I’m quite certain they are looking at the endgame. They are spending a lot of money on this, and there’s no chance the BoD would sign off on this if they didn’t see a pathway for a payout. They also have outside investors that they have to answer to to a certain extent. They might be talking a calculated risk, but they aren’t spending this kind of money if they didn’t think it would pay off.
And FWIW, Tesla has had a few whiffs too. The Solar City acquisition and the Cybertruck. Having a whiff doesn’t mean you don’t care about ROI.
100% agree. That’s the whole bet right there. That’s why the article and info graphic albaby1 shared are so interesting. If we take it at face value, Tesla had completed the faster part of the performance curve and is now in the slower part. Waymo has completed the slower part and is now in the faster part.
That’s not quite what I said. I said Waymo may not be approaching profitability in the same fashion as Tesla. By that, I mean Waymo may be willing to spend more to get there and less concerned about the sunk costs to do so or operating costs, so long as it is profitable. Tesla may simply not be in a position to spend capital in the same manner. It’s sort of like a owner of professional sports team that has a financially more important main enterprise. He or She still wants to make money but degree of profitability may not matter as much.
It is interesting but its all speculation until the results are in. I think the important comment by @albaby1 is that there may be degrees of autonomy that provide revenue generating opportunities short of Level 5. But, it begs the question of what the consumer may be willing to pay for something better than driver assist but short of the holy grail of autonomous driving. We shall see. Again, I still don’t have any idea what the profit potential looks like for either company. So, it could be a win where the winner generates little in terms of net benefit to shareholders.
Well, I mean that’s a given. Plenty of companies are generating revenue today with current autonomy tech. Whether they’ll ever generate any profits is another question entirely! I suspect they might not, but that might depend on a lot of more “squishy” things like human behavior and consumer preferences rather than technology.
I checked the six 9s claim and find it fair. What I cannot fathom is why Tesla says it needs eight 9s which is two magnitudes, 100 times safer. I had a chat with Google AI.
At first it justified the eight 9s based on the Cybercab fleet size. The human six 9s is per human, not all humans driving. It replied that Cybercab will drive many more miles than humans and therefore cause more injuries, which is a fair argument but does not justify a two magnitude increase. I asked about seven 9s. Here is the reply:
The Math of Seven Nines (99.99999%)
Definition: Seven nines reliability translates to 1 injury per 10 million miles traveled.
Cybercab Utilization: A commercial cybercab operating continuously might log 50,000 to 100,000 miles per year—significantly more than a personal vehicle.
Expected Lifespan per Injury: At 100,000 miles per year, a seven-nines cybercab would need to operate for 100 years before statistically accumulating a single injury accident. This easily outperforms an individual human’s lifetime safety record by a factor of 15 to 20.
I wonder what the Tesla executive had in mind. Any ideas?
If I had to guess, it might be a belief that autonomy will not be accepted - or that Tesla won’t choose to rapidly deploy - unless it is vastly safer than human driving. That it’s not just enough for AV’s to be as safe as a human driver, but that they really do have to be close to 100x safer than a human to be widely accepted.
That may also be why Google AI gave you the first answer. Yes, it’s true that at 100K miles per year, a seven-nines cybercab would go 100 years before accumulating a single injury accident. But if you scale to even just Waymo’s numbers (4K robotaxis doing 2 million miles a week), seven nines would mean you’re going to have someone hurt by your cabs every month or so. If you want zero accidents, as Musk has suggested, seven nines won’t let you avoid running up some injuries at any material level of scale.
The total number of deaths changes from year to year. A single crash of a large plane can cause a big jump in the numbers for that year. According to data from the Aviation Safety Network and the International Air Transport Association (IATA), recent totals include: [1]
Flying on a major commercial airline—like the ones you book for holiday trips—is much safer than flying in a tiny, private plane. [1, 2, 3]
Commercial Airlines: Major airlines go years at a time with almost zero fatal accidents. In a typical year, there are only about 5 to 8 fatal commercial airline accidents across the whole planet. [1, 2, 3]
Private Planes (General Aviation): Smaller planes, corporate jets, and medical helicopters make up the vast majority of plane crashes. In the United States alone, private plane accidents cause about 300 to 400 deaths every year. [1, 2, 3, 4]
How Safe Is Flying Really?
Flying is statistically the safest way to travel. Aviation experts calculate the risk of flying by looking at the number of deaths per mile traveled. [1, 2]
The Odds: The chance of dying in a plane crash is about 1 in 11 million to 1 in 37 million.
Compared to Cars: You are much more likely to be hurt in a car than a plane. In the United States alone, around 40,000 people die in car accidents every year. Your lifetime odds of dying in a car crash are about 1 in 93 to 1 in 107. [1, 2, 3, 4, 5]
I agree. But Musk has said that they need to be excessively vigilant during the roll-out and scaling of the robotaxi, because “if we injure even one person, it will be worldwide headline news.” I’m not sure that’s true, but it seems to be what they think is the case. So maybe that’s what the Tesla engineer was thinking of.
Sure. And it’s also possible the engineer may have said eight-nines just for PR purposes as well, and Tesla doesn’t plan to use that as a guide to their safety levels. Who knows.
I’ve noticed Tesla execs tend to be more…grounded than Elon in their public statements, but they still tell some absolute whoppers from time to time. He might have meant what he said, but he could have been telling a tall tale too.
If I apply Occam’s razor, I have to conclude that their AV ride hailing program simply doesn’t work very well, and that’s why the rollout pace is glacial.
I mean, Musk said we haven’t seen the Cybercab in operation because it is a new chassis and they have to train up the FSD. Makes sense. Except that they unveiled it almost two years ago, and presumably were working on it for some long period of time before that. How much time do they need?
I agree — it’s a confusing time and especially so for this subject. First we hear that they are completely rewriting FSD V15 for the robotaxi again. But my understanding is also that additional work adapting cybercab will take more time and will be on a timeline well beyond that. It seems like it could be at same time, but confusing as I said. What are we supposed to believe?
Yes - that data just seems to get in the way and contradict those claims. We were told v13 would be great. But it wasn’t. Then v14 was going to be great. But it still isn’t.
From the crowdsourced community tracker — and while this is not an apples to apples compare as the robotaxi testing is supposed to be v15, but v14 seems to be regressing.
The two most recent releases that have data in community tracker are v14.3.5 and 14.3.6 but they don’t have enough data yet to really judge. But v14.3.4 does.
It’s not quite the same criteria as tracking March of 9s in a more formal sense. Participants in the Community Tracker report interventions, severity, and reasons.
Right now only 98.1% of FSD users reporting on 14.3.4 report the ride was free of “critical interventions”. Supposedly, march of 9s is reliability but based on occurrences of “serious accidents” rather than 'critical interventions" . Does it really matter when it is this bad?
Regarding March of 9s. I have seen it expressed as requiring as low as 5 9s and as high as 8 9s. 6 would seem adequate to me. That Karpathy guy claimed he was trying to achieve 5 9s at Tesla but actually never came close.