2017 Investor Roundtable:General Discussion

[myusername]

Tesla has opened the floodgates of Autopilot data gathering

'
At this point, Tesla has likely well over 50,000 vehicles on the roads gathering this data and depending on the production ramp up of the Model 3, it could potentially double that by the end of the year.
'
Every one sold to an actual customer at an actual profit

But of course
GM has a couple hundred;
and Waymo's got 500 in Phoenix (with hats)
Sold to nobody, at a loss on every one...
In Tesla's AP rear view cameras,
Objects Are Definitely Smaller Than They Appear

honestly... i'm confused. Isn't Autopilot 2.0 still not as fully functional as Autopilot 1.0?... yet the "miles of data accrued" has increased substantially?

doesn't this completely destroy the entire theory?... this was either a marketing gimmick designed to make Tesla look moat-like advanced or it was something of substance.

but... they regressed. so... it appears to be a gimmick.

 

[myusername]

Your ability to predict and rationalize SP movements was repeatedly noted on this Forum, perhaps at least partially due to the confident oracle-like tone you use in making these predictions.

I am not belittling anybody. Your credibility is based on what you post, not my commentary.

nope... you're replying with a single instance of inaccuracy to help your point... that TE is going to ramp... and you continue to do it to avoid answering a question that makes you and Tesla wrong.

let's try this again:

Elon told us $2b in 2017... when do you think TE will bring in $2b?

 

[ValueAnalyst]

@FredLambert - The following sentence in this article needs correction:

"All but 20 of the 2,964 cars registered in March were not Tesla’s"
to either:
"All but 20 of the 2,964 cars registered in March were Tesla's"
or
"Only 20 of the 2,964 cars registered in March were not Tesla's"

Thanks to @MitchJi 's recommendation, I've been rigorously studying for TOEFL/SAT/GMAT.

---------

Having said that, my general view is changes in tax incentives will cause big swings in month-to-month sales, but their impact will decline as Tesla decreases the price of its cars and/or increases their value vs. ICE options in the coming years.

A back-of-the envelope calculation, combined with what we know about li-ion limits, shows Tesla can further reduce the cost of its battery packs by up to 50% as Gigafactory ramps up, which for a 70 kWh battery, could mean as much as $5-6k price reduction, which comes close to fully offsetting the impact from Federal Tax Credit phase-out in 2018/19.

Further, alien dreadnought improvements to exit velocity and volumetric efficiency of the Gigafactories will also contribute to price reductions of Tesla's products in the coming years.

 

[ValueAnalyst]

In the TE related news, according to GTM Energy Storage Monitor, in Q1 there were 234MWh worth of energy storage projects deployed in US. The total combined power rating of all installations was 71MW, i.e. aggregate MWh/MW ratio was 3.3.

According to the Q1 2017 letter Tesla installed 60MWh in Q1, likely virtually all of it in US given that 56MWh Kauai installation was completed in Q1. Based on this Tesla share of US energy storage market in Q1 was about 25%.

As Gigafactory 1 ramps up, and additional Gigafactories start coming online in 2H20, I expect Tesla's energy storage market share to increase towards 50% in the coming years.

 

[vgrinshpun]

nope... you're replying with a single instance of inaccuracy to help your point... that TE is going to ramp... and you continue to do it to avoid answering a question that makes you and Tesla wrong.

let's try this again:

Elon told us $2b in 2017... when do you think TE will bring in $2b?

Stop this nonsense, unlike your proclamations about SP, Elon's forward looking statements are not oversold.

As I noted, I expect Tesla to hold 25-30% of the BES market. TE will bring $2B when BES market grows to $6.7 - $8B.

Let me know when do you think we will arrive at "next stop 270"...

 

[techmaven]

honestly... i'm confused. Isn't Autopilot 2.0 still not as fully functional as Autopilot 1.0?... yet the "miles of data accrued" has increased substantially?

doesn't this completely destroy the entire theory?... this was either a marketing gimmick designed to make Tesla look moat-like advanced or it was something of substance.

but... they regressed. so... it appears to be a gimmick.

Not clear if you actually understand and want to just spread FUD, or you really don't understand. Tesla Autopilot vehicles collect data, both AP1 and AP2. The data they collect and send back are not the same between hardware generations and even within a hardware generation, the data collected and sent back has changed over time. Even when the vehicle is not in autopilot, there is valuable information collected and sent back. There are several different kinds of information sent that informs various parts of the autopilot program. One of the most important is validation in both active and shadow mode. But aside from that, the information about the roads themselves is also important and helps develop the high resolution road map tiles.

Getting AP2 hardware into the field to collect data was very important, as validation had to occur. The difference in hardware and sensors was too great to allow just AP1 data collection for use in AP2 development. Likely the kind of information collected with AP1 is still used for the high resolution map tiles and that data collection is ongoing. It is likely that other level 4 ADAS development programs will also have to eventually adopt Tesla's approach, as it is the only rational way to get enough experience. Therefore, what is exposed to the driver as a level 2 ADAS system is not the only code that is running and not the only ADA effort that it is informing.

One of the markers I look at is the data uplink in the car. It doesn't help if you install a Mobileye EyeQ3 chip in a shipping ADAS system and have no data uplink capability. Examine when Mobileye's REM will actually collect data from their partners. For example, BMW and VW's vehicles will only start to provide data for Mobileye's REM in 2018. They will begin to incorporate REM into some new vehicles in 2018 which means typically late 2018 given new model introduction timings. GM is working with Mobileye REM with data collection through Onstar. But note that the current REM collection is basically AP1 style data collection, not AP2, which makes sense since it's EyeQ3 which is what was one of the main building blocks of AP1. Not clear to me that any shipping GM cars are actually contributing data to Mobileye's REM yet... there was discussion of it and certainly in test vehicles last year. So likely Mobileye will get AP1 style data sometime this year with some vehicles, with more coming in 2018. But Tesla is collecting much more data in AP2 since the end of 2016. The Mobileye EyeQ4 chip is expected to ship in 2018, and that likely matches up with the real REM data collection announced with VW, BMW, Nissan, and others. Which means they are basically 2 years later than Tesla at best. Oh, and the EyeQ4 is a 2.5 Tflops chip, which is 5x slower than what is in the AP2 hardware. The EyeQ5, due in 2020, is likely comparable to AP2 hardware.

 

[scaesare]

Consider what happens if the septic leach field is under the lawn and the car drives across it and damages it instead of driving where it's supposed to. (Real issue, BTW, I have visited places where they had to warn drivers about this)

I have friends with property with the same scenario.

However, if the bar for human drivers is "Undetectable hazard, human intervention/warning required", why is it any different for an autonomous system?

The thing is I can come up with stuff like this *all day long*. There's *so many of them*.

So can I. However most examples like this would also trip up a human.

 

[kenliles]

honestly... i'm confused. Isn't Autopilot 2.0 still not as fully functional as Autopilot 1.0?... yet the "miles of data accrued" has increased substantially?

the collected 1.0 data is how they achieved current parity in only a few months- despite starting from scratch with an entirely new sensor suite and input profile.

doesn't this completely destroy the entire theory?...

what theory?

this was either a marketing gimmick designed to make Tesla look moat-like advanced or it was something of substance.
but... they regressed. so... it appears to be a gimmick.

In fact, the substance between AP1 and 2 was the issue of collected data which M-Eye was preventing.
Is your question substantive? It appears to be a gimmick
[what's your password?]

oops - I see maven beat me to the post punch

 

[racer26]

This news from HK:

In March, close to 3000 Teslas were registered (and it was believed that as many as 500 of those were registered by Tesla themselves).

In April, zero new Teslas were registered.

This makes sense. Tesla deliberately registered and paid the FRT (First Registration Tax) on those cars they registered themselves so that they could continue selling new cars (which have been registered and left sitting on Tesla's lot) without exposing their customers to the FRT hike nearly doubling the price of the car.

 

[Reciprocity]

I have friends with property with the same scenario.

However, if the bar for human drivers is "Undetectable hazard, human intervention/warning required", why is it any different for an autonomous system?


So can I. However most examples like this would also trip up a human.

For Level 4 its possible that Tesla will only allow autonomous driving when combined with Nav. This would allow the car to know what issues that require human intervention are on the path and can in some cases can route around them or take the fastest route and give plenty of warning of when a human would need to take over. Obviously this does not work for Level 5 because level 5 would not require a driver. I guess it could still work if the without a driver routed itself around any places that require a human.

I guess I dont know what a leach field is or why someone would be driving through it. Are these like flood plains where roads are often flooded? Or are these things much more transitory, meaning they can show anywhere out of the blue? Why do cars drive over someones lawn? I am very confused as to why this would be corner case at all, but either way, the above still covers it.

 

[scaesare]

I guess I dont know what a leach field is or why someone would be driving through it. Are these like flood plains where roads are often flooded? Or are these things much more transitory, meaning they can show anywhere out of the blue? Why do cars drive over someones lawn? I am very confused as to why this would be corner case at all, but either way, the above still covers it.

They are tanks buried a few feet underground to provide sewage treatment for homes not having city/county sewage service.

They are typically buried under a lawn, and aren't designed to hold the weight of a vehicle.

In the case of my friend, if he has a party and needs auxiliary parking, he'll have folks park on the grass, but you have to avoid that area.

 

[myusername]

the collected 1.0 data is how they achieved current parity in only a few months- despite starting from scratch with an entirely new sensor suite and input profile.

what theory?

In fact, the substance between AP1 and 2 was the issue of collected data which M-Eye was preventing.
Is your question substantive? It appears to be a gimmick
[what's your password?]

oops - I see maven beat me to the post punch

"parity" was not achieved by the end of December... that was 3 months and what was originally promised. why are memories around here so colluded when it comes to things like this.

Elon put out a statement at the end of Q1 saying AP2.0 will be updated and should help get back to AP1.0 capability.

Is AP2.0 equal to or greater than AP1.0 now?

 

[mmd]

As I've been saying, the supernova effect in Hong King is happening. Without incentives, electric cars don't stand a chance.

Hong Kong’s electric car market comes to emergency stop

“There was no first registration of an electric private car in April 2017,” a department spokesman confirms.

 

[kenliles]

why are memories around here so colluded when it comes to things like this.

I don't recall

Elon put out a statement at the end of Q1 saying AP2.0 will be updated and should help get back to AP1.0 capability.

Is AP2.0 equal to or greater than AP1.0 now?

very close from reports. good thing they have all the data to 'make it so'

 

[racer26]

As I've been saying, the supernova effect in Hong King is happening. Without incentives, electric cars don't stand a chance.

Hong Kong’s electric car market comes to emergency stop

You misunderstand.

In HK, prior to the expiring of that incentive, a Model S cost its owner approximately the same as a high-end Camry. The FRT is ENORMOUS, and Teslas were exempt from it.

 

[Reciprocity]

They are tanks buried a few feet underground to provide sewage treatment for homes not having city/county sewage service.

They are typically buried under a lawn, and aren't designed to hold the weight of a vehicle.

In the case of my friend, if he has a party and needs auxiliary parking, he'll have folks park on the grass, but you have to avoid that area.

Crap, never thought of that.. I have Septic tanks and drive on my lawn sometimes. I need to find out where those are.. haha. I still dont understand how that would impact driverless cars. The car wouldn't know that the lawn is an appropriate place to park as a matter of fact, it would know that it is not a place to drive. However, I could see a system where people could add geofence information to a shared repository for private properties and businesses. There are already tons of maps data around property boundaries, it would be pretty simple to allow people to define go and no go zones on their properties, which could be used mostly for their own vehicles, but could be shared to make sure no autonomous cars drive in the no-go zones. By default, the entire property would be a no-go zone and it would be up to the owner, or machine learning, to turn no go-zones into go zones. For example, if you drive your Tesla up your driveway every day for month, then it learns that path is a go zone and can define safe speed limits based on the human drivers actions. No other part of the property would be a go zone until it was either driven on some number of times by a person or specifically designated. Any driverless car company could subscribe to this data and use it to know where a car should not park.

This is why the cars data is so valuable. Almost any corner cases can be solved by showing the system thousands of examples of how the human successfully navigated the particular issue. This is how machines learn, they dont learn as quickly as we do, but you can give them very high volumes of data and they can start to learn from that data. You can then simulate how the system would deal with a specific situation. You can use the same data to simulate thousands of similar variations and determine if the system would be successful more then a human. I have no idea how long this will take, but I believe that Tesla is going to shoot for 10x safer then a human and will not wait until its perfect. They should be able to show regulators hundreds of examples of where people crashed and the system would have taken a different action. This is why I think regulation will move quickly and I think regulators will be falling over themselves to be the first to allow autonomous driving. For one because it will save lives and two for the bragging rights of being technologically progressive.

 

[Krugerrand]

nope... you're replying with a single instance of inaccuracy to help your point...

And the extraordinary level of irony continues.

 

[neroden]

I have friends with property with the same scenario.

However, if the bar for human drivers is "Undetectable hazard, human intervention/warning required", why is it any different for an autonomous system?

It's not level 5 autonomous then, is it? Also, how do you tell the system about the undetectable hazard? If you're intervening to tell the system about hazards it can't deal with without driver intervention, that makes the system level 3, by the definition used for the levels.

I'm pretty optimistic about level 3. I'm pretty pessimistic about level 5. See why?

"Level 5" will only be available under geofenced conditions on specific roads, probably excluding most driveways even within those areas, and only under certain weather conditions. For a very long time. Now, don't get me wrong, that could still be a big deal, because that could account for a large majority of all driving. But the "car without a steering wheel which takes you door to door anywhere" is fantasy at this point.

 

[winfield100]

TE took about 25% of the BES US deployments in Q1, this is in line with my expectations. The market for BES is slated for a very explosive growth. I expect Tesla to hold or grow their market share, which will translate into explosive growth of their TE business.

Do update us on "next stop 270" when you get a chance...

when are you predicting this 3:1 stock split?

 

[Krugerrand]

when are you predicting this 3:1 stock split?

The last I recall, Elon is against stock splitting.