We need to start licensing human drivers on the same axes.
Hehe, it’d be nice if we tested them more thoroughly. We kind of do have the capability to classify them thoroughly, though. There are places (even Texas) that restrict certain drivers to daytime, maximum speed, vehicle capabilities, licensed supervisors over a certain age, transmission type, no express or highway driving, etc. I would prefer that everyone who got a license had at least some real-life training on what to do when you lose traction, such as Finland requires.
In Finland you’re almost certainly going to need that skill at some point as well as in many parts of the US. In the Southern California coast, not so much.
Considering how poorly Angelenos handle roads wet with rain, we probably do.
I dunno, you don’t get that much rain, but I’ve lost traction and had to get control of a car because a layer of sand had blown onto the road. You guys have plenty of sand.
100% accurate as there’ll be one of them. It’ll be like Where’s Waldo: which one is it?
What to Know as Self-Driving Trucks Hit Texas Highways
“We start with a single one, go to tens by the end of the year and then, over the course of the next five years, go to like, hundreds and then, eventually, thousands,” he said.
Elon just announced that “soon” every car with FSD will be able to be used as a fully autonomous taxi and software will be provided to make the payment part happen.
On the Tesla fora, even the biggest fanboys are mocking him.
Where did he say that? I see sort of the opposite vibe (from a CNBC interview):
“We are actually going to deploy not to the entire Austin region, but only the parts that are the safest. So we will geofence it. It’s not going to take intersections unless we are highly confident it will do well.”
Which is what I’ve been expecting for a while. Highly geofenced in the near term.
Same CNBC interview, he says unsupervised for normal FSD owners is ~18 months away. Not really “soon.”
They’ll also have remote operators monitoring the fleet (as expected).
Oh good. I was worried he didn’t have any idea when they’d finally solve it. But this time he’s confident and I’m sure he’s correct.
He didn’t say he was confident. IIRC, he said “best estimate” or the like. And sure, I think they’ll probably miss that deadline too. I’m just curious what was meant by “soon.”
It’s just amusing that we even bother posting timelines related to anything Musk says. Posting the results of a random number generator would add the same value.
How would we know if his timelines were bad if we didn’t report on them 
?
The near-term estimates are usually ok. The claim for what we’ll get by the end of June is a small number (about 10 at first) of driverless cars in Austin in a small, geofenced area (not necessarily their Cybercab). And then slowly increasing after a few months. That’s a pretty concrete thing we can check on in ~6 weeks.
This is not a dig at you by any means, you’re just providing the quotable quote.
I bought my car with FSD in 2018. Musk was promising unattended robotaxi features within a year or two or four or whatever. 2018 was also the year my kid started kindergarten at the school across the street. Certainly by the time they went to sixth grade at the middle school 1.5 miles away, the car could do unattended dropoff and pickup.
Even with shipping dates slipping a bit, true FSD would be ready in 7-8 years, right? I never really thought it would be, but even as I was doubtful, I hoped to be proven wrong.
Just got back from picking up my kid, and this is the last week of 6th grade. So, yeah, not robotaxi yet. It did do pretty well at self driving today. I had to manually adjust the speed for school zones, and take over at the school because FSD can’t navigate the pickup line on it’s own. Today it did decide to slow down for the drainage dip across the road; about half the time I need to hit the brakes there because FSD is going fast enough to scrape.
So 18 months. Will robotaxi mode be ready before my kid has their drivers license?
My Magic 8-ball says “No F***ing way.”
I have a rather … opinionated … 8-ball. 
I’m certainly not going to defend the FSD timelines as a whole. But I want to know what someone actually said before criticizing them on that basis.
I think in retrospect we can see that FSD could not have really come much earlier no matter how much they wanted it. They needed the new generation of AI systems (both in training and inference) to have any hope. And now that they do, they have made meaningful progress. If they can maintain the same pace as they have for the last year, 18 months might well be enough. But progress tends to come in fits and starts so it’s possible they’ll hit a bottleneck that requires another rearchitecture.
What’s the threshold? I think we’re very far from telling the car to navigate an arbitrary school pickup line. And I think it’s becoming more apparent that the initial unsupervised mode will look more like Waymo than current FSD. Meaning that it can be used on relatively limited routes in limited areas. I think that if the Austin rollout works well, it’s plausible that at least some customers in some areas will be able to use their vehicle as a robotaxi within 18 months.
But that’s not the same as an L4/L5 autonomous vehicle. Can a car that only does curbside pickup/dropoff replace the need for a driver’s license? Maybe, in some situations. Not in others.
Separately, there’s the question of when they can achieve some degree of L3 driving. That is, can I take a nap on a long highway drive? That will be the point where I really feel that I got what I paid for, not a robotaxi mode.
I’m not going to criticize someone (as much) if they provide a timeline that’s based on an actual plan, and then something unexpected comes up that throws things out of whack.
But it’s obvious that Musk’s timelines are based on nothing. There’s no project plan, no detailed breakdown of the steps with estimates for each. He just makes something up.
Do you think there is some wall in a Tesla office with a big chart showing everything needed to launch unsupervised FSD, laid out month by month, and a completion milestone in November 2026? Or did he pull that number out of his ass in the conference call?
What would such a plan even look like? Back in 2018, how could anyone come up with an accurate timeline at all? Let’s see, first wait for NVIDIA to invent the H200, then…
Maybe I just have a more relaxed attitude being in software, where almost no one can predict timelines within a factor of 3 of the actual number, even for relatively straightforward projects that don’t require fundamental advances. We’re really, really bad at it, and it actually gets worse the more detail you use. In fact, making the prediction at all itself makes the prediction worse, because people start behaving differently when the prediction is made.
I’ve never seen a Gantt chart that ever had any bearing on reality; I’d trust one less than the gut instinct of a senior engineer, although I wouldn’t trust either to less than a small integer factor. If I found someone that reliably made schedule predictions for software products to within 2x of reality I’d think them an oracle.
Or a serial book-cooker. Jack Welch’s profit predictions coming to mind.
Musk’s timelines are generally highly aspirational. He thinks of those statements as inspirational. I think of them as stock price manipulation. It’s probably both.
Yeah, that too, though that’s more relevant for financial predictions. Way too consistent profit/revenue growth is a definite red flag.
There’s just not much point to stock manipulation on Musk’s part. He only rarely sells stock. The last big sale was paying for Twitter. Before then I think the last big sale was to pay taxes on some large stock grant… in which case it would have made sense to drive the price down when the shares arrived.