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

[neroden]

Indeed.

I was trying to figure out where in this thread to interject this, and your post is a great one.

Adding to it, everyone should read about how AlphaZero became the most powerful chess playing entity on the planet in just 4 hours. What's incredible about this is that the researchers didn't do this by teaching their AI how to play chess. They did it by developing a general-purpose AI advanced enough, and then simply teaching it the rules of chess, and just letting it play.

No loading initial strategies, no showing it 1000's of games to model itself after. They let it learn by playing (itself), and 4 hours later it crushed every comer, including the previous best chess programs/AIs that had been painstakingly tuned for countless hours. It took 8 hours to master Go (considered significantly harder than chess) and obliterate everyone and everything there as well.

Interestingly, in both cases the AI developed some winning strategies that were counter-intuitive to it's creators and experts, but quite effective.

So that's not to say that we can load AlphaCarZero in to Tesla's FSD computer and let it loose and call it a day. But it does underscore @eloder 's point that if " advancement of the neural net could potentially be reaching exponential increases at this point" and if so "we're in for something pretty earth-shattering" when that happens.

It may not be on the 22nd, but at some point I suspect we are going to turn a corner and the issue of "good enough" will rapidly vanish as a question. It won't be perfect, but as Elon says, a 10X reduction in auto fatalities will be hard to argue against.

I will repeat that all of this Go and Chess history is totally irrelevant to "self driving" because *we don't actually know the rules for driving* -- or at least Tesla staff don't! As soon as a suitable set of rules is devised, the problem is essentially solved, yes. But devising that set of rules is very, very hard. I don't think we've ever tried running a neural net on a game with unspecified rules which the computer isn't told and which the humans argue about.

I mean, I think it's embarassing that they're still working on computer vision and object classification. They are deluding themselves if they think solving that solves self-driving; that's just a prerequisite for even being able to *start* work on self-driving.

Tesla does appear to be far ahead of everyone else, and if they manage to simply implement automatic emergency braking that ends all "run over pedestrian" crashes, they'll have an insurance goldmine on their hands.

But "sleep in your car" self-driving is a bridge too far for the near future; from what I can tell they still haven't really started on the corner cases.

 

[The Blue Owl]

Ask yourself why we don't have driverless trains everywhere. The tech was developed in the 1970s and perfected in the 1990s. It's at least 10x safer than human drivers and substantially more efficient economically.

I await your answer.

This is a bit of a drive-by comment (no dog in this hunt), but one difference is that there are not 40k deaths every year caused by train collisions. There is a lot more to gain in the automotive world, and it goes well beyond safety, too. If we can have all trains drive themselves, we save on all those train engineers' training, salaries, and pensions. On the other hand, if we can have all cars drive themselves, well... it's a whole new game.

(To make it clear: I agree with your assessment of the huge difficulty of reaching true FSD. This is just about the differences with trains, which I think are relevant.)

 

[mongo]

I will repeat that all of this Go and Chess history is totally irrelevant to "self driving" because *we don't actually know the rules for driving* -- or at least Tesla staff don't! As soon as a suitable set of rules is devised, the problem is essentially solved, yes. But devising that set of rules is very, very hard. I don't think we've ever tried running a neural net on a game with unspecified rules which the computer isn't told and which the humans argue about.

I mean, I think it's embarassing that they're still working on computer vision and object classification. They are deluding themselves if they think solving that solves self-driving; that's just a prerequisite for even being able to *start* work on self-driving.

Tesla does appear to be far ahead of everyone else, and if they manage to simply implement automatic emergency braking that ends all "run over pedestrian" crashes, they'll have an insurance goldmine on their hands.

But "sleep in your car" self-driving is a bridge too far for the near future; from what I can tell they still haven't really started on the corner cases.

 

[Singuy]

This might be related to the lane change settings. Have you tried setting it to Mad Max? The system appears to be somewhat conservative in the other modes under the conditions you described. My LR model 3 has ~24k miles (mostly highway) and at least 70% of those miles have been driven with NoA with little to no issues save for a few interchanges that are squirrely due to heavy construction. Given the formal description of level 3 autonomy, I don’t understand how Tesla’s current system can be construed to be anything but.

It is not on mad Max, however I feel the system is already suggesting to many unnecessary lane changes already, especially near an exit. It would suggest I overtake the car in front of me when the exit is less than a mile away with my current setting. The system can't seem to make up its mind and there were times I cancelled EAP near exit thinking "wtf are you doing car?". When there's light traffuc the system work wonders and really feels like level 3.

 

[mongo]

Ask yourself why we don't have driverless trains everywhere. The tech was developed in the 1970s and perfected in the 1990s. It's at least 10x safer than human drivers and substantially more efficient economically.

I await your answer.

They had tech to detect people/ objects on the tracks or at ungated crossings in the 70s?

Do tell.

One to two engineers per train seems low cost compared to the technological expense of their full functional replacement. Even with humans in the loop, they are adding more automation and controls due to people being the root cause of many accidents (over speed/ misread signals)

 

[Navin]

The idea of a Tesla spinoff for ride-sharing software could be interesting for sure. Maybe he will just call it “x”. He owns the domain www.x.com

 

[KarenRei]

Interesting. Let's compare model 3 sales to Q3 2018. There are some similarities between the quarters.

Okay, to start off I object to the very premise, since it entirely negates all of Tesla's cost reduction progress over half a year..

Gross Margin: ? (Luv's model estimates 18.5%); production was undoubtedly more efficient in Q1 than Q3; unsure about raw materials

And my point is that this is nonsense. You're saying that since Q3 - where Tesla had a margin of over 20% - Tesla's production has gotten less efficient?

Model 3 margins in Q4 stayed at over 20% despite the ASP dropping significantly from Q3. Meaning a significant drop in the base price. Which has beyond doubt gone even lower yet. Musk has made several statements that indicate that the base SR costs less than $35k to make now (but barely); it was $38k near the end of Q4. Even at an average production cost of $41k, an ASP of $54k would mean a 32% margin.

Do I think they're going to have a 32% margin? No. But 18,5%? Less than Q4 but with a higher ASP? As though Tesla's production process has gone back in time, and completely ignored the reduction in raw materials costs? Please....

 

[UnknownSoldier]

Back on the road again. Between the Supercharger and the Autopilot, road trips in my 3 are a dream.

 

[Lessmog]

Tesla and TSLA might be better off if government approval never happens. The public will not accept profit-seeking companies unleashing killer robots onto the nation's streets and highways. These robots would maim and kill scores, then hundreds of Americans annually as they multiply. Jurors would assess huge awards that reflect their indignation. Statistics that show that autonomous vehicles are safer than human drivers would not change this, because many people don't think that way. The only way out is for states or the federal government to enact legislation that greatly limits manufacturers' liability. I will be surprised if that happens. Complete FSD will make Teslas more marketable, improving unit sales and ASP, but without level 5 capability, and no legally required driver, and greatly limited legal liability, the big bux for Tesla will remain elusive IMO.

That's really why we should go back to horses or walking. There were, of course, a great number of fatalities with cars. And broken arms starting the fossil fueled for years.

Horses caused a large number of deaths as well.

Among those, unless I am misremembering, Pierre Curie was run over by a horse-driven tram.
But then again, people get struck down by electric trams too.

 

[MP3Mike]

Autonomy Investor Day is on the evening of the 22nd

Are you sure? I haven't seen a time released yet, I was assuming it would be during the day.

 

[Fact Checking]

Ask yourself why we don't have driverless trains everywhere. The tech was developed in the 1970s and perfected in the 1990s. It's at least 10x safer than human drivers and substantially more efficient economically.

I await your answer.

Because:

• Most railway companies do not operate in a competitive environment, costs can be transferred to customers or to governments, and there's little growth and little overall innovation. Big railway projects with halo products like the TGV exist, but are generally driven by national interests, not profit seeking.
• Railway crews are generally unionized, which will resist workforce reduction. Here's an article about the NYC Subway scuttling plans to halve the per train crew from 2 to 1: "the Transportation Workers' Union, which represents New York's subway workers, fought to restore two-person crews."
• On large trains, compared to the average number of passengers, the amount of cargo and the value of the vehicle the labor cost is a comparatively small cost. For taxi passenger service wages are the main cost factor.

The third one is the killer factor: it makes sense to hire a train crew for hundreds of passengers, and the labor cost will be a comparatively small part of the trip's total costs.

But for taxi services, the average number of passengers is only ~1.4, and it requires a dedicated driver who will idle about ~50% of the time even in relatively busy metro regions.

I.e. the utilization of taxi drivers is very poor: one taxi driver is transporting only one passengers on average if we consider idle time.

So even if we assumed a competitive environment for trains (which it isn't in most markets), eliminating the driver for passenger cars via FSD is a more than 100x times higher cost advantage than for trains. (!)

 

[Bladerskb]

Came across this comment near the top on that forum: "Their approach has my interest piqued immensely, but they don't even appear close to L3 on the reduced ODD of highways despite a lot of data there already."

Anyone using navigate on autopilot knows highway driving is already now at L3. It still has the annoying nags but we are doing L3 on highways. There is a large gap in perception vs reality of where Tesla is with autonomous driving technology. The demonstration on 4/22 is going to reduce that gap immensely. I believe it will make it clear that Tesla is at least neck and neck with Waymo if not ahead on FSD development. Obviously, the market definitely does not believe that right now or the stock price would be much higher.

Level 3 Def:
A Level 3 vehicle is capable of taking full control and operating during select parts of a journey when certain operating conditions are met. For example, a vehicle that is capable of managing itself on a freeway journey, excluding on- and off-ramps and city driving, might be considered Level 3 automated.

Do you currently have to pay attention and monitor the environment and the system's performance?
If Yes its still SAE level 2, if No, then its SAE level 3.

Level 3 allows you to watch a movie, read a book, etc because you no longer have to pay attention.

 

[Fact Checking]

Do I think they're going to have a 35% margin? No. But 18,5%? Less than Q4 but with a higher ASP? As though Tesla's production process has gone back in time, and completely ignored the reduction in raw materials costs? Please....

U.S. sales benefit from GHG credit sales, which increased Q4 ASPs. Without the FCA income in Q1 the higher consumer ASP in Europe is thought to roughly counter-balance the lack of GHG.

What is hurting Q1 most is half of S/X sales missing, and over 10k units in transit. That's a hit of over $2b in missing revenue, and at least a $500m hit to GAAP income.

A lot of good things need to happen in Q1 for those two strong forces to be countered.

 

[PhaseWhite]

Regarding Tesla's progress to Level 3 autonomy, I see some posters here making statements like were already there. That's just not true and hopefully through this post I can bring you back down to reality. I use NOA frequently and 2019.8.5 was a big jump in capability however there are still many gaps in functionality that prevent it from being a true Level 3 system. Just in the last few days driving around the Twin Cities I've encountered multiple scenarios where I the human had to take *immediate action* (not the graceful handoff with warning that Level 3 requires).

1. Taking a left hand exit, the car started to exit, then served back (for an unknown reason) so I had to take control to make my exit. It has worked before in that spot but not that day.

2. In Minneapolis area there are a ton of dual purpose exit/on ramp merge lanes that combine cars exiting at 75mph with cars entering at 40mph that are ridiculously short (like 100 feet) that NOA cannot handle at all. These are very difficult for human drivers as well. Sometimes NOA is smart enough to warn me that it won't be able to handle this via a "Unsupported maneuver warning" but other times it just tries and fails in a bad way.

3. NOA is only as good as the maps it's based on, and the maps are not being updated frequently enough. There's an exit lane that I need to take frequently but it's not aware that it's there so it takes a different route. In other parts of town, there's long term (2 years) construction projects on the interstate that have altered the exits and routes you can take. It's been this way for a year already but NOA has no knowledge of them.

For NOA to become L3 capable it will require some sort of crowd sourced mapping data from the cars on the road. I do think Tesla's vision system has improved leaps and bounds but other areas seem immature and I'm not sure they have the right software teams in place to solve all these data / computer science problems that need to be solved to enable L3 driving, which are separate from the vision problem.

Lastly, one of my biggest frustrations is that I cannot even set waypoints, so if I enable NOA it will drive me the way it wants to go and I have no way to set the route.

 

[EVMeister]

Are you sure? I haven't seen a time released yet, I was assuming it would be during the day.

The original news release:

"With a number of very exciting developments coming in the weeks and months ahead, Tesla will host investors on the morning of April 19th at our headquarters in Palo Alto to provide a deep dive into our self-driving technology and road map."

So I would assume it will still be in the morning (PDT), but 22nd now.

 

[MP3Mike]

A spin-off was my first thought the way Musk’s tweets were worded yesterday. Makes a lot of sense if the market is valuing self driving future capability of other companies at tens of billions of dollars, but that value is not currently reflected in TSLA SP.

Might also make sense in terms of getting in front of anticompetitive concerns. If Tesla Network might be a monopoly, and Tesla cars are the leading EV manufacturer, there would be a lot of pressure to break that up anyway.

Also makes sense in terms of a capital raise. Musk can continue his desire not to raise more capital for his car company, while Tesla Network Newco could raise capital for its fleet and tech development.

What if the Tesla Network spin-off is going to be the one leasing the Model 3s. That gets the bad effects of leasing off the Tesla books, and the spin-off will own the vehicles to put them in ride-share mode at the end. (And would be able to actually utilize the federal tax credits it earns.)

 

[MP3Mike]

Spin off doesn't make sense, as the data from Tesla cars produced are critical for AV. In terms of antitrust / monopolistic concerns, it will only come if Tesla sells Autopilot tech to other firms. Tesla doesn't need to sell this until it is at a mature state and there is regulatory approval.

The spin-off wouldn't own the AP tech, Tesla would still be in control of all that just like they are now when they don't own all the vehicles. The spin-off would just own a fleet of Teslas to run on the Tesla Network.

 

[Bladerskb]

The bullet points are my arguments and the performance and latency arguments as well - which aspect do you find inaccurate?

• There are ~10 billion neurons in the human visual cortex
• Each neuron recieves ~10,000 synaptic inputs
• Each synapse requires ~10 FLOPS based on a 1 Hz firing rate
• Thus: ~10 G neurons x 10 K synapses x 10 FLOPS = 1 PFLOP

Also HW3 (FSD Computer) is actually approximately 80-100 TOPS. This will be confirmed when its more in the wide and someone can crack it open.

 

[Navin]

A tweet worth a 1000 words.

 

[bhtooefr]

If they sell at current prices, even if the vehicle is shared through the TN, they will still be giving up a substantial amount of economic value.

Unless the FSD taxi market gets saturated. Then the only way to create more economic value is to sell to the non-taxi market.

Additionally, there could be value to reducing competitors' addressable markets even if it doesn't maximize Tesla's economic value directly.

And, of course, optimizing for economic value is not necessarily the best approach for the people and the world.