2017 Investor Roundtable:General Discussion

[brian45011]

I have no idea what you just said so I am just going to nod and agree quietly.

When Tesla does a direct, in-house lease (as opposed to banking affiliate leases) the tax credit on those vehicles goes unused, but still counts against the 200,000 because leased cars must be titled by Tesla (as opposed to service loaners and show room models).

On direct leases, Tesla adds back the $7,500 to the estimated FMV at the end of the lease term, ie the "residual." This reduces the lessee's monthly payment but induces lessees to return cars at the end of the lease terms since they would have to pay $7,500 over the used car market to retain the vehicle.

 

[Oil4AsphaultOnly]

The people programming have never designed a Bridgeport milling machine. Or taken classes about how you do it.

With a Milling machine there are 2 cylindrical rails that the table travels on as you advance the piece through the cutter. There is no way to hold a tolerance between 2 rails for the entire travel of the table. The only way to get a precision path is to use a circular hole on one of the rails to follow that one closely, and a slot riding the other rail. The slot keeps the table level, but does not require the distance between the two rails be precise.

Auto pilot should work the same way. Use the dashed center lines to construct a smooth path line. Register the vehicle a fixed distance off that line. No ping pong.

Adjust the target distance off the center line only with "out of control limit" changes in lane width.

The software does not demonstrate as much cross training and breadth of experience as I would expect.

That's not how machine learning is done. That's determinate programming, which is truly error-prone since it has to account for all the exception cases.

 

[22522]

That's not how machine learning is done. That's determinate programming, which is truly error-prone since it has to account for all the exception cases.

Machine learning does not required a desired path?

Now I know why Elon does not trust it!

 

[ValueAnalyst]

The people programming have never designed a Bridgeport milling machine. Or taken classes about how you do it.

With a Milling machine there are 2 cylindrical rails that the table travels on as you advance the piece through the cutter. There is no way to hold a tolerance between 2 rails for the entire travel of the table. The only way to get a precision path is to use a circular hole on one of the rails to follow that one closely, and a slot riding the other rail. The slot keeps the table level, but does not require the distance between the two rails be precise.

Auto pilot should work the same way. Use the dashed center lines to construct a smooth path line. Register the vehicle a fixed distance off that line. No ping pong.

Adjust the target distance off the center line only with "out of control limit" changes in lane width.

The software does not demonstrate as much cross training and breadth of experience as I would expect.

The new videos will likely change your opinion.

 

[Intl Professor]

We will make great pets. We'll make great pets. Ohh ahhh.. we will make great pets.

Maybe watching us will be their TV. There'll always be jobs for entertainers; gladiators for example. But will they always be so able as we to differentiate TV from reality. Hmmmmmm?

 

[winfield100]

My background molecular biochemistry and I was extremely skeptical of sequencing human genome in 1980 now only extremely embarrassed to have thought that back then

@Chickenlittle, you are in NoVa?, i worked @ NIH, bldg 10, 49, 15K, few others. wife hand sequenced 46,000 bp both ways, from 91-94, XRay film, hand count, now do 3-4 yrs work in lazy afternoon, HGP on 6th flr bldg 49. big hoohaa! lotsa brainz. 40 -100 gigbit ring, fun times, totally surprised we were at speed of sequencing. punctuated unevenly distributed singularities, eh? check out fMRI and neural reads, no writes ...yet

 

[dc_h]

what are you guys saying? No one wants a Bolt because BEV's are the wave of the future?

One of my neighbors has reserved a TM3 which they tell him he'll get in February 2018. I asked him if he'd test driven a Bolt and he said no - he had tried the BMW i3 and didn't like it. He thought the differentiator was the Tesla charging network. So, that's a point for the longs, the charging network is a huge differentiator.

Superchargers: No one else has global highway charging
Design: BEV cars that look like cool cars (not a Panda on wheels, or a Sonic)
Sales: ICE dealers hate BEV
Service: ICE dealers hate no service fees
Branding: One car has the cache of being the technology, design and engineering leader
Scale: Only one mfr intends to build 100,000's of thousands of BEVs
Scale: Only one mfr is building the largest battery plant and gaining a 30-50% cost advantage on batteries
Purpose: Only one mfr is currently building only BEVs, all other western mfrs are still building compliance cars. Plans for mainstream vehicles are still beyond 2020.
Tech: Over the air updates allow for continuous improvement
Tech: Metadata capture is allowing Tesla to gather the most robust street view data in the world
AI: Elon's investment in OpenAI provides a revolving door to new tech directly to Tesla
SpaceX: 3D printing and new material sciences are helping Tesla innovate faster


Other than than the above details, Tesla is neck and neck with the rest of the industry. I think the 3 key issues are Superchargers, design and costs. Creating a global driving infrastructure for a new power source (for cars) was the chicken and egg moment. Tesla solved both at the same time. Design has created the brand image of an awesome and beautiful car. Perhaps its just getting from point a to point b for many people, but not for the people who are responsible for all of the profits in the auto industry. GM doesn't pay the bills with the Sonic, nor Ford with the Fusion. Driving down costs is preparing for the end of subsidies and the most exciting time for long term Tesla fans. When margins on BEV's surpass ICE's, the middle of the story begins. At that point all the other manufacturers have to acknowledge their transmission and engine plants and technology have no future value and are obsolete. At this point the footrace reaches another level and it becomes about the survival of the closer followers and the demise of the deniers.
The second half of 2018 will start getting interesting. Tesla's continued investments in new plants may continue to feed the FUD that they are losing money on every car sold. I think they are finding money in every sale and using it as quickly as possible to fund the most rapid expansion possible. It may not be as fast as some (VA?), but it will be much sooner than others believe.

Best of luck with your personal investing. I don't expect to make any adjustments in the near future. I don't see another short squeeze funding the next Gigafactory, or having any lasting downward pressure on the stock. The next 6 months is about scaling up the Model 3 and maintaining the brand value. 2018 will be about margins on S, 3 & X and the emerging buzz about the Semi and timing of the Model Y. If they hit targets on volume and margin for the Model 3 the stock should rise through the year. If they nail the Semi reveal and start setting release dates for the Model Y in early 2018, next year should be a great year for longs.

I hope you are not over invested in your short theory. I don't like over leveraged longs on the Forum any more than paid bloggers using the site for fodder. I do appreciate some back and forth on the site, as it can encourage thinking through the risks and opportunities in the near and longer term.

 

[JRP3]

My guess, and this is completely a guess based on what he said at ted, they may incorporate a demo of dynamic routing during the trip IF they are comfortable enough with the system.

Even so the route will be restricted to major roads with supercharger access, so they can "train" it for any possible routing ahead of time.

 

[MartinAustin]

‘Fossil fuels are dead’ says rail baron who hauls 800,000 carloads of coal a year

 

[winfield100]

since folks have digressed and are talking about "friendly robots" read Asimov's "liar, liar", and run screaming from Jack Williamson's "with folded hands", and "the humanoids" (humans don't win,) or even Chas Stross "Singularity sky", very bad for humans
back quiet and lurking, maybe

 

[JRP3]

On direct leases, Tesla adds back the $7,500 to the estimated FMV at the end of the lease term, ie the "residual." This reduces the lessee's monthly payment but induces lessees to return cars at the end of the lease terms since they would have to pay $7,500 over the used car market to retain the vehicle.

So does that mean Tesla eats the $7,500 when you turn the car in, since they don't get the tax credit?

 

[MP3Mike]

So does that mean Tesla eats the $7,500 when you turn the car in, since they don't get the tax credit?

If they directly leased it, yes. That is why they are not the primary leasing company. (US Bank, etc. are the primary leasing agents and get to take the $7,500 credit.)

 

[Reciprocity]

Thanks for the response. Perhaps they really do have a lot more going on beyond the AP2 system. Any thoughts on why something as simple as lane keeping on curving roads is too challenging at this point for AP2? My AP1 does fine on curving roads.

I have never had problems on curvy roads, so I can't really speak to it. I think it has become smoother and maybe people didn't feel as comfortable until recently. It's not easy to replace a product like mobileye, which was recently acquired by Intel for $15B. This shows you the value of what Tesla has recreated in just 9 months. What some call disappointing, I call extremely impressive. To put this into context, Tesla has two good driver assist systems and no one else has one that's nearly as good.

 

[kbM3]

I don't think parity is of very much importance to Tesla right now, in the sense that they are really more interested in much bigger things. Why worry about that if you are going to drop the mother of all updates soon? Either that or they really are struggling, but if that were the case I don't think you see Musk reiterating the coast to coast trip, and I think you see the head of auto pilot getting fired rather than starting his own thing.

I think maybe there is a misunderstanding or something. The 6B isn't delusional, it's the gameplan. My understanding is that 6B has been adding up for a while while ap is being used and even when it is not, it started a while ago, and is talking about level 4 and then I'm not sure how soon after level 5 will arrive. My understanding is they already pretty much have the ability for level 4, it's just the accumulation that they need to prove safety beyond a doubt before it can be released to the cars. Tesla has now 1.3 billion miles of Autopilot data going into its new self-driving program

As I understand it, the 6B miles is to prove that the system is safer than a human. You can read in the auto pilot threads about many times where the driver is having to intervene.

If they are using the current mileage, how are they counting all of the times when a human driver is taking control? Are those counting as fatal accidents? Or are those counting as minor accidents? I don't think that the mileage they have so far will go towards demonstrating that the car auto pilot system is safer.

 

[Waiting4M3]

The people programming have never designed a Bridgeport milling machine. Or taken classes about how you do it.

With a Milling machine there are 2 cylindrical rails that the table travels on as you advance the piece through the cutter. There is no way to hold a tolerance between 2 rails for the entire travel of the table. The only way to get a precision path is to use a circular hole on one of the rails to follow that one closely, and a slot riding the other rail. The slot keeps the table level, but does not require the distance between the two rails be precise.

Auto pilot should work the same way. Use the dashed center lines to construct a smooth path line. Register the vehicle a fixed distance off that line. No ping pong.

Adjust the target distance off the center line only with "out of control limit" changes in lane width.

The software does not demonstrate as much cross training and breadth of experience as I would expect.

Another degree of difficulty is that a road is not always flat. The lane lines show up in a 2D camera image, and the AI needs to interpret the 2D image into a 3D environment, if the road is tilting up or down, or in the case of cresting a hilltop, the road is tilting up and then down, so the perspective of the camera is constantly changing. IMO it's way more complex than to follow a line on a flat table.

 

[brian45011]

Elon Musk says he has approval to build tunnel that would cut NYC to DC commute to 29 minutes

That is bizarre. There are dozens, if not hundreds, of jurisdictions whose written approval (and compliance with their specifications) would be required if the tube(s) were solely under publicly controlled right-of-ways. Passing under private property would be several orders of magnitude more complex. In the US, conventional (somewhat simplistic) real property law is that a landowner's sub-surface rights extend from the surface to the "core of the earth." http://scholarlycommons.pacific.edu/cgi/viewcontent.cgi?article=1141&context=facultyarticles

He's channeling someone else's gratuitous tweets who types before engaging gray matter. Focus on the M3 ramp is essential!

 

[winfield100]

In comments to my articles, I've noticed a surge in FUD on Solar Roofs.

FUD on Model 3 seems to have abated.

a very simple answer is to look at the $4.66 rise in stock price today, the rise of the last 4-5 days, and do a scan of SA and whose short and how much they are panicking hurting

 

[Waiting4M3]

That is bizarre. There are dozens, if not hundreds, of jurisdictions whose written approval (and compliance with their specifications) would be required if the tube(s) were solely under publicly controlled right-of-ways. Passing under private property would be several orders of magnitude more complex. In the US, conventional (somewhat simplistic) real property law is that a landowner's sub-surface rights extend from the surface to the "core of the earth." http://scholarlycommons.pacific.edu/cgi/viewcontent.cgi?article=1141&context=facultyarticles

He's channeling someone else's gratuitous posts who types before engaging gray matter. Focus on the M3 ramp is essential!

If he drills under HW95, wouldn't he only need the DoT's approval?

Edit, then plus the last mile going to the city center, again it will be likely under public road (elevators go straight up to the road), so just need the city to approve.

 

[tander]

As I understand it, the 6B miles is to prove that the system is safer than a human. You can read in the auto pilot threads about many times where the driver is having to intervene.

If they are using the current mileage, how are they counting all of the times when a human driver is taking control? Are those counting as fatal accidents? Or are those counting as minor accidents? I don't think that the mileage they have so far will go towards demonstrating that the car auto pilot system is safer.

I think it's referring to miles collected when autopilot is in use and even when it is turned off. So it's learning even when not in use, kind of like a 15-year-old sitting in the passenger seat might be watching what the driver does in various situations and even thinking of ways they could do it better. We're talking about auto pilot 3 here or whatever they call it, So when a human driver has to take control that's with autopilot 1 or 2, 3 is in the passenger seat sponging up all the info, it still hasn't been in the drivers seat yet. So if they can say auto pilot 3 ( which is been running in the background all the time doing a simulation) would have prevented this accident etc. if it was driving, that's how to demonstrate it's safer. So it's like having the old program actually being used in the cars right now, but the new one is in there logging what it would do if it were in control. So eventually the new one gets to a point where it's consistently making better decisions than the old one and better than the human too after 6B miles ( before imo, 6B is to prove beyond doubt). That's the plan as I understand it .

The bigger point too, is that once you get to significantly safer than a human, the question has to be raised is should it be mandated to be put in every car where it could save thousands of lives a year? I bet Tesla might even give the safety part of it away for free like Volvo gave away 3 point seat belts.

 

[mongo]

When Tesla does a direct, in-house lease (as opposed to banking affiliate leases) the tax credit on those vehicles goes unused, but still counts against the 200,000 because leased cars must be titled by Tesla (as opposed to service loaners and show room models).

On direct leases, Tesla adds back the $7,500 to the estimated FMV at the end of the lease term, ie the "residual." This reduces the lessee's monthly payment but induces lessees to return cars at the end of the lease terms since they would have to pay $7,500 over the used car market to retain the vehicle.

I can understand adding the 7500 to the residual to calculate lease payments. Due to financing and insurance, they would not want to take the 7,500 off the sales price.
Then, at the end of lease, the bank can't say the residual is different that stated at the beginning of the contract. So you don't buy out the lease, then it reverts to the bank and is sold at the market value, possibly 15k less than original residual (7,500 from the adjust up in value and 7,500 from the credit the bank received).
If this is correct, leasing with the intent to buy doesn't make fiscal sense at this point. Yah?