FSD v12.x (end to end AI)

[AlanSubie4Life]

Aha, so this is probably what I was remembering. Isn't it legit, as long as you understand it's not referring to unsupervised operation?

No. It is subject to massive selection bias. There is no way to compare these categories. There is probably nothing wrong with the data - you just cannot compare the length of the bars - for fairly obvious reasons.

This is the data I was referring to earlier. I am not sure why they publish it since it is meaningless.

They are probably able to do a more exacting analysis in similar use cases which can actually be compared observationally - they have a lot of data and there are ways to do this - but they have not published any such data.

To be clear, I suspect FSD reduces accident rates. But I do not know and I have never seen any data to support that.

 

[FSDtester#1]

Don't worry, they save all of those emails in a massive WOM, with is techie speak for write only memory. ;-)

Thanks for the acronym explanation, others on this thread aren't so nice.

..... @JHCCAZ

See nice posters do exist.

 

[mgs333]

I seriously hope you 2 are joking.

How many people are fully paying attention behind the wheel??

It likely saved at least a serious collision.

Don't believe me, next time you're in the passenger seat watch how many drivers are
staring at their phone while driving in the city, on the highway, and at lights when the lights turn green and the driver doesn't notice and someone has to honk to get their attention.

Want a few real world video examples of inattentive drivers smashing into innocent people? Watch a few episodes of wham bam teslacam.

Any system that has the ability to eliminate or reduce the chance of a collision is better for everyone.

It might just be your bacon ADAS saves one day.

Yes, I know many people are not paying attention but I was talking about the overreaction from the car in response to a slow lane change.

 

[TK211X]

They better send us ASS on 12.4 to make up for this.

 

[TK211X]

Banish is going to be so much fun. You get to find out with ASS if your car hit and run someone.

Can't wait for a Tesla to autonomously take out someones grandma and make them a trust fund baby. Who knew when she was young, frolicking in the 1960s grass, that when she turns senile a giant triangle driven by nobody is going to delete her.

 

[FSDtester#1]

Yes, I know many people are not paying attention but I was talking about the overreaction from the car in response to a slow lane change.

Thanks for clarification, I make a fool of myself here with poor humor.
Safety is the only thing I won't take for granted.

 

[mgs333]

Did anyone touch base on what exactly 2024.8.7 is?

All my friends that are non subscribers just got it.

2024.8.4 Official Tesla Release Notes - Software Updates

Tesla software update 2024.8.4 includes Vehicle Alarm Includes Trailer, One-Time Charge Limit, View or Reset Tire Mileage, Rear Passenger Headphones, Auto Wipers, Danger Zones on Your Route, Adaptive High Beams, Speed Cameras on Your Route, Security Improvements, Windshield Washer, Fold Mirrors...

www.notateslaapp.com

2024.8.7 Official Tesla Release Notes - Software Updates

Tesla software update 2024.8.7 includes Vehicle Alarm Includes Trailer, One-Time Charge Limit, View or Reset Tire Mileage, Rear Passenger Headphones, Auto Wipers, Danger Zones on Your Route, Adaptive High Beams, Speed Cameras on Your Route, Security Improvements, Windshield Washer, Fold Mirrors...

www.notateslaapp.com

Looks like the notes are virtually (if not exactly) identical, seems like security and bug fixes.

I think they've got to be working on 2024.8.7.1 or such that has FSD v12.3.2.1.0 on it as they seem to be moving most non-FSD cars to 8.7.

 

[FSDtester#1]

Banish is going to be so much fun. You get to find out with ASS if your car hit and run someone.

Can't wait for a Tesla to autonomously take out someones grandma and make them a trust fund baby. Who knew when she was young, frolicking in the 1960s grass, that when she turns senile a giant triangle driven by nobody is going to delete her.

Is this cryptic engineer talk?

Where is @JHCCAZ for an explanation?

 

[TK211X]

Is this cryptic engineer talk?

 

[OxBrew]

But, what if the data is released by Tesla? This links to data through Q4 2023.

Tesla Vehicle Safety Report | Tesla

At Tesla, we believe that technology can help improve safety. That’s why Tesla vehicles are engineered to be the safest cars in the world. We believe the unique combination of passive safety, active safety, and automated driver assistance is crucial for keeping not just Tesla drivers and...

www.tesla.com

LOL no. As has been hashed out ad nauseam all over TMC, they don't break out AP into TACC, Lane Keeping, EAP, FSD, or FSDb.

They do not account for the probable bias for safer conditions in general being over-sampled (people disengage when it gets unsafe, and use it on the highway more, which also could generally be safer).

This is what I meant by TRANSPARENT data sharing. That tells us almost nothing. (and the statement by Tesla is not actually data, just PR fluff about some purported data they have. Show me the actual data, with all the original meta data intact, including location, road type, speed, everything.).

Looking at what Tesla said, the entire increase in safety on "AP" could be 100% down to people using it ONLY for the safe easy parts of driving.

I challenge any one here to prove that last statement wrong, using actual data.

 

[JHCCAZ]

Is this cryptic engineer talk?

Where is @JHCCAZ for an explanation?

I think that was more like a playback of a Psychiatry session. Combining ancestor resentment with Cybertruck envy and get-rich-quick daydreaming.

Maybe when @TK211X finally gets his update, there needs to be a series of on-screen ink blots so that Grok can decide if it's safe to turn over control. Sort of a Freudian breathalyzer routine.

 

[JohnB007]

I did another 40 mins drive, and still not much of a surprise for me. It has remained competent in relatively complex intersections.

Turning off auto-Speed worked for me. Now I can go back to controlling Max Speed, except in this case, I'm telling Auto-Speed to not go over my Max speed. It worked well but will take some time to get used to it.

So I now use accel tap and scroll to control how the car behaves if its default speed isn't my liking. I think it will be very useful when I visit new towns as I already make all kind of dumb mistakes when I drive manually.

Tomorrow will be a big day - totaling more than 400km drive to visit a small USA town (a tiny one called Watertown in upstate NY). I plan to use v12.3.1 extensively. Wish me luck!

 

[Pdubs]

Have you tried Hydrocodone

I wanted to put more than on reaction to this. I went with the crowd and did the laughing guy. You're really winning lately.

 

[JHCCAZ]

On v11, my top frustration was that it would fight against my attempts to get into, and stay in, the correct lane ahead of a turn. Not everywhere, but very repeatable incorrect behavior in certain places.

v12 has almost entirely fixed this problem, fantastic! It doesn't fight me anymore once I'm in the correct lane. However, it still seems to be too relaxed about initiating the needed line change on its own. And it doesn't always cooperate with my turn-signal stalk command to perform a requested lane change.

v11 would promptly perform a requested lane change as long as it was safe to do so - even if it subsequently tried to undo that change. v12.3 often refuses even when everything is clear. And if the change does happen, there's typically a little lateral hiccup first.

BTW I also cannot say that Minimal Lane Changes has any particular effect, based on daily drives where sometimes I set MLC and sometimes not. This is all in-town suburban type driving, no freeways.

I'm definitely in the camp that v12 is a big step forward, but I haven't figured out the logic of the automatic nor semi-automatic lane changes.

 

[primedive]

Haha, archaic maybe compared to an LLM that can just write software to do whatever I want with zero effort on my part, but C++ is still AFAIK the first choice for high-performance, low-latency, low-power applications. Just because it's fragile and high maintenance doesn't make C++ archaic when there are no better tools for the many jobs C++ is still the best for...

Ok ok, but what about Rust? C++ is probably why the old code kept misbehaving…
In my own domain, I actually find C (or C++ with a C ABI) to be the most useful as you can drop it in almost anywhere and usually without having to drag in a build system for the ride.

 

[willow_hiller]

they don't break out AP into TACC, Lane Keeping, EAP, FSD, or FSDb.

They do not account for the probable bias for safer conditions in general being over-sampled (people disengage when it gets unsafe, and use it on the highway more, which also could generally be safer).

Tesla once did break out FSD Beta. At the time, it was 0.31 accidents per million miles, compared to 0.18 accidents per million miles on Autopilot, and 0.68 accidents per million miles for Teslas without active safety features.

You cannot compare it to the US fleet average data as Tesla tries (1.53 accidents per million miles), but we can absolutely say that Tesla drivers with FSD Beta and Autopilot are both safer than Tesla drivers without Autopilot.

 

[AlanSubie4Life]

but we can absolutely say that Tesla drivers with FSD Beta are safer than Tesla drivers without Autopilot.

How do you figure? It’s not comparable.

Describe exactly what it is that you are comparing.

 

[aronth5]

Awesome. Looking forward to it.



Awesome!

Looking forward to seeing everyone’s issues resolved.



Not too surprising if this ends up being the case. Elon did say 12.3.1 would address it, after all.

Sometimes I wonder if there is something wrong with your car since your experience seems so different than many others. The fixation on stopping smoothness doesn't help either. I'd rank that near the bottom of my enhancement requests.

 

[willow_hiller]

How do you figure? It’s not comparable.

Describe exactly what it is that you are comparing.

I'm not saying FSD Beta is the cause of the increased safety. I'm making an observation of the available data. There does not need to be comparability.

"The miles driven by Teslas on FSD Beta have resulted in fewer accidents per million than the miles driven by Teslas without active safety features." That's a fact, given the data.

You can argue that you think the data is fabricated. You can argue that the additional safety is caused by other factors that are spuriously correlated with FSD Beta. But you cannot say that the miles driven by FSD Beta have resulted in more accidents per mile than Teslas without active safety features.

 

[AlanSubie4Life]

Sometimes I wonder if there is something wrong with your car since your experience seems so different than many others. The fixation on stopping smoothness doesn't help either. I'd rank that near the bottom of my enhancement requests.

My car is fine. It has modern extremely capable hardware, capable of full autonomy: HW3.

Drive on a road at 50-60mph with traffic lights every half mile or so and report back.

I am curious if HW4 does better, actually, since I believe part of this could potentially be usable perception range. But I do think it is more likely a software limit, not perceptoon.