The funny thing is that I've owned Gen 3 FSD for five years now and only in the last few months has it gotten noticeably better. It's actually respectable at driving around my city now, can handle roundabouts, make turns, etc.

I have noticed the same - Tesla FSD has improved remarkably in the last couple of versions, to the point where it almost never requires interventions.

That said, I still don't "trust" FSD to the point where I could comfortably be in the backseat taking a nap just yet. 99% is not "good enough" for something as critical as driving.

I actually started to trust V12. V11 was a complete joke vaporware. V12 is a complete game changer that actually drives on par if not better than human. Not a tesla stock holder(all of it is a scam per my retarded opinion), but I am now more than confident that they will start doing self driving this year. Maybe isolated and with some remote interventions here and there, but it feels on par with Waymo. My friend was just testing Waymo and by his words it is not as aggressive (human like) like Tesla on the road where you need to assert yourself to merge.

I’ve been reading this comment for years, just with version numbers changed. I expect in two years I’ll be hearing that V14 sucked, but now with v15, actual level 4 FSD is just around the bend!

Probably, but what they are saying is true. It recently went from uncomfortable verging on unsafe, to moderately comfortable.

> That said, I still don't "trust" FSD to the point where I could comfortably be in the backseat taking a nap just yet. 99% is not "good enough" for something as critical as driving.

You should only trust it after Tesla accepts full liability when FSD is enabled.

The funny thing is that a nearly perfect template of your comment appears on nearly every post which includes a criticism of FSD. It's always magically "just recently got better!" As well as "being able to do some selected basic driving despite being called FULL self drive."

Maybe because it is true?

If it's true now then it means it was false before. Unless the claim is that every few months there's a noticeable spike in capabilities however the grandparent commenter seems to claim otherwise (ie, that only recently did it improve).

> Maybe because it is true?

It reeks of manual fitting. Tesla surveys an area and makes manual adjustments that keep the system from misbehaving for a while until something changes and then it’s back to shit.

There is definitely some amount of that. On each training run it gets better at some things, worse at others (this is mostly limited to its lane handling more recently).

And in the 12 months I've had the car, it has gotten noticeably better maybe 2 times. Once last year, and then once in February. After the iteration last year, it drove fairly well, but I still had to intervene once or twice per trip. Now I make it through half of my trips without intervening, and when I do intervene it tends to be to avoid embarrassment vs safety (not committing, wrong lane for an upcoming turn, etc).

Given how ai is advancing, I am pretty confident they will get this working fully autonomously with just vision, it's just a question of when.

Well that’s what people pay for when they buy a Tesla. The holistic experience of self driving. Somebody who pays for self driving alone pays Waymos.

I have no 'recently' to compare to, as I'd never been in a Tesla before. I was dubious because of all the "this year for sure" history, but after test driving one, I bought the new Model Y, and it now pretty much drives me to work and back every day with little to no intervention.

Knowing how prone to exaggeration Elon is, my bar was low. But it blew me away honestly. After nearly 30 years working in software and with some background in machine learning and computer vision and generally just trying to make software that works reliably, it's a pretty jaw dropping experience.

Would I take a nap in the back seat and let it drive? No. But does it allow me to sit there focused on a technical podcast or an audiobook so I feel like I'm getting back an hour or two a day instead of worrying about driving? Absolutely.

Look, everyone on the planet agrees that calling it FSD was over-promising. You are not adding anything to the conversation by pointing that out, or by confusing technology with "magic". The fact remains that software developed by a competent team can actually improve over time, despite the insanity of corporate leadership, and that's what is happening here.

The fact also remains that fraud (even if it's not prosecuted) is fraud. You promise X and deliver... not X, for years? You're a fraud. Even if you are getting better at delivering Y, for some value of Y which is Far, Far less than X.

> It's actually respectable at driving around my city now, can handle roundabouts, make turns, etc.

This year for sure, guys, for real this time!

-- Elon

lol the bar is so low

The funny thing is that just this weekend FSD was in the news for a crash that put a young man in a coma.

But when the usual outcome is death, i suppose a coma is an improvement.

As someone who also purchased FSD, I was able to drive hundreds of kilometers without any disengagements even though my car is an older HW3 with 12.6. I fully believe that with some Waymo-like conservative route planning and teleoperator backup, they will be able to achieve their robotaxi goals this year.

Tesla is going nowhere without LiDAR.

I strongly disagree. Go look at the videos of the current iteration of FSD.

Yes, then read this article: https://www.jalopnik.com/teslas-fsd-is-biased-towards-influe...

Go and actually test drive one and let it FSD you around. Videos don’t do justice. There is always MKBHD kind of biased videos.

O3 can now play geoguessr better than master level human players. It can also beat master level Codeforces competitive programmers. I wouldn't discount the ability of AI to make sense of images far better than humans possibly could, all the while beating them at logical thinking, especially in a restricted domain like driving.

AI isn’t magic. If there isn’t enough information in the inputs, you can’t expect reliable results. It’s the same principle in all of software: garbage in, garbage out.

If there simply isn’t enough visual information, vision-only will fail.

https://youtu.be/IQJL3htsDyQ

Theoretically, if there's not enough visual information for AI drivers, then there's not enough visual information for human drivers, and that's a problem with the road. (Which, to be sure, occasionally there are roads like this: e.g. merging onto a higher-speed thoroughfare from a lower level, with a very short distance between "where you're in a position to see the merging traffic (and not that much of it)" and "where the roads have fully merged (and there's no shoulder)".)

This Mark Rober video has been debunked several times:

https://www.youtube.com/watch?v=TzZhIsGFL6g

https://www.youtube.com/watch?v=7cxTO8g47_k

That is not a debunking... That's someone running a similar experiment and getting a different result. That would debunk the claim that Teslas can never detect a painted wall. It does not debunk the claim that Teslas will sometimes fail to detect a painted wall.

And in a safety-critical system, the distinction is not mere pedantry.

I mean he didn't even use FSD.

[delayed]

headlights can make cameras see in the dark

Left and right of the car? What about behind it?

Waymo has 360-degree LIDAR and, not coincidentally, is the state of the art.

It also has the help of being able to do web searches and when o3 fails, it doesn’t kill people.

Spinning lidar is not a panacea.

Sadly the US is quite corrupt, so I wouldn't bet against the car company whose CEO has a close personal relationship with the US president.

Unlike China, right?

Toyota is Chinese now?

I'm just saying that out of two American companies, Waymo and Tesla, Tesla is more likely to benefit from government favors.

I still can't believe they were taking one-time purchases for FSD like... 6 years ago.