Hacker News Reader

Comment by modeless | Hacker News Reader

It's not going to be in mass production until 2027 at the earliest, realistically. Will they have unsupervised FSD "next year" as claimed? I doubt it. But by 2027? I think there's a strong possibility. I've been testing FSD since it was released and lately the pace of improvement has gotten a lot faster. And the Cybercab is going to have a much faster onboard computer, and probably more/better cameras.

View on HN

It is interesting that you are confident in that matter. But you have been playing with it for a long time. Someone like myself with zero experience with it am very skeptical about how much I can trust it. We have seen several accidents over the years in the news. How do we convince the masses that this car is safe and this car will not suddenly drive off the road? I do think self driving cars are as a whole a lot safer but I also consider myself a good driver so it would be hard to give up that control. When I would be sold on the tech is when it is so good I am legally allowed to sit back and sleep between destinations. Wake up in a new city each day.

I've had FSD (as branded by Tesla) for probably five years; it's gone from "actively trying to kill you constantly" to "very, very, good" to "probably safer than my teenager (who is a good driver)". I banned my current youngest driver from using the old version, but encourage use of the latest version for night driving -- it's really excellent. In most circumstances, it's probably a better driver than me.

That said, my kid told me last night in the rain with some cars slowing, it tried to pull left into oncoming traffic, and needed a quick recovery. We seem to be at stuff like that every thousand or so miles, down from every mile five years ago. it is WAY WAY WAY ahead of any other car I've driven or ridden in that can be bought. I understand Waymo in SF is significantly better. But, compared to Rivian, Ford, Volvo, Mercedes, it's years ahead.

> and needed a quick recovery. We seem to be at stuff like that every thousand or so miles, down from every mile five years ago. it is WAY WAY WAY ahead of any other car I've driven or ridden in that can be bought.

That's orders of magnitudes better than other FSD users.

Independent testing of over 1,000 miles through Southern California required 77 interventions, at an average of once every 13 miles.

I suspect your "one intervention every one thousand miles" might be a little optimistic.

Maybe. But on what version of FSD? Recent ones are radically better than older ones. And of course roads and situations vary.

12.5.3, so :(

https://arstechnica.com/cars/2024/09/tesla-full-self-driving...

I just took it on the way home, (1 mile), and I disengaged it. But I'm not sure if I needed to; I was in a roundabout and somebody tried to wedge in from another direction; the car braked, but I decided it was safer to accelerate through. So, now I'm down to at least 1 in 500 :p

There are more than enough on the road already to know exactly how safe they are. FSD has already driven a billion miles. Once it is good enough they will have the statistics to prove it.

Tesla has, as far as I can tell looking through the reports, done not a single mile or kilometer under the Californian DMV Autonomous Vehicle Tester (AVT) Program[0]. Nor have they partaken in any other program of this kind that would enforce public reporting or any measure of transparency. There is no public data to gauge safety or reliability as of yet.

[0] https://www.dmv.ca.gov/portal/vehicle-industry-services/auto...

You are right, there is no public data, they've kept it all private. FSD is not anywhere near good enough to be unsupervised right now so there's no real point to doing autonomous testing.

I'm simply saying that when it improves to the point where it can be unsupervised, they will definitely have the data to prove it.

It's interesting, because FSD has probably driven a billion miles, but I couldn't find any useful statistics yet.

So... I think it's a bit early to start believing the hype :)

It is far from good enough today, no doubt. But the rate of improvement is what matters, and from personal experience I believe it is quite high. I do wish Tesla would release some better statistics.

Closest we have is this: https://arstechnica.com/cars/2024/09/tesla-full-self-driving...

Sadly, it's not the real statistics. Also, it's not really good.

I got to try fsd (supervised) recently, and although it wasn't perfect, it was pretty good.

I also had to learn to to enable adaptive driving (or whatever it is called) to let the car go slightly over the speed limit and go with the flow of traffic, otherwise it would only go the speed limit and people would rage-pass.

I suspect if it works well 99% of the time, which is pretty good, they're about half way to their goals. Making it work well 99.9999% is probably a lot harder.

If they were that close, why did Karpathy leave 2 years ago?

You'd have to ask him. He worked there for 5 years, maybe he just wanted a change instead of a second 5 years of sprinting to the goal? In interviews he says he is still bullish on Tesla.

There is no reason for him to burn bridges so being bullish doesn't really mean much imo. It sounds like what any socially capable person would say.

To me it feels like the traditional auto manufacturers are catching up to Tesla and now they need the next hype to stay ahead of the game. It keeps the stock price high. I am aware this is not a new goal though. I very much doubt it's within reach by 2027. I'm happy to be proven wrong though, driving a car is a bit tedious imo.

Check the video here. https://youtu.be/hM_h0UA7upI?si=Tt9HqQoxceXKHmiI&t=140 He's not just politely saying he likes Tesla, he's talking for 5 minutes in detail about exactly why he believes Tesla is ahead of Waymo and will beat them to scale.

He's forced into supporting vison-only sensors as superior for political reasons but humans can't see in the dark. I don't want a self driving car that's as good as me at night, I want one that's better.

Of course he is bullish on Tesla, probably still has some stocks or stock options. Otherwise saying that he is bullish costs nothing to him.

And "wanted a change" my *ss. If you believe the hype that autonomous driving is really just around the corner (has been since 2015) AND you are leading the R&D of the company that does this, would you want to jump ship before the product is shipped? Do you think Jobs would have left Apple in 2006, just before the Iphone announcement because "he wanted change"? If you do, I have a bridge to sell you.

Could also be a non-disparagement clause in a contract he signed.

No idea.

But, as always, I consider how hard working at Tesla is, and what percent of the total economic value of working at Tesla Karpathy had acquired at the time he left. You don't need to imagine anything other than "cool, I'm good, see you guys around" for such a decision to make sense.

We saw a bunch of execs leave shortly after Tesla presold 500k Model 3s. Super sensible -- they were vested most likely, and other industry execs could be retained for the herculean lift that getting scaled up for the M3 was going to be. Why kill yourself? And from Tesla's point of view, why overpay in the market for those guys? You can hire someone from Audi (which they did) for much less on the back of the successful pre-order.

Reply on news.ycombinator.com