Is it worth $3k for FSD for EAP owners

[Octo]

By discounts on FSD

That’s some sugary pay.

 

[daniel]

Your statement about planes is not totally true. In decent conditions the computer will do better than a human being, however when you are not within certain limits, the computer can't safely land the plane, the pilots have to do it, like strong winds.
My point is : if plane automation can't be 100% reliable after decades of utilization, will car automation succeed in a couple years or even a decade ?
Just a honest question.

Thank you. In correcting me about airplanes, you've made my point stronger.

why not invest your 3k$ in TSLA stock instead?. If they do come out with a reasonable FSD software suit, their stock will double or triple in no time. You can then sell your stock and possbily get FSD software for free or pretty close.

If, in the other hand, they can't deliver on their promises, worst case you can sell your TSLA shares and get most of your money back.

I do not see how handing away your money blindly can be a better option than buying TSLA.

This is so brilliant I wish I'd thought of it myself. Well, I do own some TSLA. This is the perfect answer to people who say they're buying FSD so as not to be hit with price hikes. If this thing works, that stock is going through the roof and would pay for a brand new car.

BTW, there's a guy here on TMC who bought TSLA when they first announced they would build the Model S. When his turn came to order his car, the stock paid for the car and still left him with more than when he went in. We met some years back. Don't remember his name.

Tesla seems to be taking an incremental approach to self-driving. Tesla is releasing features/improvements one by one, under driver supervision, validating each piece as they go. Some of these FSD features/improvements will be released relatively soon, like in the next 6-12 months, some FSD features/improvements will be released later. In terms of paying $3000 for FSD, it really boils down to how much progress you think Tesla will make in the short term versus the long term and what you are hoping to get out of the FSD software. Personally, I have no doubt that current owners today will certainly get some FSD features and benefits on the existing hardware even if their cars don't become FSD. Whether those benefits are worth $3000 is totally up to each individual.

Don't pay $3000 IF:
- You want a robotaxi next year.
- You think the progress will be very slow and very few FSD features will be released in the next 1-3 years.
- You are leasing your car.
- You are planning to trade in your car soon.

Do pay $3000 IF:
- You plan to keep your car long term.
- You are ok with not getting a robotaxi next year.
- You want the latest FSD features even if they require driver supervision.
- You like being a guinea pig for new AP/FSD features.
- You think Tesla will make enough progress that you will get $3000 worth of AP/FSD features in the next 1-3 years.

We've come to total agreement. I love it when that happens!

I am still trying to figure out how these features, by themselves, will be useful. Unless the car can also properly yield to pedestrians in or about to enter a crosswalk, handle a four-way stop, and so on, I won't be engaging any of the automated systems. I want to know what it does AFTER it sees it has a green light, yet there's someone in the crosswalk that you have to cross to make the desired right turn. How's it going to handle a four-way stop when two drivers arrive at the same moment?

All issues that have to be solved, and they won't be solved overnight.

That said, I did buy FSD for the simple reason that it will, in my opinion, likely result in better EAP behavior on the highway, which is where it's really useful to me. I don't care about parlor tricks and if the car never drives itself on city streets, I really won't care.

I've seen this idea put out before, that the HW3 computer will improve EAP performance. I know of no reason to think this would be true. It would depend on whether the software is designed in such a way that it can benefit from a faster computer. I don't think we know this.

Ummmmmm You do realize that Tesla has been running FSD on some of it's cars (including Elon's) for MONTHS now, and it is CURRENTLY running in "shadow mode" on the entire fleet right?

No.

What is running on Elon's car might be an alpha test version of the features they're working on, or some of those features. It's not FSD. And "shadow mode" means it doesn't work yet so they're using your car to run a distributed computing test of the software. If it worked they'd release it.

The questions are, 1: Will they be able to make this software work? (I say yes, they will.) And 2: When will it be ready to release to customers' cars? (I say a year for the first "feature" not included in EAP which will still be Level 2; and five to ten years for Level 3 -- driver can read a book while sitting in the driver's seat.)

 

[evannole]

ve seen this idea put out before, that the HW3 computer will improve EAP performance. I know of no reason to think this would be true. It would depend on whether the software is designed in such a way that it can benefit from a faster computer. I don't think we know this.

No, we don't know it for sure, but the Autonomy Day presentation made plenty of references to HW3's ability to process images and make decisions at a much greater rate, which can't hurt performance on the highway, and I think it's reasonable to believe that it would help. The subsequent demonstration drive after the presentation was actually largely on the highway and the car appeared much more capable than my EAP car is today.

 

[daniel]

No, we don't know it for sure, but the Autonomy Day presentation made plenty of references to HW3's ability to process images and make decisions at a much greater rate, which can't hurt performance on the highway, and I think it's reasonable to believe that it would help. The subsequent demonstration drive after the presentation was actually largely on the highway and the car appeared much more capable than my EAP car is today.

Maybe it will, and maybe it won't. It's a big chunk of money to spend on a "maybe." Of course, when they finally get around to installing your new computer you'll also get the new software. I expect to trade up to a new car in five years, which is when I expect there to be new features I actually want, running on HW4, with better sensors.

But then, I buy things for what they can do, not for what the dreamers think they might be able to do someday.

 

[evannole]

Maybe it will, and maybe it won't. It's a big chunk of money to spend on a "maybe." Of course, when they finally get around to installing your new computer you'll also get the new software. I expect to trade up to a new car in five years, which is when I expect there to be new features I actually want, running on HW4, with better sensors.

But then, I buy things for what they can do, not for what the dreamers think they might be able to do someday.

Fair enough. And I buy cars and keep them for far longer than five years.

 

[smodtactical]

$3000? Wow... in Canada this upgrade is $7900 CAD.. brutal.

 

[Big Dog]

$3000? Wow... in Canada this upgrade is $7900 CAD.. brutal.

It's only $3k (USD) for those who already have EAP. If you have AP, its $6k to upgrade.

 

[pscarroll]

As someone who bought FSD at purchase time in 2018, I'd say that you should not spend the $3000. There have been zero updates that have benefited us that paid last year (or even before), even the promise of getting early updates has been ignored.

 

[Billyk24]

As someone who bought FSD at purchase time in 2018, I'd say that you should not spend the $3000. There have been zero updates that have benefited us that paid last year (or even before), even the promise of getting early updates has been ignored. ---------I call this a form of Capital Raise. No one knows when or if or how much is updated.

 

[rickjames]

[QUOTE="daniel, post: 3915233, member: 1500"
I think that @AlanSubie4Life is spot-on when he says that the first people with self-driving cars will not be the ones who buy the FSD package today; they'll be the ones who buy the new cars that have the necessary hardware, because our cars don't, and it was simply wrong to say that our cars had all the needed hardware before the software existed. Before you have the software you cannot know what hardware will be needed.

Ummmmmm You do realize that Tesla has been running FSD on some of it's cars (including Elon's) for MONTHS now, and it is CURRENTLY running in "shadow mode" on the entire fleet right?

This entire notion of "based on what I see now it's nowhere near ready so let me speculate beyond infinity as to what I think will make it work" is nonsense. The software in your car is not even close to what will be in V10 V11 and V12 running HW3. To compare the two is like comparing a MacBook to a Speak and Spell.

Tesla (Elon) is saying "Have faith in me and you will be rewarded with a discounted price, wait for me to show you I can land a rocket on a remote control ship 400 miles off the coast, and you pay retail."

YES! It's worth it. Duh.

DISCLOSURES: I already bought the upgrade at discount.

I have a few issues with most arguments people are making.

1. When the software releases that requires HW3 they aren't going to stop producing new cars to give the 10s of thousands of people the upgraded CPU. You will wait in "service hell" to get it installed whenever they get around to it, which will probably take several months. This is evident given the issues in the parts supply chain and repair process now. Priority of parts always goes to producing an entire car for sale over repairing one that someone already paid for.

2. Using NOA, the car still can't decide if it wants to change lanes under several repeatable conditions for reasons that just don't make any sense. For example, while driving in lane 2, a motorcycle in the HOV lane to the left, let's call it lane B of A&B of the available lanes, was paralleling at the same speed as me. A whole lane separates us. While switching from lane 2 to lane 1 on NOA, the vehicle executed the lane change and aborts while halfway over the lane marker, returning to original lane. It though the bike was in the lane it was switching to. It made a 2nd attempt before I disengaged and took over. I've also had it swerve as detailed above at night in the desert with nothing around at all.

3. There is still phantom braking. It is drastically reduced, but still exists. For some reason in the right lighting conditions, a large vehicle maintaining the same speed or slightly slower than your own and drifting just enough to touch a lane marker is enough to trigger it. I have decent success repeating this in traffic on my drive home.

I love the car, so far and I am sure Tesla will get there eventually, but these are glaring issues that still aren't fixed since I downloaded the update with NOA.

Until then, I think most people's money would be better served in other places, like ppf, tint or maybe a vacation.

 

[mongo]

Using NOA, the car still can't decide if it wants to change lanes under several repeatable conditions for reasons that just don't make any sense. For example, while driving in lane 2, a motorcycle in the HOV lane to the left, let's call it lane B of A&B of the available lanes, was paralleling at the same speed as me. A whole lane separates us. While switching from lane 2 to lane 1 on NOA, the vehicle executed the lane change and aborts while halfway over the lane marker, returning to original lane. It though the bike was in the lane it was switching to. It made a 2nd attempt before I disengaged and took over. I've also had it swerve as detailed above at night in the desert with nothing around at all.

This was discussed at one of the autonomy presentations. They are training the NN to detect vehicles that are going to change lanes even if they are not signaling. Guessing the motorcyle version of that feature is too sensitive.

Night swerving with nothing around sounds like low light induced artifacts or unseen (by you) creatures. It wasn't trying to lane change, was it?

 

[Saghost]

DISCLOSURES: I already bought the upgrade at discount.

I have a few issues with most arguments people are making.

1. When the software releases that requires HW3 they aren't going to stop producing new cars to give the 10s of thousands of people the upgraded CPU. You will wait in "service hell" to get it installed whenever they get around to it, which will probably take several months. This is evident given the issues in the parts supply chain and repair process now. Priority of parts always goes to producing an entire car for sale over repairing one that someone already paid for.

2. Using NOA, the car still can't decide if it wants to change lanes under several repeatable conditions for reasons that just don't make any sense. For example, while driving in lane 2, a motorcycle in the HOV lane to the left, let's call it lane B of A&B of the available lanes, was paralleling at the same speed as me. A whole lane separates us. While switching from lane 2 to lane 1 on NOA, the vehicle executed the lane change and aborts while halfway over the lane marker, returning to original lane. It though the bike was in the lane it was switching to. It made a 2nd attempt before I disengaged and took over. I've also had it swerve as detailed above at night in the desert with nothing around at all.

3. There is still phantom braking. It is drastically reduced, but still exists. For some reason in the right lighting conditions, a large vehicle maintaining the same speed or slightly slower than your own and drifting just enough to touch a lane marker is enough to trigger it. I have decent success repeating this in traffic on my drive home.

I love the car, so far and I am sure Tesla will get there eventually, but these are glaring issues that still aren't fixed since I downloaded the update with NOA.

Until then, I think most people's money would be better served in other places, like ppf, tint or maybe a vacation.

I haven't been terribly happy with NoA's strategic decision making so far.

Having said that, it's not obvious to me how many of the issues we have now will still happen when the car is running full resolution, high frame rate FSD Neural Nets.

Simply seeing the world more clearly and more quickly may improve the resulting performance substantially.

 

[daniel]

Fair enough. And I buy cars and keep them for far longer than five years.

I normally do, also. 12 to 15 years for most of the cars I've owned. But I do not believe the significant upgrades (L3, FSD) will be possible with the hardware in my car, so I expect to trade in the car in 5 years, at which time it will be 6 1/2 years old. I really think that in 5 years the hardware (both sensors and computers) will have improved so much that it will not be possible to bring a 2018 car up to the new standard of performance. Rather than upgrade the cars, Tesla will refund the money to people who paid for robo-taxi-level FSD.

I haven't been terribly happy with NoA's strategic decision making so far.

Having said that, it's not obvious to me how many of the issues we have now will still happen when the car is running full resolution, high frame rate FSD Neural Nets.

Simply seeing the world more clearly and more quickly may improve the resulting performance substantially.

I'm sure it will. The big question is: Will it improve enough to get the car to Level 3? Honestly, I do not believe that HW3 and the current suite of sensors will be able to get it to NoA/city Level 3. If I'm wrong, and it does, I'll trade up my car. I really want to be wrong on this one. So I'm not paying for FSD now, and I win either way: If they can't manage it, I've saved $3,000. If they do manage it, I'll be so happy with the new car I won't care about the price. And since it's clear that retro-fitting will take many months or even a year or more, trading up will be the quickest way to get to the new level of performance.

 

[diplomat33]

Honestly, I do not believe that HW3 and the current suite of sensors will be able to get it to NoA/city Level 3.

I am curious. What hardware do you think would be good enough for City L3? Do you believe the cars need LIDAR or do you think extra radar and extra high definition cameras would be good enough?

 

[daniel]

I am curious. What hardware do you think would be good enough for City L3? Do you believe the cars need LIDAR or do you think extra radar and extra high definition cameras would be good enough?

I think it needs full-surround radar or lidar (I don't know enough to say whether lidar is superior to radar, since both bounce photons off of targets) as well as full-surround cameras. I don't think that at present it has enough of either. I also think they are two generations shy of sufficient computing power (maybe HW5 will get there). And finally, I think that neural network technology is not yet mature enough.

I think we're five years away from Level 3 and ten years away from no-driver-needed.

And I really hope I'm wrong. I want to have to come back here in December of this year and post that you were right all along.

 

[willow_hiller]

I think it needs full-surround radar or lidar (I don't know enough to say whether lidar is superior to radar, since both bounce photons off of targets) as well as full-surround cameras.

Cannot speak to the need for surround lidar/radar, but the camera coverage in HW3 is pretty much surround:

It would be nice to see a demonstration of whether any blind-spots exist, but as far as I can see there are none.

 

[diplomat33]

I think it needs full-surround radar or lidar (I don't know enough to say whether lidar is superior to radar, since both bounce photons off of targets) as well as full-surround cameras. I don't think that at present it has enough of either. I also think they are two generations shy of sufficient computing power (maybe HW5 will get there). And finally, I think that neural network technology is not yet mature enough.

I think we're five years away from Level 3 and ten years away from no-driver-needed.

And I really hope I'm wrong. I want to have to come back here in December of this year and post that you were right all along.

I think you are too pessimistic. Teslas already have full surround cameras. And the FSD computer is good enough IMO. The NN is just software that will mature as Tesla plugs more fleet data into the machine learning. I do agree that extra radar would be useful. I go back and forth on lidar. I think you can probably use radar instead of lidar.

No way it will take 10 years to get to no driver self-driving. I think you are underestimating the power of the Teslas fleet data to validate FSD. Right now, we are in the slow, laborious phase of FSD development because Tesla still has to finish the individual pieces of FSD. But once that is complete, it will take much less time to validate if FSD is reliable or not thanks to the size of the fleet. With millions of miles of driving probably per day, it won't take very long for Tesla to see if FSD is reliable or not and tweak the software.

I say we are 4-6 years away from L4 autonomy in our Teslas.

 

[pscarroll]

Purchasing FSD will, as of today, give you absolutely nothing extra. No new computer. No early access program. No new or exclusive features. You're basically giving Tesla $3000 in advance, hoping that they will release new features and give you a free computer upgrade, plus the promises of being in the early access program that seems to only materialize for those who "know someone".

 

[daniel]

I think the cameras need redundancy, not just full coverage. What happens to your no-driver car when a splot of mud hits the camera lens?

... I think you are underestimating the power of the Teslas fleet data to validate FSD. [...]

I say we are 4-6 years away from L4 autonomy in our Teslas.

I think the fleet data will be able to validate FSD very quickly. I don't think the programmers are going to solve the programming difficulties very quickly.

But your "4-6 years away from L4" is really not all that different from my slightly more pessimistic 5 years to L3 and 10 years to L5. Our positions seem not that far apart. As noted earlier, I like that. Maybe where we are further apart is that I don't think my 2018 Model 3 will be upgradable to L5. Or if it is, it will be an inferior version of L5 compared to the new cars that are built after the technology is actually ready for consumer cars.

 

[diplomat33]

But your "4-6 years away from L4" is really not all that different from my slightly more pessimistic 5 years to L3 and 10 years to L5. Our positions seem not that far apart. As noted earlier, I like that. Maybe where we are further apart is that I don't think my 2018 Model 3 will be upgradable to L5. Or if it is, it will be an inferior version of L5 compared to the new cars that are built after the technology is actually ready for consumer cars.

No, we are not that far apart.

Tech is always changing. And Tesla, especially, likes to innovate and upgrade their cars as they go. So it is a given that Teslas in 5-10 years from now will have much better FSD hardware than what our 2018 cars have. So yes, I think we can be pretty sure that whatever FSD we get will be inferior to the FSD in 5-10 years, just because of the nature of technological progress. Whether our current cars will ever be able to do L5, that is a slightly different question. I am on record as saying that our current 2018 cars will probably never do L5. However, I do believe that our current cars can be L3 and probably L4.

And just to make sure we are on the same page: L3 is full self-driving within a limited ODD, with a human driver as fall back, L4 is full self-driving within a limited ODD with system as fall back, and L5 is full self-driving with no limits on ODD and with system as fall back.