My FSDb Experience vs YouTube Videos

[PianoAl]

I've watched a lot of YouTube FSDb videos, but my first experiences seem a lot worse than theirs. For example, today it missed the road it was supposed to turn onto, went into a parking lot, said, "In 500 feet, your destination will be on the left," and started accelerating. I've had it be in the wrong lane for turning, almost stop in an intersection, etc.

In fact, in almost every instance that I turned it on, I had to disengage within a minute. Note that I'm in a rural area without that many Teslas.

Am I just not used to it, or is there some other reason my experience doesn't match that see in the YouTube videos?

 

[diplomat33]

I've watched a lot of YouTube FSDb videos, but my first experiences seem a lot worse than theirs. For example, today it missed the road it was supposed to turn onto, went into a parking lot, said, "In 500 feet, your destination will be on the left," and started accelerating. I've had it be in the wrong lane for turning, almost stop in an intersection, etc.

In fact, in almost every instance that I turned it on, I had to disengage within a minute. Note that I'm in a rural area without that many Teslas.

Am I just not used to it, or is there some other reason my experience doesn't match that see in the YouTube videos?

It is completely normal that your experience does not match the youtube videos. For one, those videos are a small sample. They will only show how FSD beta works in that particular area and under those particular conditions. They won't show the whole picture of how FSD beta works everywhere. Second, the videos may be curated to only show you when FSD beta was the best. Lastly, FSD Beta will not behave the same everywhere since roads, traffic etc are different from place to place. And it will depend on what data Tesla used to train their system. FSD Beta will work better in areas with more training data. You are in a rural area without a lot of training data so Tesla has likely not trained FSD Beta to handle your area. A lot of the youtube videos are from areas with a lot of training data so Tesla has trained FSD beta to handle those areas. Hence why FSD works better in those areas compares to your area.

 

[bayareaever]

I’m in the SF Bay Area with Teslas just about everywhere. My experience also doesn’t match the YT videos.

It does fairly well in light traffic but as soon as there is lots of cars/pedestrians/bikes, not so much. Phantom/exaggerated braking is still the biggest problem. Others include blowing through a faded stop sign (daytime only), unsafely blowing through a yield sign, wanting to blow through a red light after stopping for it (the same one every time). I’ve been testing for 16 months so I’ve seen the progression which is unmistakable, but there is a ways to go. Hopefully single stick will be a big improvement!

 

[Tam]

...Am I just not used to it...

Those YouTubers are experts because Dan O'Dowd used mannequins, but these Youtubers even use their very own small children in front of the car to test out the FSD beta.

 

[kabin]

Those YouTubers are experts because Dan O'Dowd used mannequins, but these Youtubers even use their very own small children in front of the car to test out the FSD beta.

Aka skin in the game.

 

[TS.42]

LOL this thread wins! The YouTube garbage is just shilling for views. RL FSD tries to kill you constantly. It starts out surprisingly good then just falls apart.

 

[PianoAl]

Are you saying that Dirty Tesla, for example, edits out problems??

 

[Snerruc]

I've watched a lot of YouTube FSDb videos, but my first experiences seem a lot worse than theirs. For example, today it missed the road it was supposed to turn onto, went into a parking lot, said, "In 500 feet, your destination will be on the left," and started accelerating. I've had it be in the wrong lane for turning, almost stop in an intersection, etc.

In fact, in almost every instance that I turned it on, I had to disengage within a minute. Note that I'm in a rural area without that many Teslas.

Am I just not used to it, or is there some other reason my experience doesn't match that see in the YouTube videos?

Your experience is similar to mine. I have turned it off and won’t try again until Tesla puts skin in the game by assuming responsibility for its actions. I suggest you shut it off completely and forget it’s there.

 

[bayareaever]

Your experience is similar to mine. I have turned it off and won’t try again until Tesla puts skin in the game by assuming responsibility for its actions. I suggest you shut it off completely and forget it’s there.

From what I understand, it won’t improve if no one uses it, so I personally keep using it. There is some work involved for beta testers though, no doubt. It does really impressive things sometimes, but just not consistently.

 

[srs5694]

From what I understand, it won’t improve if no one uses it, so I personally keep using it. There is some work involved for beta testers though, no doubt. It does really impressive things sometimes, but just not consistently.

There's not just work for beta testers, there's risk. If your car gets into an accident on FSD because FSD does something wrong, then you're responsible. Even if nobody is injured or killed, the points will go on your driving license, your insurance rates will go up, etc. FSD beta testers should never forget this.

I haven't seen any scientific studies of just how risky the FSD beta is, but my subjective impression is that it's very poor compared to even an average human driver. My own nickname for FSD is "Flaky Student Driver," and I think that sums it up better than any other three words I've heard. In my experience, it's more stressful and difficult to drive with FSD beta active than to drive manually, because I have to do all the mental tasks of driving normally (watching traffic and pedestrians, planning what to do if somebody does something stupid, etc.), plus try to anticipate what my own car will do and plan for its stupid moves.

That said, FSD is technically impressive, despite its flaws. It does things that were science fiction not so long ago. (Remember the TV show "Knight Rider?") Being technically impressive and being useful for everyday driving are two different things, though.

Despite being pretty "down" on the state of FSD as a useful feature, I do think that it needs to be tested. I do believe that self-driving technology (whether Tesla's or somebody else's, and most likely multiple development paths) has the potential to save a lot of lives in the long run. I don't know of a way to get there without testing the technology on real roads. What I have a problem with is Tesla's (and especially Musk's) marketing and sales practices, which make FSD seem more capable and less risky than it is, resulting in faster and more widespread deployment than is advisable.

 

[PianoAl]

Your experience is similar to mine. I have turned it off and won’t try again until Tesla puts skin in the game by assuming responsibility for its actions. I suggest you shut it off completely and forget it’s there.

That's what I'll do, except in one instance, which is the reason I got FSDb:

On a route I travel frequently, the car interprets the truck limit as the regular limit. FSDb allows me to override that, and drive any speed I want. I plan to turn on FSDb on that route alone. I've done it a few times, and it's been flawless.

 

[Dewg]

There's not just work for beta testers, there's risk. If your car gets into an accident on FSD because FSD does something wrong, then you're responsible. Even if nobody is injured or killed, the points will go on your driving license, your insurance rates will go up, etc. FSD beta testers should never forget this.

I haven't seen any scientific studies of just how risky the FSD beta is, but my subjective impression is that it's very poor compared to even an average human driver. My own nickname for FSD is "Flaky Student Driver," and I think that sums it up better than any other three words I've heard. In my experience, it's more stressful and difficult to drive with FSD beta active than to drive manually, because I have to do all the mental tasks of driving normally (watching traffic and pedestrians, planning what to do if somebody does something stupid, etc.), plus try to anticipate what my own car will do and plan for its stupid moves.

That said, FSD is technically impressive, despite its flaws. It does things that were science fiction not so long ago. (Remember the TV show "Knight Rider?") Being technically impressive and being useful for everyday driving are two different things, though.

Despite being pretty "down" on the state of FSD as a useful feature, I do think that it needs to be tested. I do believe that self-driving technology (whether Tesla's or somebody else's, and most likely multiple development paths) has the potential to save a lot of lives in the long run. I don't know of a way to get there without testing the technology on real roads. What I have a problem with is Tesla's (and especially Musk's) marketing and sales practices, which make FSD seem more capable and less risky than it is, resulting in faster and more widespread deployment than is advisable.

I think you hit on a good point - Beta is an early release for testing. It takes a certain type of person who wants to test the system and help improve it. I wrote an entire article about being careful what you wish for, as people were clamoring to get into the Beta in 2021. My opinion is that the vast majority of people in the Beta should not be in the Beta. It's not for them. It's for people who have a serious commitment to improving the software and are willing to do so with thoughtful and careful intent.

There is a big difference between:

1) Yay! I got into Beta. OMG, the car just tried to turn while traffic was coming and nearly killed me! This thing sucks - why is this even being allowed?!?

2) I've requested access to Beta and been invited. I'm going to plan routes and test features as they're released. The car just attempted a right turn when traffic was coming. I had to intervene and disengage the software. I've noted the condition and reported the incident to Tesla. I will continue testing the turn to see what conditions lead to success. I'll then learn what situations lead to repeatable failures, and test those scenarios with each update, continuing to report my findings.

With the most recent updates, the reporting feature was removed, but the disengagement still sends valuable data, and I continue to do so as I test.