Tesla says there are 60,000 FSD Beta testers now

[EVNow]

Just slight improvements every version, culminating to an improvement of a disengagement rate from (just making numbers up) 1 per mile to maybe 10 per mile.

You mean from 1 every mile to 1 every 10 miles ?

BTW, Elon made somewhat of a similar mistake on the call. He talked about disengagements per mile ... though he could have been talking in fractions (I hope!). More likely he was talking about may be disengagements per 100 miles or something like that.

 

[JulienW]

I don't believe there will be any jump. Following FSD Beta's development from late 2020 to now, there have been no major jumps. Just slight improvements every version, culminating to an improvement of a disengagement rate from (just making numbers up) 1 per mile to maybe 10 per mile. I think it'll just continue to gradually increase, and POSSIBLY, by the end of the year, it'll get to a point where they'll open up FSD Beta to everyone as an advanced L2 driving assist, and say it's safer than the human driver in the same way they say about Autopilot.

Yea, I bet that if you took 5 different version of Beta starting with 1.0 from almost 1½ years ago and 10.9 with 3 in-between and did a double blind test on all 5 most people would not be able to rank them in order. So if it is soooo close to the 1½ year Beta version how is it supposed to VASTLY improve to Level 4 equivalent in less than a year? Hate to say it and hope I'm wrong but I bet we are looking at Level 3 in about 2 years and probably not reach Level 4 with the current hardware. I just can't see the current cars EVER reaching Robotaxi Level 5 as they are.

Having said that I will be happy with a strong Level 3 and hope Tesla cracks Level 4 (and 5) even if it takes major hardware upgrades on future cars.

 

[Rice&Curry]

From the shareholder deck that was just released for the earnings call later today, Tesla says that there are now 60,000 FSD beta testers:



Earnings Call Release

I am very pleasantly surprised. That number is higher that I thought it would be.

Another way to look at it is that we atleast have 60K very safe, cautious and dependable Tesla drivers on the road. ( I am not one of them - opted out based on my score)

 

[momo3605]

You mean from 1 every mile to 1 every 10 miles ?

Yes, i'm just making numbers up, but anecdotally, with my bay area suburb driving, i'm probably close to one disengagement every 10 miles. (intervening with gas pedal probably more)

 

[momo3605]

Another way to look at it is that we atleast have 60K very safe, cautious and dependable Tesla drivers on the road. ( I am not one of them - opted out based on my score)

why not stay opted in? If your safety score is low, you won't get it till it's improved significantly. It's only available for 97 or 98-100 right now i think.

 

[momo3605]

Yea, I bet that if you took 5 different version of Beta starting with 1.0 from almost 1½ years ago and 10.9 with 3 in-between and did a double blind test on all 5 most people would not be able to rank them in order. So if it is soooo close to the 1½ year Beta version how is it supposed to VASTLY improve to Level 4 equivalent in less than a year? Hate to say it and hope I'm wrong but I bet we are looking at Level 3 in about 2 years and probably not reach Level 4 with the current hardware. I just can't see the current cars EVER reaching Robotaxi Level 5 as they are.

Having said that I will be happy with a strong Level 3 and hope Tesla cracks Level 4 (and 5) even if it takes major hardware upgrades on future cars.

I'd be happy with some limited L3 as well, but i'm skeptical because Elon has never mentioned interest in L3 at all. Always had an all-or-nothing mentality. But maybe if it starts to look like there's no path to L4 or L5, maybe he'll change his tune and settle with L3 as the goal.

 

[FloridaJohn]

Yes, i'm just making numbers up, but anecdotally, with my bay area suburb driving, i'm probably close to one disengagement every 10 miles. (intervening with gas pedal probably more)

I can't even get out of my neighborhood without a disengagement (maybe 1 mile, probably less). I have a T intersection that the car has yet to complete properly. It comes to a stop (properly), and then creeps out into the intersection to see if it is safe to go. By the time it determines it is safe, it is so far out that it is blocking both lanes. If there are any cars coming from either direction I have to disengage.

 

[momo3605]

I can't even get out of my neighborhood without a disengagement (maybe 1 mile, probably less). I have a T intersection that the car has yet to complete properly. It comes to a stop (properly), and then creeps out into the intersection to see if it is safe to go. By the time it determines it is safe, it is so far out that it is blocking both lanes. If there are any cars coming from either direction I have to disengage.

what about after you exit your neighborhood? I have a similar issue, neighborhood speed limit is 15mph with 5mph speed bumps, but tesla maps says it's 25. Tries to blow way too fast through the neighborhood. At this point, i don't activate FSD until i get out of my neighborhood (about 500 feet)

 

[Sporty]

I can't even get out of my neighborhood without a disengagement (maybe 1 mile, probably less). I have a T intersection that the car has yet to complete properly. It comes to a stop (properly), and then creeps out into the intersection to see if it is safe to go. By the time it determines it is safe, it is so far out that it is blocking both lanes. If there are any cars coming from either direction I have to disengage.

It can’t drive in my neighborhood. Among other things it stops for incorrect reasons (pedestrian cross walks, bikes, …). I’ve read “just push the go peddle” for many “it won’t go” or “stopping” scenarios. I don’t really get that view. If feels like a cheat. Yes I want the car to stop if it doesn’t know what the safe thing to do is. But I also want it to not need interventions all the time. It literally can not drive up the road without interventions.

Even the phantom breaking on the highway is too much. It’s probably 1 in a couple hundred miles for me, but that is too much. It’s in no way better then how I can drive.

Jointly together we might be better (but it’s a hassle for me due to its behavior). To hear Elon talk he thinks it will do better then normal drivers would do. Its behavior is very far away from such a thing.

 

[Spuzzz]

It’s about to be 59,999 will be trading mine in on Sunday and did not buy FSD on new one. No regrets … I enjoyed being part of the process. I know exactly where they are in development I don’t feel I’ll be really missing out on much for some time. I did see real progress, but they’ll be chasing 9s for awhile. I do wish they offered EAP…I would have bought that.

 

[FloridaJohn]

what about after you exit your neighborhood? I have a similar issue, neighborhood speed limit is 15mph with 5mph speed bumps, but tesla maps says it's 25. Tries to blow way too fast through the neighborhood. At this point, i don't activate FSD until i get out of my neighborhood (about 500 feet)

Not much better, I'm afraid.

On my short 5 mile non-highway commute, FSD Beta regularly fails at the flashing light in front of the fire station, the school zone, the roundabout, and the four-way stop.

That doesn't even count the fact that it can't pull into the parking lot at work, or pull into the garage when coming home. Both of which are required for a fully autonomous vehicle.

 

[Battpower]

flashing light in front of the fire station, the school zone, the roundabout, and the four-way stop.

add to that my experience with good ole 'not FSD' (in UK) railroad crossing lights and semi truck tail lights (at high and low level on same truck) and it feels like the NN really struggles with light context and flashing Vs steady.

Listening to the not so rosey FSD Beta reports it often feels like many of the current beta issues could be fundamental to Tesla's model..... which again at a fundamental level maybe hasn't changed as much as they would have us believe.

Even with the super simplistic NN / machine learning models I have tinkered with, I was surprised at how an apparently well trained NN that knew it's stuff could be thrown right off by seemingly random changes. It wasn't easy to know for sure what attributes were being depended on most heavily to determine between certain characteristics.

 

[saflak]

I hope you are right as I have it on two cars and many people have waited five years for the product. I wish I could say city streets is close to being usable without making big mistakes. But it makes big mistakes and unpredictable big mistakes. For example, a couple days ago it “tried” to do an unprotected left in my neighborhood and thank heavens an approaching car slowed. It hesitated and then froze — and this is an intersection I have had no problems with for weeks. I do pay constant attention and probably had time to avoid a collision but I really believe a 15 year old with a new learners permit would do better.

In summary, I would never say I should have the product when others don’t, but I see a rash of accidents if this becomes general release soon.

Autopilot also can be glitchy, and risky. But that should not keep it from a general release to those who paid for it. The benefits far outweigh the risks. And the concept that a few bad apples ruin it for the rest of us, is not a justifiable reason to not release the FSD to those who paid for it many years ago. If we are proven to be irresponsible end users of the FSD, then kick us out but don't make us presumed guilty until proven innocent!

 

[FelixORiley]

Autopilot also can be glitchy, and risky. But that should not keep it from a general release to those who paid for it. The benefits far outweigh the risks. And the concept that a few bad apples ruin it for the rest of us, is not a justifiable reason to not release the FSD to those who paid for it many years ago. If we are proven to be irresponsible end users of the FSD, then kick us out but don't make us presumed guilty until proven innocent!

I’m with you. I’ll add as a fallback position that I’m okay if Tesla uses Safety Score as a means to mitigate liability/publicity impacts, but once someone has achieved whatever criteria (e.g. 98 with 100+ miles and 7+ days of driving since requesting FSD beta), they should get an automatic invite.

 

[EVNow]

Autopilot also can be glitchy, and risky. But that should not keep it from a general release to those who paid for it. The benefits far outweigh the risks. And the concept that a few bad apples ruin it for the rest of us, is not a justifiable reason to not release the FSD to those who paid for it many years ago. If we are proven to be irresponsible end users of the FSD, then kick us out but don't make us presumed guilty until proven innocent!

Currently the benefits do not far outweigh the risks. It is simply not ready for wider release. I think wider release should be only after disengagement rates are on par with human errors - something like 1 in 10,000 miles. Still need 1,000x improvement for that.

 

[Fernand]

Well, maybe it's a specialty item. I hear two voices. One group complains that it's not close enough to "full autonomy". Another group seldom takes part in these discussions, but mostly uses it and simply works around the flaws. I suspect this is the majority.

I'm in the latter demographic and for over 3 years have used it all the time on the freeways, and now in town. I would be devastated if I lost it. It would be like losing half the miracle of the Tesla. It's especially welcome on the many unprotected highways that don't qualify for freeway status and used to only get the measly "base autopilot" treatment.

Of course it's a driver assist that requires literally a few weeks of operator re-learning and adaptation to a different (supervisory) driving role, and then attentive operation during use; it's not a Level 5 robotaxi. Nobody has gotten there yet. Development has gradually revealed how tricky that ever-receding objective is. Yet if they said that the current system is all we get for now, I'd be ok with that; it's very helpful to me. And well worth the money, BTW.

 

[momo3605]

Currently the benefits do not far outweigh the risks. It is simply not ready for wider release. I think wider release should be only after disengagement rates are on par with human errors - something like 1 in 10,000 miles. Still need 1,000x improvement for that.

Elon already believes FSD Beta disengagements are super rare. By whatever internal metrics autopilot team feeds Elon, he may not realize the general fleet disengagement rate is closer to 1 in <10 miles.

 

[azred]

Elon already believes FSD Beta disengagements are super rare. By whatever internal metrics autopilot team feeds Elon, he may not realize the general fleet disengagement rate is closer to 1 in <10 miles.

I seriously doubt Elon believes disengagements are rare. He is Tesla’s chief PR guy among his many hats so some lying is expected. My personal experience with two cars on FSD Beta is close to the 1 in 10 miles rate you mentioned. And this experience is mostly on the easy, modern, wide streets of South Chandler, Arizona, which Waymo probably chose for testing at least partially for that reason. Yet I still can’t go far without intervening. I cannot imagine the frequency of issues if I primarily drove on two lane roads or the narrow horse and buggy trails typical of many older cities.

 

[EVNow]

Elon already believes FSD Beta disengagements are super rare. By whatever internal metrics autopilot team feeds Elon, he may not realize the general fleet disengagement rate is closer to 1 in <10 miles.

I think its all relative - if he had to take over once a mile early during FSD beta - now can drive for ten miles, it feels "rare" !

 

[pitabun]

If pressing the accelerator = intervening, then yes it happens all the time, with each round-about, or if turning and a car's behind me.

One thing that makes this dead in the water are right hand turns from bike lanes. Turning right, Tesla stays in the lane. Other cars behind me are confused by my turn signal, slowly passing me to take a right in the bike lane.