FSD v12.x (end to end AI)

[rlsd]

Sentry mode in the dark: I just looked at videos of my car in the garage at night. The cameras showed the darkness around the car. However the were some short frames showing clear images of every object as if there was light. How could it be? Does sentry mode incorrectly send images that were captured before nightfall?

 

[JHCCAZ]

Sentry mode in the dark: I just looked at videos of my car in the garage at night. The cameras showed the darkness around the car. However the were some short frames showing clear images of every object as if there was light. How could it be? Does sentry mode incorrectly send images that were captured before nightfall?

I believe you are seeing the results of headlight flashing, possibly also reverse lights, upon some trigger. I'm not actually sure all these events are true Sentry incidents recorded as clips.

I do know that after dark, when I activate the Live Camera feature in the app with the car sitting in the driveway, I often see random bright flashing from the headlights and the corresponding lit-up video display in the app. Not always, and not in any really detectable pattern.

If I'm just standing outside, away from the car and not using the remote Live Camera view, I don't see these occurrences (unless there's an obvious cause from a person or large animal approaching the car).

My guess, and that's all it is, is that the motion detection sensitivity trigger is set to a very low threshold (i.e. very sensitive) while the Live Camera view is active. I think it returns to normal (i.e. less sensitive) when you stop using the Live Camera remote viewing feature.

Why would it do this? Again a guess: it may be that the onset of a Sentry event requires significant motion, but then for a period of time the sensitivity is increased to catch related follow-on activity more reliably. If this is true, then it's quite possible that the action of the user initiating a Live Camera session in the app, even though it wasn't initially triggered by detected motion around the car, then follows the same routine - prone to these apparently false triggerings and flashes.

I actually wish that Tesla would allow us to intentionally illuminate the headlights and reverse lights, as an optional feature while viewing, because the nighttime view can be pretty dark and useless unless you happen to get one of these random flashes. The illumination could probably be somewhat dimmed-down but steady, and would still be very useful for seeing what is going on at night. The cameras have decent night vision as long as they have some light to work with. The illumination should remain an option, as you might sometimes want to activate the camera observation without alerting bad people or bothering good people who happen to be around the car. As it is now, it's prone to those unpredictable, non-useful and annoying flashing events when you pull up the Live Camera view.

 

[sleepydoc]

True, except for how really no instructions needed; it happens automatically.

We picked up my MY from getting the alignment done and my wife drove behind me on the way home. I had FSD on for part of the trip and drove my self for the rest. After getting home I asked my wife if she could tell when I was driving and when FSD was. She couldn’t.

 

[zoomer0056]

There are many times the car is still in the far left lane of the highway when the exit is just less than a quarter mile. With only 300 feet or so to go. it tries to shift with a lot of difficulty due to heavy traffic and misses the exit completely and takes the next exit. I don't have the patience for such idiocy, and disengage FSD, shift lanes and then turn on FSD.

I don't have FSD, only tried free FSDS for a month. But I am curious here. When FSD messes up like this, is there risk of an accident? Does FSD act impatient, get increasingly aggressive and try to make the exit? Or remain safe and patient, miss the exit and reroute?

 

[sleepydoc]

It makes no sense and infuriates other drivers (I have never tested this).

Wait - you have never tested it? How do you know?

How can you possibly know that if you haven't tested it.

I am just not even following at this point. I don’t need to test something if I know what will happen.

Just to lay it out, you have clearly stated that you are drawing conclusions from your own assumptions. The fact that you can’t follow this line of reasoning is why people say you have no credibility.

As you know, I have been a consistent contributor here, offering factual assessments of FSD for those who may not have it - for years now. Credibility levels quite high, even if not everyone agrees with the assessments.

Nothing of late has altered that, in spite of people’s incoherent recent efforts to claim that…based on…something?

I will continue to contribute here as I see fit, with the continued high credibility, when 12.4 and 12.5 arrive.

As always, I will admit when I am wrong. It’s one of my favorite things to do. I’m wrong all the time! So maybe 12.4 and 12.5 will surprise me. I’ve previously offered my 12.4 estimates. We’ll see! Hopefully everyone has them bookmarked.

Everyone should make more use of bookmarks and hold each other accountable. Remember that some here believe Elon went forward with the 8/8 Robotaxi date based on the strength of what he had seen from 12.4 and 12.5! Will that assessment be correct? Let’s see!

I hope I am WAY off base and 12.4 is the huge step forward that people predicted, due to the rapid retraining of the NN enabled by AI with more training clips, or whatever the reason is.

You have been a constant contributor but of late your posts have been getting progressively more negative and your reasoning more distorted. I posted an example above. It’s almost like you’ve been ‘radicalized’ by an anti FSD cult. I’m not saying this to troll you or pile on, it’s just an observation I’ve made.

 

[sleepydoc]

Just to add onto Alan's reply, the braking behavior is terrible in my car as well. It'll slow down too early too abruptly, then coasts painfully slowly all the while the braking pulses stronger/weaker/stronger/weaker. It's demonstrated very well here:

https://twitter.com/x/status/1785099119306158519

I don’t have an accelerometer to test and graph stops but that does not match my experience. Maybe I’m just more tolerant of small variations in deceleration, maybe my car performs better or maybe it’s a regional difference. I have no way of knowing; I can just post my subjective experience.

 

[MP3Mike]

Credibility levels quite high,

with the continued high credibility

You don't get to decide your level of credibility... That is up to each individual reading your posts.

 

[zoomer0056]

You don't get to decide your level of credibility... That is up to each individual reading your posts.

Definitely not meek or humble. There's value in that too. However, it's nice to be able to skim or even skip at will. By now that's achieved with just a look from the wife. Sorry, I piled on here just for grins.

 

[kabin]

Intervention is where you interact with the system ( pressing brakes or accelerator or turning the wheel) for a situation unrelated to safety . Thinks like, ‘this is not the correct exit, the car is waiting for too long at this Xn’.

Disengagement is where you take control because you think you will end up in an accident if you let FSD continue to drive.

Anything you do while driving is related to safety. Intervening when FSD accelerates too fast, tailgates, excessively decelerates, phantom brakes, picks the wrong lane, crosses over the double yellow line on a left turn is related to safety.

FSD supervisors are suppose to be that last piece of swiss cheese without an aligned hole.

 

[kabin]

As you know, I have been a consistent contributor here, offering factual assessments of FSD for those who may not have it - for years now. Credibility levels quite high, even if not everyone agrees with the assessments.

Nothing of late has altered that, in spite of people’s incoherent recent efforts to claim that…based on…something?

I will continue to contribute here as I see fit, with the continued high credibility, when 12.4 and 12.5 arrive.

As always, I will admit when I am wrong. It’s one of my favorite things to do. I’m wrong all the time! So maybe 12.4 and 12.5 will surprise me. I’ve previously offered my 12.4 estimates. We’ll see! Hopefully everyone has them bookmarked.

Everyone should make more use of bookmarks and hold each other accountable. Remember that some here believe Elon went forward with the 8/8 Robotaxi date based on the strength of what he had seen from 12.4 and 12.5! Will that assessment be correct? Let’s see!

I hope I am WAY off base and 12.4 is the huge step forward that people predicted, due to the rapid retraining of the NN enabled by AI with more training clips, or whatever the reason is.

No worries. The truth eventually arrives - it's just taking a bit longer with v12.4. I'm sure it will be equally mind blowing.

 

[sleepydoc]

I just had my first safety disengagement in a while. On the interstate FSD wanted to move to the left lane to get around some slower traffic. It signaled and started to merge even though there was a car right off my rear bumper. It was even highlighted red on the display.

I also disengaged because it chose the wrong turn lane at a tricky intersection. The Nav system had the correct lane highlighted, FSD just didn’t follow its advice.

I just drove 41 miles with 3 stops on all types of roads including stop and go traffic and construction on the interstate. With the exception of the above it was perfect and incredibly relaxing.

 

[AlanSubie4Life]

You don't get to decide your level of credibility... That is up to each individual reading your posts.

Agreed. We’ll see how it turns out.

 

[skytool]

From Elon's post in X a minute ago, "FSD 12.4.1 release today to Tesla employees. If that goes well, then it will be released to a limited number of external customers this weekend. There are massive number of chagnes to this build....."

 

[uscbucsfan]

From Elon's post in X a minute ago, "FSD 12.4.1 release today to Tesla employees. If that goes well, then it will be released to a limited number of external customers this weekend. There are massive number of chagnes to this build....."

https://twitter.com/x/status/1798374945644277841

His, "it will take 1 year of driving to get a disengagement" is the type of comment his haters love, because it's incredibly implausible.

 

[JulienW]

Please chime in Gottagofast now that 12.4.1 is releasing. So when have YOU approved 12.4.1 to start downloading to our cars?

Elon "arguably" just can't resist saying the next version "arguably should be called" 69.420.

 

[JulienW]

What does Elon's Tweet say? We both must be high because this can't be what he said and I MUST be reading incorrectly.

".....Two other versions are in earlier stages of testing: 12.5 and 12.6, which could be called v14 and v15. We are starting to get to the point where, once known bugs are fixed, it will take over a year of driving to get even one intervention."

Say WHAT???????

That is bound to disappoint Alan unless he drives ½ mile a year.

 

[maddogs]

What does Elon's Tweet say? We both must be high because this can't be what he said and I MUST be reading incorrectly.

".....Two other versions are in earlier stages of testing: 12.5 and 12.6, which could be called v14 and v15. We are starting to get to the point where, once known bugs are fixed, it will take over a year of driving to get even one intervention."

Say WHAT???????

That is bound to disappoint Alan unless he drives ½ mile a year.

He didn't say how much the person drives in that year.

 

[OldManCan]

He didn't say how much the person drives in that year.

He also didn't say how many "known bugs" need to be fixed and how long he thinks it will take to get there...

 

[gottagofast]

so it’s really version 13, that’s why it was delayed

 

[AlanSubie4Life]

The fact that you can’t follow this line of reasoning is why people say you have no credibility.

Let me know when you can explain to me the need for testing.

because it's incredibly implausible.

It’s somewhat plausible “once known bugs are fixed.” That phrase is doing a lot of work.