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

[SOULPEDL]

Shorty get rich quick scheme : buy a used Tesla, short a bunch of shares on a Friday, set it on fire over the weekend, put a video of it on YouTube. Media will take care of rest. Cover on Monday.

Exactly, and bring in the movie folks for a few Million. Easy. Also, there was some footage in that story if you keep clicking the links showing car parts scattered everywhere, including the cars next to it... flashlight inspection post fire. To do that, there would have needed to be an even bigger explosion after the initial one shown. But you can see the cars in tact after this fire. Does not compute.
I saw the second video late last night, anyone else see it? I havent check back since.

 

[Dangerous Fish]

I applaud your dedication and eye for detail but …

• … if it smells like a conspiracy, it often isn't.
  
  
• … if footage looks kinda off and some things don't add up, it's often due to some unlikely probabilities (eg. garbage CCTV tech).
  
  
• … if people think it's real, it doesn't matter if it was staged or not – until proven otherwise. The damage is already done.

You're third point is valid and I can only hope the Tesla investigation reaches a swift conclusion and gets to the truth. You're first two points are why I initially said here that I thought it was genuine. Fake videos are usually obvious on just the first or second viewing to those who work in this medium and I admit I was fooled, but there are too many anomalies I can see now which convince me it's been tampered with.

 

[powertoold]

That Tesla CCTV footage is soooo fishy. Here's a similar whiteout from an explosion, but the whiteout comes as a gradient rather than a fade to white:

 

[CaliBear1]

Unfortunately, we need an "OUTRAGEOUS" emoji option!

I totally agree. Every time something happens to Tesla, however trivial, it is news and TSLA takes a nosedive. Other companies have worse news and nothing happens at all. A fire of a car of another brand would not be in the news.,

I saw on the Dutch news site on the front page about the Tesla shrinking of the board, That on their front page!! As if a board change of any other middle sized American company would be that news worthy in the Netherlands.. You never hear that. But when it is Tesla...

 

[MarcusMaximus]

I appreciate the links. The details are very helpful. This absolutely proves that I'm right.

This is exactly what's very disappointing. They need to eventually hire some people with field-specific expertise in driving. They got none so far. Tell me when they hire some.


It's a "copy the incompetent" technique. It is literally impossible to make a good self-driving car this way.

Behavior cloning is fine. The problem is, quite specifically, whose behavior you're cloning; and who evaluates whose behavior you are cloning.

If you're going by "the majority of people on the road", well, we already know the majority of people on the road are bad drivers.


This is correct. Unfortunately, they're getting data from bad human driving, and then the data is being evaluated by bad human drivers.


That's all correct, but they're training it wrong. (To be clear about this, they're probably doing more or less OK on the vision / object identification stuff, it's the driving policy decision side I don't see any work on.)

To be clear, they can take the same neural network architecture and retrain it, probably in a matter of weeks, using competent people -- which may put them well ahead of any "competitors" -- but they haven't started yet.


As far as I can tell, he's totally unqualified to make those decisions. Zero qualification in his LinkedIn profile for making such decisions -- zero. They need someone who actually knows what the vehicle really ought to do, not some random engineer.

They're making path planning decisions based on "average driver behavior" (terrible), and the random whims and tastes of engineers at Tesla with no particular driving qualification (perhaps better but still very bad). That's just wrong. Hire some professionals to answer the path planning questions, and let the engineers stick to implementing it.

I don't think the field of safe driving commands huge salaries; Tesla should be able to afford a few specialists.

In terms of control(as opposed to classification), I don’t believe we know how they’re training them. The most likely method would be to use unsupervised reinforcement learning, which generally doesn’t require human labels at all.

They likely define a set of “bad” states(crash, near miss, unexpectedly leaving lane, missing required exit, entering pedestrian crosswalk at intersection, etc) and just train to avoid those states while still making optimal time. With classification already working and “known-good”, those states become trivial to detect automatically. Since the ultimate goal is to improve on safety, crash and near miss, alone, become signals for any needed improvement in other areas. With as much data as they have, anything truly unsafe will result in one of those two.

All this is why I brought up my earlier point: they may end up with a system that is extremely safe(far safer than any human driver) and efficient, but doesn’t necessarily follow all the rules of the road. Best way to deal with that would actually be to change the rules of the road(or provide exceptions for autonomous cars).

 

[Pras]

Just for fun I bought a few long call spreads for July 315-365 for $6.75.

 

[anthonyj]

Alright, Market Makers sold enough puts, time to launch

 

[neroden]

humans aren't bad at driving. they drive extraordinarily well most of the time. stay with me: if all humans always drove the way they do *most* of the time, there would be virtually no accidents.

Nice theory, and if it were true you'd be right... but it's false

 

[humbaba]

I don’t buy neroden’s argument at all! AI engineers need to be experts in neural networks, training data, and analyzing results. No need to be a professional driver, lol.

The AI being fed enough data will always outperform engineers trying to understand the world to come up with some smart algorithm. This is called software 2.0, the data is writing the code, not humans.

Your basic point as applied to FSD is what Karpathy addressed in a video linked a bit ago. According to him, the programmers are the data labelers. Yeah, in some pedantic sense you are right, but in a practical sense humans are still doing the programming. If you alter the labeling (code) then the results of the neural net training (compiler) are different.


umm... topicality? erm... Tesla's quest for FSD and their process for doing it which appears to be different (more reliant on neural nets than others) which will be in the forefront in about 90 minutes.

 

[Dangerous Fish]

Another video taken of the clip has emerged. Now I'm not sure again.

 

[SOULPEDL]

My hope for the event today and the earnings report on Wednesday is that they announce an intent to spin-off the autonomous driving division, as well as announce that Fiat-Chrysler (or some other vehicle manufacturer(s) ) are going to license their autonomous technology.

A guy can dream can't he !

If not this week, I think what I mention above is inevitable at some point in time.

Wow, people disagreeing to a possibility? I like the ideas myself.
I think we will be surprized today. I already posted my predictions in detail, so we'll circle back later today on that...

 

[neroden]

OT
TBH, this looks like a large volume propane release (low level fog, flame spread).
Massive propane leak Thursday in Manchester

Yes, that would do it. Rigging a propane tank to catch fire isn't that hard and the camera angle is certainly set to hide any explosives or flammables which were placed under the car.

 

[AudubonB]

A heads-up to cheer you up:

If any company competently can determine what's what with smoke patterns, spread, timing and so forth, it would be a company that has unparalleled expertise in the forensics of rocket RUDs. Does anyone know of any such?

 

[mongo]

All this is why I brought up my earlier point: they may end up with a system that is extremely safe(far safer than any human driver) and efficient, but doesn’t necessarily follow all the rules of the road. Best way to deal with that would actually be to change the rules of the road(or provide exceptions for autonomous cars).

Or realize people don't either and that some situation require not following the normal rules of the road. (two lane roundabout that is only one lane due to construction)

My concern with using 'experts' to train driving is that the cars are driving with 'non experts'.
marquis of queensbury rules vs street brawler.
Driving:
1. Don't hit anything
2. Obey laws
3. Head toward your destination
(with 2 and 3 intermingled)

 

[EVNow]

Unfortunately, they're getting data from bad human driving, and then the data is being evaluated by bad human drivers.

That is really speculation, right ? You are basically assuming worst. Nothing I've seen in AP makes me feel it is being trained by bad drivers (rather than problems with recognizing objects).

I mean, sometimes I may drive badly (usually when distracted) - but can definitely figure out what is good and what is bad driving. We should expect that even average drivers can pick out what is good driving and use that for training the network - because it is easy to recognize and throw away obvious bad driving.

 

[AlexS]

I am extremely (perhaps overly?) skeptical of conspiracy theories. With that said, the way that smoke appears from one from frame to the next, and the way it looks right before the whiteout, looks very fake to me. Edited to add: the whiteout also strikes me as fake.

I think the video is real thing. Other thing if it was sabotaged or just battery failed, investigation will tell.

 

[scaesare]

It's not like "expert" drivers are applicable to regular driving either. Sure a professional driver is probably far safer than the average person, but it's not because they know how to drive better, it's because of greater situational awareness and attention.

Well, I'm not entirely sure that's true. In addition to being aware of a situation, you also have to have the skill/expertise to then deal with it.

Example: there are lot of drivers aware they are in a skid, but don't have the expertise to counteract it.

 

[MarcusMaximus]

That is really speculation, right ? You are basically assuming worst. Nothing I've seen in AP makes me feel it is being trained by bad drivers (rather than problems with recognizing objects).

I mean, sometimes I may drive badly (usually when distracted) - but can definitely figure out what is good and what is bad driving. We should expect that even average drivers can pick out what is good driving and use that for training the network - because it is easy to recognize and throw away obvious bad driving.

@verygreen can probably verify, but I don’t believe the network is currently doing any controlling of the car. It’s just detecting and classifying objects and lane lines.

 

[mongo]

Well, I'm not entirely sure that's true. In addition to being aware of a situation, you also have to have the skill/expertise to then deal with it.

Example: there are lot of drivers aware they are in a skid, but don't have the expertise to counteract it.

Ah, but that is stability control, and they do have experts on vehicle dynamics code that part.

 

[lascavarian]

Whoa there—citation please? I don’t recall this. It would be major news and would conflict with the Tesla position on service since time immemorial.

Regarding Tesla designing their corporate structure such that service is not a profit center. This draws a striking contrast to the current dealership system of traditional manufacturers where service is a large profit contributor. Dealers don't do it this way because they are stupid. The dealer system works but perversely IMO.

But delivering great service can be handicapped if there is no incentive for profit. Profit sharing helps a bit but really portions of a service delivery enterprise work best if on the profit line where incentives are ever present and growth is self-funded. Leaving service on the expense line makes it ripe for budget cuts at the worst possible times (in my experience).

One possible solution:

Tesla provides a product with a warranty. Tesla can do direct warranty service for which they fund internally.

Tesla could defer from doing out of warranty service but establish certified providers that are geographically diverse. Local service enterprises can fill in with certified parts profitably supplied by Tesla as well as Tesla certified training. This would be a small business at first but grow over time. (This is close to what the rangers seem to be filling in for as the company grows)

Tesla could warrant certain SW for life so they remain fully involved with all things SW related.

As a manager spending a few decades in “service”, I will say that the most efficient way to deliver poor service is to take it off the profit line.

This is controversial particularly with sales and marketing and sometimes engineering but my personal experience supports it. Nothing wrong with making a profit for doing an outstanding job for customers (this requires independent verification).

Profit is a tool that delivers many subtle benefits if managed wisely. Structurally, service has to be organized so that baseline profit flows from an incentive for reliability (what the customer really wants) and not from unreliability. Just my thoughts....