Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register
This site may earn commission on affiliate links.
Large deployment of 2024.14.3 going out now.

TeslaFi.com Firmware Tracker
Looks like the EU, mainly, and they don't, AFAIK, have any FSDS 12.x branches available. The loads before the 2024.14.3 have 11.4.9 baked in.

Looks like better eye candy on the screen but no current FSDS 12.x. We'll see when TeslaFi catches up.
 
It would have driven 38mph most of the time and there would be nothing you could do.

Another win for manual mode.
I use auto max speed control all the time. Does it always drive the same speed that I would on city/streets? Nope, but other drivers don't always drive the speed I do either. Biggest problem is when FSD drives too slow but that happens when there are no other cars around me to annoy. So I just let it be. Once there is another car in front or behind me FSD adjusts the speed pretty well to match theirs. I've noticed several times a car gaining on me but once the car gets closer FSD goes a little faster and the car never gets too close. Does auto max control need to be improved? Of course, but it's ok where I drive.

Only time I change speed is on the highway.
 
  • Like
Reactions: JB47394 and GWord
That road appears to be nothing but blind turns. How long are you willing to sit behind a cyclist before attempting to pass them?

In the end, if FSD is capable of returning to the correct lane shortly after spotting on-coming traffic, then there's no issue. But we don't know if it would have returned because he disengaged.

It's nothing but click-bait for him to claim he prevented a head-on collision.
The reason for double yellow solid lines is because passing is unsafe. So I will sit behind the cyclist until they get off the road and stop so I can pass without getting out of my lane, or until either they or I get to our respective destination. Assuming we don't get killed by some fool crossing the double solid yellow line before then.
 
So I have to wonder - why don't they just put a 'speed limit 55' sign instead of 'end of 25' sign? The cost is the same but one option is completely clear while the other is very ambiguous and requires you know what the previous limit was.
Sometimes you need to know both. A few years ago my family was driving through a small town, and we stopped for a break most of the way through town, parked on the shoulder. (I had slept on the way in.) On the way out, it was my turn to drive. It was wide-open highway, and I saw a 55mph sign several hundred feet ahead, so I accelerated toward it, but was immediately flashed and pulled over by the local sheriff, who ticketed me for doing 37 in a 25.

I wrote to the county judge pleading my case; that there was no indication from my point of view that the speed limit was anything other than 55. He agreed to dismiss the points on my driving record, but still made me pay the fine. In this situation, an "End 25" sign would certainly have helped!
 
  • Informative
Reactions: primedive
Another drive with v12.3.6. My car was waiting to make a left turn from a suicide lane onto a residential side-street, and there was a short window with no cars, but a jogger was about to cross the street. The car abruptly went for it, and I had to slam the brakes. (Had it continued, and presumably braked just in time to avoid the oblivious jogger, it would have rudely and dangerously blocked fast-moving oncoming traffic.) Then, half a block later, a phantom forward collision warning when nothing was in front of me at all, going perfectly straight on a straight and empty suburban street. L4 consistency feels a very long way off!
 
Last edited:
  • Informative
Reactions: primedive
Found it by accident. Not sure they are documenting all capabilities anymore.

On 12.3.6 nothing mentioned the autopark improvement in showing all the spots. It wasn't until I pulled in to a parking lot that I saw they now displayed. Didn't see that on 12.3.4 or see it mentioned in the release notes. I may have missed it in the release notes.
I didn't notice any Autopark improvement between 12.3.4 and 12.3.6. You mean it shows more parking spots? Is this confirmed?
 
Last edited:
I use auto max speed control all the time. Does it always drive the same speed that I would on city/streets? Nope, but other drivers don't always drive the speed I do either. Biggest problem is when FSD drives too slow but that happens when there are no other cars around me to annoy. So I just let it be. Once there is another car in front or behind me FSD adjusts the speed pretty well to match theirs. I've noticed several times a car gaining on me but once the car gets closer FSD goes a little faster and the car never gets too close. Does auto max control need to be improved? Of course, but it's ok where I drive.

Only time I change speed is on the highway.
I enabled auto speed today in a 35 zone. I turned it off when my speed reached 48, and still slowly accelerating. It seems to want to go as fast as the fastest car on the road, not the average.
 
  • Informative
Reactions: primedive
Of course the driver had plenty of time to intervene before. And cyclists always have the right to play out their inner Darwin on tight roads in these days of distracted drivers.

Otherwise it's an unsafe move for FSD to attempt but we've seen it before with blind curves.

I might have used the same language as the driver. :)

I'm disappointed by how poorly FSD seems to understand visibility and obscuration. Is this a completely unsolved problem in computer science? Previous versions of FSD have never been able to reliably determine whether the car has the requisite amount of visibility in order to make a decision such as crossing at a stop sign. v11 would rarely creep out far enough before "going for it". The situation improved after moving to E2E NN, but not because it understands visibility. Moreso because it is mimicing the positioning used by other drivers, giving the illusion that it knows it has a clear line of sight.
 
I'm disappointed by how poorly FSD seems to understand visibility and obscuration. Is this a completely unsolved problem in computer science? Previous versions of FSD have never been able to reliably determine whether the car has the requisite amount of visibility in order to make a decision such as crossing at a stop sign. v11 would rarely creep out far enough before "going for it". The situation improved after moving to E2E NN, but not because it understands visibility. Moreso because it is mimicing the positioning used by other drivers, giving the illusion that it knows it has a clear line of sight.
Any sufficiently advanced mimicry is indistinguishable from intelligence.

E2E is a black box all the way through; it may be nonsensical to think of it as "knowing why" it's doing something, or to have any way of finding out in any simplifiable or explainable way. This is the double-edged sword of using deep neural networks. But as you observed, it is getting better at it, even if there's no way to peer into its "mind" to know why. There may be a neuron in there whose signals correlate very strongly with "visibility". (Likely there is.) Or maybe not. But... If you can't tell, does it matter?
 
Any sufficiently advanced mimicry is indistinguishable from intelligence.

E2E is a black box all the way through; it may be nonsensical to think of it as "knowing why" it's doing something, or to have any way of finding out in any simplifiable or explainable way. This is the double-edged sword of using deep neural networks. But as you observed, it is getting better at it, even if there's no way to peer into its "mind" to know why. There may be a neuron in there whose signals correlate very strongly with "visibility". (Likely there is.) Or maybe not. But... If you can't tell, does it matter?
I might agree that humans are after all just glorified pattern matchers.

Off topic, but I wonder if Tesla has cut down the number of software engineers on staff. I figure once the NN code is laid down, the rest of development boils down to data curation. I imagine "programming" via data instead of coded logic is unfulfilling for an engineer.
 
I use auto max speed control all the time. Does it always drive the same speed that I would on city/streets? Nope, but other drivers don't always drive the speed I do either. Biggest problem is when FSD drives too slow but that happens when there are no other cars around me to annoy. So I just let it be. Once there is another car in front or behind me FSD adjusts the speed pretty well to match theirs. I've noticed several times a car gaining on me but once the car gets closer FSD goes a little faster and the car never gets too close. Does auto max control need to be improved? Of course, but it's ok where I drive.

Only time I change speed is on the highway.
Yes. What you describe is exactly the same as manual mode, of course.

Any area where the issues above arises and you won’t be able to use auto mode. Because it won’t exceed 38mph.

In general they are the same though and behave as you describe. But not this case.

I enabled auto speed today in a 35 zone. I turned it off when my speed reached 48, and still slowly accelerating. It seems to want to go as fast as the fastest car on the road, not the average.
The max speed is 53mph.
 
Last edited:
I might agree that humans are after all just glorified pattern matchers.

Off topic, but I wonder if Tesla has cut down the number of software engineers on staff. I figure once the NN code is laid down, the rest of development boils down to data curation. I imagine "programming" via data instead of coded logic is unfulfilling for an engineer.
There's plenty of good old-fashioned engineering still to be done: to figure out how to embed the network into Dojo and train it there at scale, how to iterate the architecture of the network to see which way works best, and of course to find ways to translate what the network "sees" into the on-screen visualizations. (I'm not sure whether this data is somehow extracted from the primary E2E network, or is separately computed and visualized by an independent network.) And all that other non-FSD stuff that the Tesla touchscreen does. There's also the synthetic data generation system, the backend to handle ever-increasing volumes of training data from the fleet, and the systems to analyze, categorize, and curate it. There's also the need to simulate and virtually test possible future hardware and sensor suites. There is no lack of work to do. But you're right, those 300,000 lines of C++ control code they stripped out from v11 -> v12 are not coming back.

Once FSD reliability reaches a steady-state of near-perfection, future work may boil down to just data curation and retraining; e.g. to handle new car form factors, and by then our general-purpose AGI systems may handle all of that monotonous work for us. But we are probably a couple decades from that point still.
 
  • Like
Reactions: cyborgLIS
So I have to wonder - why don't they just put a 'speed limit 55' sign instead of 'end of 25' sign? The cost is the same but one option is completely clear while the other is very ambiguous and requires you know what the previous limit was.
Speed limit 55 might be interpreted by some as an endorsement of that as a safe speed. So maybe they don't really want people to speed up to 55 on such a twisty road. Or, maybe, back when they were making the specs for these signs, there was talk about changing the default speed, which would require replacing the signs. Who knows. I expect the old, wise hands who decided on this solution are long gone, or else we could ask them.

But as it stands, it is one of those countless cases where the roads and signs, as they are now, work "ok" for us humans. (As if ~50,000 deaths per year on road in th US is "ok".) Teaching FSD to deal with this, hard as it seems to be, is more likely than changing all the quirky laws and quirky signs and quirky intersections in the world to accommodate AI. In the long term, I expect it'll be a process of compromises. (If only the real world was a simple as Elon imagines it to be. Landing a rocket pointy end up is now demonstrably far easier than understanding all the weirdnesses that highway engineers have come up with.)
 
  • Like
Reactions: cyborgLIS
The reason for double yellow solid lines is because passing is unsafe. So I will sit behind the cyclist until they get off the road and stop so I can pass without getting out of my lane, or until either they or I get to our respective destination. Assuming we don't get killed by some fool crossing the double solid yellow line before then.

As long as it takes to reach a passing zone. Crossing a double-yellow is illegal and a stupid move.

We've had this exact same discussion in this thread before. It is not always illegal to cross a double yellow. Passing a bicycle is a perfectly legal reason to do so.

In fact, in my state, the law is you must give at least 3 feet of space to a cyclist while passing. And there's an explicit carve-out in the double-yellow law to allow you to cross it to give cyclists safe space.
 
Yes. What you describe is exactly the same as manual mode, of course.

Any area where the issues above arises and you won’t be able to use auto mode. Because it won’t exceed 38mph.

In general they are the same though and behave as you describe. But not this case.

The max speed is 53mph.
Would you explain your response further. What does 38mph and 53mph have to do with anything?
 
So I have to wonder - why don't they just put a 'speed limit 55' sign instead of 'end of 25' sign? The cost is the same but one option is completely clear while the other is very ambiguous and requires you know what the previous limit was.
Here's one simple reason: suppose they change the default speed limit for that road type? Now they don't have to change those signs.