained on vision only, then a second test with a NN fully trained on the existing poor radar + vision (that will be a different optimal training). Ideally one would then run a third test with a NN trained with a good (AESA) radar + vision. Each test needs to explore a lot of cases, not quite an exhaustive search but quite likely nearly.

Funnily enough this is research I did (badly) as an engineering student 30-years ago. The underlying issue of how to "prove" that a NN is working is quite difficult, it was certainly too difficult for me back then. It is pretty much the same problem as trying a prove that any given human will not become mentally unstable in some particularly stressful edge case. Basically we don't know how to do either.

These are difficult problems and I am staggered at the fantastic progress that Tesla are making.

 

[22522]

Not since 2011 or so, I expected more trucks and SUV's and fewer cars, especially for construction workers.

You might be onto something.

 

[CLK350]

A discussion about the benefits of radar should account for how modern AI based on neural networks makes decisions. Understanding this clarifies Elon's comment about not needing radar and thinking probabilistically.

The NN AIs in Teslas are basically probability calculators. The decision making in NNs is probability based, not binary or rule based logic (e.g. not a rule like check under the car in front and see if another car is slowing down). NNs have weights (coefficients) applied to each input (pixel) of each sensor (camera, radar, etc) which generate outputs that are fed into more layers that have weights that effectively use input to predict what the proper objects are or behavior should be.

Determining the coefficients and NN models is hard, but it's easy to see that Tesla can assess the contribution of radar to the system. Try one pass without the radar, try it again with the radar, then compare the accuracy of predictions.

Arguing that radar handles certain cases is binary thinking because it assumes that a particular "case" is recognizable (aka TRUE/FALSE) and would handle it a certain way. However, NNs assign probabilities to predictions (e.g. 82.35% chance of being this situation), so it's not black or white.

There's also the matter of ROI when adding sensors. Using completely made up numbers, let say radar ($100) improves NN predictions by 5%, but adding two more cameras ($20) improves NN predictions by 10%. Clearly, Tesla is better off installing two more cameras instead of one expensive radar.

If two cameras improve NN predictions 10%, is that worth $20? After all, maybe accidents would drop from 10,000 to 9,000. That sounds like a lot, but autonomy will require 5 or more orders of magnitude improvement. In other words, reducing errors from 10,000 -> 1,000, then 1,000 -> 100, etc until it's .1 error. With such large reductions, 10% would not make a difference, because the 100,000% improvement needed from better AI models and training would dwarf a 10% improvement through additional sensors.

My numbers are made up, but Tesla knows what the real numbers are. If Tesla drops radar, it's because they have done the math that shows it does not add a material improvement, relative to improvements in FSD software.

Consider ALSO that the NN would need to learn mostly in corner cases (rare!) how radar would improve over non-radar, these cases I can sort of see would be VERY rare.
I suggest maybe the mods ought to stop further posts trying to out think Tesla's decision - some is fun, but really .. would we know better than Tesla's A team of AI /NN researchers ???

 

[22522]

I was surprised at the ratio of passenger cars to pickups/SUV's there were in the construction worker parking lots. Even more surprised how few were Tesla (less than 1%). This shows just how early we are in the transition.

I was surprised that a crane could move 6 pieces of steel from the ground to the correct place in the structure in a single lift like a string of beads. See 3rd from the left.