FSD v12.x (end to end AI)

[AlanSubie4Life]

But it's increasingly diminishing returns for values of "better" as you drop down.

OT, my last post on this: This is not quite true in the region of 55%-65% for NCA packs (it is a bit higher for NMC/LFP packs - curves are different but they show a similar peak in “returns”). This is pretty well established now and empirical data indicates this is how Teslas behave with calendar aging (dominant contributor to aging for typical use). Except in extreme heat, there is not a huge benefit to lower charge levels until you keep the car below around 55% for NCA. It’s not smoothly diminishing returns, at all. Note that charging to 100% as needed is no problem at all; it has minimal impact. Just don’t keep it there. And keep cycles as shallow as possible (not as important, and prioritize keeping below the 55%). Again: storage charge level (time-averaged SOC) is most important.

For further discussion can move to one of many threads on the topic (this is not the best example thread; search for @AAKEE posts for more options):
Does leaving a battery at low state of charge cause degradation or just inconvenience?

 

[Daniel in SD]

Is there a tweet or something of Elon saying the goal for v12 is unsupervised operation?

The OEDR hasn’t changed, all the limitations specified in these 2020 letters still exist four years later. FSD is better at doing turns and navigating roundabouts, but those were never part of the limitations — those are customer experience improvements.

There's always a tweet:

https://twitter.com/x/status/1727484899374899687

It's funny to me that people think Elon is telling his engineers to make driver assist software when it's so obvious they're trying to make a robotaxi.

 

[zoomer0056]

Both of my cars insist on 80%

Things have changed since the 90%. Tesla has read all the posts here on TMC from @AAKEE and have learned a thing or two. They have since changer their recommendation to 80%

 

[sleepydoc]

Are these interventions for coasting at a higher speed vs starting sooner (from 0 or creeping) vs keeping motion (instead of allowing 12.x to slow too much) vs other? I tap on the accelerator for all 3 of those but most common would be the coasting where at least 12.3 then maintains the higher speed for at least a little bit.

Sometimes I'm also surprised that it decides by itself to go 40+ in a 30mph such as over bridges, but that's how nearby vehicles are driving too, so then I get a bit concerned about the potential speed trap getting off the bridge. Although at least testing so far, 12.3 also seems to realize it needs to slow back down closer to 30mph.

All (or almost all) of them have been to increase the speed because it's driving 10 MPH below the posted speed.

 

[zoomer0056]

But why are we taking about batteries?

That and "why doesn't my car wakeup". You should see the investment thread. It is like the hive ... thousands of voices speaking about everything. It's truly schizophrenic.

In this thread there's many pages of "levels", definition of "drive" and so forth. It still pales to that investment thread.

 

[Ramphex]

Beating the same v12.3 horse gets old so people discuss other things until v12.3.2.1.0 rolls out

 

[chado]

Last night drove a model 3 with 12.3.
For reference I use autopilot everyday and love it, makes things easier, reduces stress while driving.

This was my first FSD experience and I don't know how you guys do it, sometimes it works nicely but then it will do something so unexpectedly wrong; keeps you on edge all the time, not relaxing at all.
I think possibly testing it at night was a bad thing as a first experience, but this has completely removed my want to pick up another model 3 with FSD.
FSD fixed my insane obsessive want for FSD lol

 

[Ramphex]

Last night drove a model 3 with 12.3.
For reference I use autopilot everyday and love it, makes things easier, reduces stress while driving.

This was my first FSD experience and I don't know how you guys do it, sometimes it works nicely but then it will do something so unexpectedly wrong; keeps you on edge all the time, not relaxing at all.
I think possibly testing it at night was a bad thing as a first experience, but this has completely removed my want to pick up another model 3 with FSD.
FSD fixed my insane obsessive want for FSD lol

I honestly still like driving. Driving on FSD to me is more work than driving manually.

When I’m driving manually, I’m responsible for my actions and reactions to other people’s actions.

When FSD is driving I have to supervise and predict what unpredictable thing it’ll do next, supervise its actions and its reactions to other people’s actions, plus also have to please it by jerking off the steering wheel every 15-20 seconds and constantly looking straight to avoid “pay attention” nags.

Is it cool technology? Yes. Is it more work? To me, yes.

It does not allow me to relax. It does not allow me to not pay attention. It does not allow me to use a phone. With that being said, I still have to pay attention when I’m driving manually, but at least I don’t have to masturbate the steering wheel and can look at my phone if I want to without something complaining.

 

[Knightshade]

There's always a tweet:

https://twitter.com/x/status/1727484899374899687

It's funny to me that people think Elon is telling his engineers to make driver assist software when it's so obvious they're trying to make a robotaxi.

TBF Tesla already said, in legal documents, that Elons claims get ahead of engineering reality.

 

[sleepydoc]

Right. What I want is simple: Drive at the MAX speed if safe to do so.

In the world of programming, that's (relatively) trivial.

it seems like it should be, especially since speed control has pretty much been a non-issue in all the previous versions.

I suspect the issues we're seeing now are related to using neural net programming for everything and/or trying to do more adjustments for driving conditions. If the speed control is part of the neural net then isolating it out and chaining it may not be as trivial as it would first seen. This my uneducated impression - I don't have any experience with neural nets, so I may be way off, too.

 

[sleepydoc]

That and "why doesn't my car wakeup". You should see the investment thread. It is like the hive ... thousands of voices speaking about everything. It's truly schizophrenic.

In this thread there's many pages of "levels", definition of "drive" and so forth. It still pales to that investment thread.

well, the car not waking up could be part of the software update, even if it's not part of FSD so at least it's. tangentially related.

I'll stay off the investment thread, though.

 

[lzolman]

Last night drove a model 3 with 12.3.
For reference I use autopilot everyday and love it, makes things easier, reduces stress while driving.

This was my first FSD experience and I don't know how you guys do it, sometimes it works nicely but then it will do something so unexpectedly wrong; keeps you on edge all the time, not relaxing at all.
I think possibly testing it at night was a bad thing as a first experience, but this has completely removed my want to pick up another model 3 with FSD.
FSD fixed my insane obsessive want for FSD lol

I think a large part of the charm of being a beta tester (that applies to anyone using FSD anything) is the opportunity to contribute to the cause. Our feedback can supposedly make a difference, help make it better and stronger, faster. I don't "need" FSD on city streets as much as I "need" it on the highway, but I enjoy being part of this first-ever grand experiment. I don't enjoy the being "on edge all the time" (I refer to it as being on shpilkes) so I haven't always had FSD going when I could. I want to just purely enjoy driving my S a lot of the time.

But now my ratio of FSD driving to non-FSD driving is going up, because it's getting better. The more it improves, the more I'll be using it because the car just starts feeling like more of a partner. The human-FSD collaboration is slowly gaining ground on the "me alone" experience. This is a Good Thing and the source of my optimism.

 

[OxBrew]

You cant use supervised FSD miles/accident accident rates to say anything about unsupervised FSD safety. Supervision removes the failures from the testing data.

10 times safer than the average human driver equals one accident per 2 million miles.

We are all down here squatting on the sidewalk talking about how wide the cracks are (sub weeds level) when we need to be taking the 40,000 ft view. But we can't, because the fine micro cracks in the side walk keep making it fail.

3 months ago the most optimistic cheerleaders were saying v12 is robotaxi by the end of the year.

I'm still in the somewhere between 10 years and never camp, and never on this hardware.

 

[AAKEE]

Things have changed since the 90%. Tesla has read all the posts here on TMC from @AAKEE and have learned a thing or two. They have since changer their recommendation to 80%

Note that the notification does not show until exceeding 80%.
Even if it doesn’t really say clearly so, they mean 80% and below.

Just like the 90% never was really 90% but [below 90%], which means any value from 89% and below.

 

[kabin]

Here's another case of v12.3 not detecting or responding to a cement median while attempting an unprotected left turn. Heavy reliance on map data and little on vision. I experienced the same thing on one of my first v12 drives.

https://twitter.com/x/status/1773102329211785442

 

[kpanda17]

There's always a tweet:

https://twitter.com/x/status/1727484899374899687

It's funny to me that people think Elon is telling his engineers to make driver assist software when it's so obvious they're trying to make a robotaxi.

Similar that Neuralink and Boring Tech are really to applied to the Mars colonization

But for us to accept now, Elon is applying here
Give the mere mortals something now while Elon, rhe 4D long game player, is charting Mars with his and or one of his 11 children leading



Wow

 

[Daniel in SD]

You cant use supervised FSD miles/accident accident rates to say anything about unsupervised FSD safety. Supervision removes the failures from the testing data.

10 times safer than the average human driver equals one accident per 2 million miles.

We are all down here squatting on the sidewalk talking about how wide the cracks are (sub weeds level) when we need to be taking the 40,000 ft view. But we can't, because the fine micro cracks in the side walk keep making it fail.

3 months ago the most optimistic cheerleaders were saying v12 is robotaxi by the end of the year.

I'm still in the somewhere between 10 years and never camp, and never on this hardware.

I assume they are taking a random sample of disengagements and analyzing them to determine the percentage that are necessary.
I agree that people don't realize how hard it is to get human level safety which according to Tesla data is 1 serious collision (>~12mph) per million miles.

 

[JB47394]

This was my first FSD experience and I don't know how you guys do it

We do it by having gained familiarity with the system. There have been many posts by people who have their first FSD experience, then declare FSD to be a dead product. It wasn't an effortless Apple product with a refined experience, and that's what they demand when they put down their hard-earned money. As a result, it's a non-starter for them. FSD demands things of the driver, and if you're willing to go through the learning experience, you'll have gained a new tool for driving.

We're all hoping that, in time, that tool will be thoroughly natural to all drivers.

And, yes, making your first FSD experience take place at night is like jumping straight onto black diamond ski trails and being completely turned off to skiing as a result.

 

[PianoAl]

... the opportunity to contribute to the cause. Our feedback can supposedly make a difference, help make it better and stronger, faster.

Very true, and this has made me realize the following:

Tesla could improve this by telling us in detail how our report-button comments are used. Are they all transcribed by computer? Are there people who listen and categorize the feedback? Are there key words? If we make the same comment for the same problem multiple times, is it given more weight? Ignored?

IOW, if Tesla teaches us how best to use the feedback, everyone benefits.

 

[AlanSubie4Life]

For me V12 is absolutely a game changer.

Honestly, for me it has been a game changer.

Awesome. Looking forward to it.

and 12.3 fixed all of his persistent disengagements:

Awesome!

Looking forward to seeing everyone’s issues resolved.

I just watched a video of v12.3.2.1. Looks like they still have issues with auto-speed.

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