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

[EVNow]

Tesla is not seriously working on policy. They do not have a single employee with the expertise to do so. You can't have a bunch of Silicon Valley engineers and programmers work on designing driving policy -- they do not have the knowledge or skill set. When they start advertising for expert driving instructors, you'll know they're working on policy.

But then - driving instructors specialize in teaching you to pass the driving test.

Why would deriving the policy using normal driving - with human oversight by regular drivers to get training data be all that bad ? This would give real world rules instead of bookish ones.

 

[ggies07]

I might have missed that this was post already, but if not.....

 

[anthonyj]

I read some articles written by a great hedge fund manager named David Einhorn and an auto/EV professional Anton Wahlman. They have great track records and their articles are featured on Google News so it must be legitimate. After reading these fine, unbiased articles, I’m inclined to sell all of my shares and short the stock via Puts! I would like to do this before the 22nd, because there are some genuine people on Twitter saying that Tesla may announce bankruptcy on April 22nd. After doing some digging, I’m finding that this may be the case. See attached:

Tomorrow I will be selling all of my stock and buying a ton of $10 puts with my sale. I’m so happy to find my new friends at $TSLAQ!

 

[Steve m]

But the point is that spot is irrelevant to short term supply disruptions in Tesla's supply chain, no matter the price. Firstly because the majority of the "battery grade" material on the open market is not "EV grade", but also because not even "EV grade" lithium can be substituted into EV batteries at short notice - all lithium from different plants is different and it takes a long time to be qualified.

This sounds like a rehashing of Jon Petersen's Tesla and the Cobalt cliff.

 

[sundaymorning]

I went back and looked at Navigant’s “who’s leading in autonomy” chart. It’s a joke.

Does ANYONE really believe Nissan and Toyota are ahead of Tesla on this front? Anyone?

Tesla is nearly dead last on that chart. Yet they are the ONLY manufacturer that can even change lanes autonomously, much less navigate freeway interchanges and exits. Does anyone dispute that?

Cadillac’s Supercruise illustrates the narrow thinking. Yes, you can map some highways and get a reliable system on that. But that solution is not scalable. At all.

Waymo can map a city and effectively drive it autonomously. Great. That makes you appear to be in front. It’s not scalable. At all.

We are at the cusp where the GENERAL solution that Tesla has will start to outperform the specific solutions that Cadillac, Waymo, and Cruise employ.

Even Navigant doesn’t seem to get this yet.

Recall that all of the experts thought that what happened with Falcon Heavy this week was impossible.

It happened, and SpaceX has now left EVERYONE else in the dust. No company or nation on earth can match them.

It will be the same with Tesla’s AP.

Elon’s strength is in looking beyond the self-imposed boundaries that other solutions take and approaching the problem from a fresh direction. Doing so makes you look behind the others at first. After all, Cruise is already driving around autonomously in San Francisco, right? Look at all those Falcon 9s crashing into the barge!

We are getting close to the point where the Falcon 9 stuck its first landing. There will be an inevitable sea change in opinion where Tesla goes from appearing to be way behind (which itself is ludicrous) to being recognized as so far ahead that they can’t be caught.

In AI, data is king. Nobody has more than Tesla.

Elon’s greatest asset and perhaps also his own cryptonite is that he can’t be axed or let go. This allows him to push AP and new ideas to its boundaries and limitations. Any director, executive, CEO would be too chicken scared to push the company’s technology to its limits. Because waiting on the other side of success is failure, and if you fail as a corporate executive, you’ll be pushed out/shown the door and burned at the stake. For this reason I believe Elon is able to take risks that others are unwilling to do.

 

[ZsoZso]

You are clearly misunderstanding the document.

Yes at L3 the system won't be able to HANDLE all situations that will stop it from completing the full DDT and the fallback is to hand over to the person behind the driver seat. But the system MUST be capable of recognizing EVERY CIRCUMSTANCE that it needs to hand over control. That hand over is not instant either, its not immediate.

It seems to me, you are the one who misunderstands the document. The example I quoted was exactly pointing out a situation where the system is NOT ABLE to recognize the problem, such as a vehicle mechanical failure, so the human has to recognize it and take over without ADS issueing a request to intervene.

NOTE 3: "The driver state or condition of being receptive to alerts or other indicators of a DDT performance-relevant system failure, as assumed in level 3, is not a form of monitoring. The difference between receptivity and monitoring is best illustrated by example: A person who becomes aware of a fire alarm or a telephone ringing may not necessarily have been monitoring the fire alarm or the telephone. Likewise, a user who becomes aware of a trailer hitch falling off may not necessarily have been monitoring the trailer hitch. By contrast, a driver in a vehicle with an active level 1 adaptive cruise control (ACC) system is expected to monitor both the driving environment and the ACC performance and otherwise not to wait for an alert to draw his/her attention to a situation requiring a response (see 3.23)."​

Are you claiming, that you can see a trailer hitch falling off a truck in front of you when you are watching a movie and not paying attention to the road ?

The document says:

"However, for level 3 ADSs, as well as for level 1 and 2 driving automation systems, the human driver is assumed to be receptive to vehicle conditions that adversely affect performance of the DDT."​

Exactly, the actual definition of the level 3 system is on Page 22, which clearly states that the DDT fallback-ready user must be receptive and take over in a timely manner.

Our disagreement comes down to the interpretation of receptive DDT fallback-ready user.
You think a person watching a movie or reading a book qualifies, and I absolutely disagree based on several examples cited by the SAE document. If the "driver" is not paying attention, then s/he will not be able to notice a problem / system-failure fast enough and take over in a timely manner as required by the SAE definition.

I'm guessing Toyota, Audi, Mercedes, Mobileye and others who have been trying to asses/launch a Level 3 system and are working with the regulators to do so are wrong?

The Toyota research exec on the video is a clear example of deliberately spreading FUD, trying to explain away why they do not have and not even working on a level 3 system, because it is "too hard" in his incorrect interpretation. He was trying to make a ridiculous claim that the system would need to recognize ~1500 yard ahead that there will be a problem so it can alert the movie-watching user to take over in time. By that interpretation nobody can ever make a level 3 system, because it is simply impossible.

 

[M3Rider]

I don't think AIs have a chance in hell in health care; we don't really know what we're doing in health care, like, at all, and AIs are totally not anywhere near able to handle that level of lack of information and misinformation and wild guesswork. (There are some areas of health care which badly need computerized search databases, however.) AIs are potentially quite useful for some types of education, but probably unusable for others. Social services are similar; much of social services should really be automated (again, no AI needed), and is only non-automated because of government policy decisions to try to deny people coverage. Other parts (dealing with hallucinating people is the first one which comes to mind) are so individualized there's really no scope for an AI.

These are some quite wrong assumptions...
Have you checked Elon's tweets lately? Elon Musk on Twitter
Having experience of many doctors, I would trust an AI to make a better diagnosis/prediction than any single doctor with a limited experience going on a faulty memory.

Also, this doesn't need to be an AI, but a scalpel in a robotic arm can surely make a lot less mistakes than some surgeons whose hands are shaking due to numerous factors that could lead to that... “Every surgeon carries within himself a small cemetery"

 

[Mike Smith]

Yes, Dana Hull gets paid more if her stories move the market:

Bloomberg News Pays Reporters More If Their Stories Move Markets

This is an obvious conflict of interest for a reporter. The more embellished your claims, the more you get paid.

 

[PhaseWhite]

Elon's tweeting today has me wondering if hes got a new agreement signed with the SEC set to go in effect in the near future (and in hes taking advantage of his leeway ahead of that). If not he is certainly kicking the SEC hornets nest today with not 1 but 2 potentially material tweets about the actual gigafactory capacity and 500k cars n the next 12 months statements. Yeah, the 2nd one may be up for debate but somehow I doubt hes been reviewing any tweets with legal council today.

Really hope hes got a deal as this will give fresh ammo to the SEC to argue for sanctions in front of the judge and that wouldnt be very smart.

 

[Singer3000]

Or, indeed, the existing global *road* inventory... this is why it's going to start out working only on particular roads.

Versus your unmarked dirt track comparison, it is intuitively far easier to train a NN to drive in the centre of a nicely defined highway lane with good clear white lines and cats-eyes every 10 metres. But it's not obvious to me why you think it would be impossible to train a NN to recognise where the road edge begins and the grass verge starts. That seems a far easier thing to achieve than tarmacking and line-painting every obscure rural road in the world.

The bit I can see that is trickier, is for example the behaviour response when driving on a very narrow country road, when an oncoming agricultural vehicle cannot pass without you first reversing halfway off the road onto the grass bank. But as I said in my previous message, what percentage of journeys carry a greater than zero probability of this type of event?

 

[Silent Ludicrosy]

I.e. unless I'm missing something the correct calculation is more like 1 Tesla-FLOPS per synapse, and that's probably generous due to the low information content of human synaptic strength.

Tesla-FLOPS? New beach footwear to go with short-shorts?

 

[bdy0627]

Are you certain there is no reason to slow? Safe drivers will slow in certain situations where other drivers see only clear lane ahead.

E.g. You are about to pass a slower car in adjacent lane. Behind you is a truck. The good driver assesses to probability that the slow car will pull into his lane and that he will be rear ended by the truck. So he passes the slow car with a reduced speed differential. After passing the slow car he returns to the speed limit. At no point was there anything other than clear lane ahead.

No apparent reason. Shadows maybe? Breaking occasionally is applied quite firmly and startlingly, without apparent reason. That's not something I really ever do on the highway unless an accident is imminent. Seems to be improving with updates.

 

[bdy0627]

He he "niggly" my new favorite word. Thank you!

Careful with use of that one. It might not go over well in some situations.

 

[Bladerskb]

It seems to me, you are the one who misunderstands the document. The example I quoted was exactly pointing out a situation where the system is NOT ABLE to recognize the problem, such as a vehicle mechanical failure, so the human has to recognize it and take over without ADS issueing a request to intervene.

Frankly it looks like you are refusing to read the document. You are literally posting a soundbite and claiming the soundbite is actually saying what you want it to say rather than reading the entire document as a whole and taking it in context.

Are you claiming, that you can see a trailer hitch falling off a truck in front of you when you are watching a movie and not paying attention to the road ?

This is a classic example of you not understanding what the document is saying. Its explaining the concept of monitoring versus being receptive. That a user being receptive is NOT MONITORING. So if you were reading a book and your car starts beeping, you will become aware of it. That's what receptive is.

EDIT: It uses an example of you being in a car where the trailer hitch fell off. Its not talking about a trailer hitch of a car next/in-front of you. But a trailer hitch of your car that you are driving and yes you will feel it if your trailer hitch fell off. But you weren't monitoring it, just like you will be aware of a fire alarm but you weren't monitoring it before hand.

Exactly, the actual definition of the level 3 system is on Page 22, which clearly states that the DDT fallback-ready user must be receptive and take over in a timely manner.

receptive IS NOT MONITORING.

Our disagreement comes down to the interpretation of receptive DDT fallback-ready user.
You think a person watching a movie or reading a book qualifies, and I absolutely disagree based on several examples cited by the SAE document. If the "driver" is not paying attention, then s/he will not be able to notice a problem / system-failure fast enough and take over in a timely manner as required by the SAE definition.

There is no disagreement, you are simply wrong. The driver DOESNT pay attention. The user is NOT SUPPOSED to notice a system failure, the ADS system does and handles it while providing sufficient time to hand over back to the user.

There are two responsibility of a user in a Level 3 system.

1) Be receptive of system requests to take over which happens through audible,visual, or and vibration feedback.
2) Be receptive to "kinesthetically-apparent vehicle failure" like a "broken tire rod", "broken suspension component"

"A vehicle with an engaged level 3 ADS experiences a broken tie rod, which causes the vehicle to handle very poorly, giving the fallback-ready user ample kinesthetic feedback indicating a vehicle malfunction necessitating intervention. The fallback-ready user responds by resuming the DDT, turning on the hazard lamps, and pulling the vehicle onto the closest road shoulder, thereby achieving a minimal risk condition."

​

If the "driver" is not paying attention, then s/he will not be able to notice a problem / system-failure fast enough and take over in a timely manner as required by the SAE definition.

This completely contradicts the SAE document as it explicitly says that the user DOES NOT monitor the environment nor the performance of the ADS system.

The Toyota research exec on the video is a clear example of deliberately spreading FUD, trying to explain away why they do not have and not even working on a level 3 system, because it is "too hard" in his incorrect interpretation. He was trying to make a ridiculous claim that the system would need to recognize ~1500 yard ahead that there will be a problem so it can alert the movie-watching user to take over in time. By that interpretation nobody can ever make a level 3 system, because it is simply impossible.

That's exactly what a high speed Level 3 system will need to provide. That's not FUD, that's just you refusing to read the document. A traffic jam level 3 can get away with something like 5-7 seconds. But a 80 MPH for example would need 10+ seconds.

"In the event of a DDT performance-relevant system failure in a level 3 ADS or in the event that the ADS will soon exit its ODD, the ADS will issue a request to intervene within sufficient time for a typical person to respond appropriately to the driving situation at hand"​

Notice how the driver is NOT paying attention after activating the system and is instead playing with his kid in the back-seat?

 

[Zaxxon]

They are really, really bad at software.

The worst. Except for all the rest.

 

[ZeApelido]

Hai again, did Elon do something to a horse?

 

[mblakele]

If FSD requires developing a "driving policy", then I guess Tesla's FSD will fail. Because Andrej Karpathy is convinced that "Gradient descent can write better code than you. I'm sorry."

I don't think he'll change his mind until and unless events force him to. I'm not 100% convinced he's right, but I'm disinclined to bet against him.

Andrej Karpathy: Software 2.0

 

[Causalien]

It's a little early, it should be August 29. Or, maybe it goes live on April 22 and becomes self-aware on August 29. Hmmm....

I think the movie august 29th is judgment day. Self aware was august 4th. So we got about 4 months.

 

[CalX]

35k model 3 delivered. From the Facebook group:

 

[printf42]

Not sure why people are so hung up on the L1-L5 definition now.
People need to understand sometimes industry standards are not drafted by the real innovators in the forefront of a fast changing field.

Sometimes an established industry would fail miserably at predicting and standardizing future technology.
I am not saying the L1-L5 autonomous driving classification had failed, I am only trying to make the point that these classification and standards and regulations are not set in stone, and sometimes would need to be pushed and sometimes re-wrriten.

A closer example could be the OSI 7 layer network model, it was drafted with good understanding of the technology at the time, but later with many innovations that's blending layers or creating services that does not belong to any defined layers, many are arguing whether the 7 layer model is still relevant or not.

I think similar would happen to the L1-L5 definition, with Tesla as the main driving force that could lead to a more practical path to clear legislative obstacles, and redefining the classification system on its way there.

tl;dr:
What Tesla is going to build could be hard to put into any of the L1-L5 Autonomous classification, so don't try to force it into one.