Search results

The exam can be given orally. It's even written into law in some states. FSD can be illiterate and still function as a robotaxi. And since FSD could be technically classified as dyslexic, the robotaxi would qualify under the Americans with Disabilities Act and be entitled to some dandy...

Understanding semantics is a nice-to-have feature. Like I said, illiterate people can drive and be granted a driver's license. It's not a serious obstacle to robotaxi rollout.

As I understand it, FSD doesn't need to understand the semantics of language on road signs at all. It just mimics what a human does after seeing a particular sign. For instance, it knows the design of a yield sign and the shape of the word "Yield" but it has no idea what words and letters...

Yeah, 12.3 can do a little bit of pulling over. But it's still got a long way to go before it could do everything it needs to do for picking up passengers. I'm hoping that Tesla has already been working on this. We do know that Actually Smart Summon is being developed and that gives some of...

I'm not sure if this counts, but I know that the Chevy Bolt and Chevy Sonic once shared an assembly line at the Orion plant in MI. They were both designed by Daewoo in South Korea, but the Bolt was an EV and the Sonic was ICE. They were of similar size, but I'm not sure if they really had that...

From the transcript: I could have sworn the answer was "No."? Which is correct?

A lot of Uber drivers use Teslas. So Tesla may already have enough data to do the robotaxi-specific training. But I don't think that training is reflected in the current FSD build. The car needs to be able to pick out who the passenger is on the side of the road and figure out the safest...

The app is the easy part. The back end is a little harder. But the part I'm most interested in is the neural net. We don't know if Tesla has started teaching the car how to be a robotaxi with pickups and drop-offs. I'm hoping we learn more about that on 8/8.

Me too. It is fascinating how after this conference call it seems like a cloud has been lifted. While these past few weeks have been trying, nobody has been more steady and reassuring than @Krugerrand. He is one cool cat.

I think you just answered your own question. SpaceX was already in Texas. I've seen a lot of committees scout out new locations for a company headquarters. They consider all the economic factors and what is best for the business. Somehow, that almost always happens to be close to the place...

Yes, the hybrid part was definitely political. It is there because of Joe Manchin (D-WV). He was the swing vote and he was carrying water for Toyota, who has a large presence in his state. IMHO, the materials sourcing requirements are good policy. The US needs to develop its own supply chain.

Correct. This is why the levels don't matter in the real world. For example, Tesla FSD is far, far superior to the Mercedes system that is classified as L3.

Nobody, including Waymo, has actually explained Waymo's path to profitability. Their weaker technology is holding them back. The vehicles are too expensive to produce and the system requires too much human intervention. Waymo just doesn't have enough intelligence inside the vehicle. Elon...

You have stopped responding because you can not explain to us how Waymo becomes profitable. If there is hope for Waymo, I'd really like to know about it.

This sounds great. But Waymo still has to figure out how to operate profitably. As long as frequent human intervention and monitoring is required behind the scenes, Waymo can never be profitable. The only way to cut out the expensive humans is for the "Waymo Driver" to make better decisions...

According to this, Tesla is already stockpiling cells from LG. I think they are 2170s? https://www.teslarati.com/tesla-stockpiled-batteries-lg-energy-solution-q1-report/ I think there might be two reasons for the delays in building out the Semi factory: 1. Elon is trying to save cash in the...

So I think this is the first we have heard about robotaxi-specific neural nets? The system needs to learn how to pick up and drop off passengers. It could learn this from the many Uber and Lyft drivers using Teslas. I have assumed that Tesla was working on some robotaxi-specific nets, but...

More cars on the road just means more humans in the background monitoring each ride and coming to the rescue of stuck vehicles. The bigger the fleet, the more money Waymo loses. If that was not the case then Waymo would already be massively growing their fleet. The only way Waymo can win is...

As I understand it, Tesla does exactly what you describe. But Tesla still needs to test the opposite scenario to answer the question, "What does FSD do that makes users disengage?" That's a different metric from just finding out the delta between what the human did and what FSD would have done.

It's actually kind of weird that I constantly run into Tesla-praise from Drive-Thru workers. They see cars all day long, and I'm always getting, "Hey, I love your Tesla." Happened again just yesterday in my Model 3.

That's what a stupid engineer might do. Say what you want about Elon's politics. He's not a stupid engineer.

Exactly! FSD is essentially a software play. And we have a saying in the software business. "Software is like dealing drugs. Our customers are called users and the first hit is free."

My personal theory is that Tesla lowered the FSD (supervised) subscription price to $99 is because they want to show OEMs lots of customers who love it, use it, and pay for it. The OEMs will have to do a licensing deal with Tesla or risk obsolescence. The reason the price of fully-paid FSD...

Tesla gets different data from subscribers and non-subcribers. From subscribers, Tesla gets disengagement data. From this, Tesla learns about errors in the current system. From non-subscribers, Tesla gets human behavior data. From this, FSD learns to drive like a human. Both are important...

Ah. So "in-field retrieval". Thanks for the info.

So here is what it said about Cruise: But we know that almost every drive was being guided remotely by a human at some point. So what counts as a disengagement? And every video I have watched recently of a Waymo drive has some part where the car is stopped and a message comes up and says, "We...

RT is a great business as long as you can make a profit. Waymo has not shown a path to profitability because their technology is inferior. Not only are Waymo's specially-equipped vehicles expensive, their operation is still highly dependent on lots of human labor monitoring behind the scenes...

So they will spend $6 billion in order to lose $12 billion? I don't think Google will go for it. Honestly, we don't know the numbers. But until Waymo says, "We see a clear path to profitability.", there is no point in discussing Waymo as a competitor to Tesla.

So let's say it turns out the HW3 vehicles are never able to be used as robotaxis. Is Tesla open to a class action lawsuit? Well, they could try. But Tesla never put any robotaxi promise in writing. They would have to argue that Elon's public pronouncements constitute a verbal contract. And...

Hoes have a status?

There is a decent chance that will happen this year. But you still shouldn't sell.

Excellent post. This is exactly why going balls to the wall for autonomy is such a blindingly obvious move. We are going to see massive improvement at a very fast pace. I don't see how Tesla gets to another local maximum any time soon unless they run into hardware limitations. But even then...

True, but it is how humans troll.

Or it's the opposite. Would it be possible for Elon haters to buy up shares, vote their shares, and then sell off before the vote? I guess this wouldn't be enough to actually influence the final vote unless it's really close.

Not sure if this is good news or bad news.

Good point. It might.

Yeah, I saw Dillon talk about this last night. He presented a good case as to why we should be worried. So I'd really like to see a breakdown of who controls or influences the voting for large blocks of shares and how they are likely to vote.

That sounds great, but I don't think Tesla can cover all the liability of unsupervised FSD at only $99/month unless they get a huge take rate. I expect that it will take awhile get the take rate high enough. In the mean time, Tesla would be liable for everything from curb rash to fatalities.

I haven't seen anyone do an actual breakdown of how we get to a majority vote. While I don't think the board would have put this on the ballot if they didn't think they would win, I'd like to be reassured with some actual numbers showing how many shares are a definite "yes". Has anyone here...

Or maybe not?

He also says this about road trips: People often ask me about charging on a road trip. They will say, "How do you know where the chargers are?" I've come up with something even better. No need to mention route planning. I say: I've literally had people stand there with their mouth open.

Not huge, but at least going in the right direction. Tesla Model 3 wait times get pushed back in China

Oh, I know. Pick me Mr. Kotter. Pick me. Money.

FSD is a lot like landing rockets. It's a breakthrough technology that nobody will copy for a very long time. FSD is also like landing rockets in that it unlocks profit opportunities that go beyond the technology itself. Just as landing rockets enables Starlink, FSD enables robotaxis and OEM...

Yeah, if you are looking for a quick turnaround for the stock then one FSD OEM announcement would do it.

Awesome! This has been low hanging fruit for a long time. Tesla knows exactly when you will arrive at the Supercharger, so why not let customers order lunch to be hot and ready? I don't know how much profit Tesla will be able to generate from this service, but it's so good for the customer...

These numbers were just an example. Presumably, Tesla will develop very good data about the risk of unsupervised FSD. They will price it such that they can pay out claims and still make a healthy profit. But what you say may be true. It may be that unsupervised FSD has double risk of...

Thanks for that. Though an experiment of one has obvious flaws, I think the Uber guy provides better data. At least we can assume his reporting is consistent.

I wouldn't be at all surprised to see Tesla make a deal with BYD to put FSD in its cars. But I think it's far-fetched to think that Tesla would outsource the manufacture of dedicated robotaxis.

12.3.4 is about the same or slightly better than 12.3.3 for me. Is your experience with V12.3.4 worse than V10 and V11?