FSD Upgrade Price reduced from $5K to $3K in MyTesla Page for Model S

[ucmndd]

It's a trust issue. Tesla says alot of things especially regarding fsd.

Certainly, but generally not things that can be considered cut and dry fraud, such as telling people unequivocally that the hardware upgrade is included in the price and then deciding later to charge thousands of dollars for it.

If we don’t trust the company in the first place, WTF do we hope to gain by clogging up their support resources with stupid questions asking them to endlessly confirm what they’ve already said clear as day?

 

[docbrown]

... what they’ve already said clear as day?

I hardly think that a response tweet is 'clear as day'. If it's so clear and obvious, why the hell don't they state this on the order page?? It's a pretty big selling point.

 

[ucmndd]

I hardly think that a response tweet is 'clear as day'. If it's so clear and obvious, why the hell don't they state this on the order page?? It's a pretty big selling point.

The feature is worthless without hardware to support it. Elon has said at least 3 times that FSD orders receive new hardware if necessary to enable the features you bought with the FSD package. The confusion on this topic is manufactured.

Not that I'm buying it either way...

 

[BigTonyTones]

The feature is worthless without hardware to support it. Elon has said at least 3 times that FSD orders receive new hardware if necessary to enable the features you bought with the FSD package. The confusion on this topic is manufactured.

Not that I'm buying it either way...

The issue is they changed their wording on FSD. The old description stated much more. I suspect they can accomplish their new revised FSD feature list with the current hardware and further out they will have an FSD + that will require newer hardware to work properly.

 

[SucreTease]

Why are people still confused about this? How much clearer can they be?

There is an even clear statement from Elon:

 

[rnortman]

The feature is worthless without hardware to support it. Elon has said at least 3 times that FSD orders receive new hardware if necessary to enable the features you bought with the FSD package. The confusion on this topic is manufactured.

Not that I'm buying it either way...

There is a very good chance that FSD as currently defined (after the recent re-definition) can be supported on HW2/2.5. This is the "danger" for people who click the order button today, because they are buying what is currently described on the order page, not what the old order page described. If Tesla can deliver that on HW2/2.5, they are not going to upgrade people to HW3 -- even if they can do it better on HW3.

To me, it's as simple as that, no matter what Elon tweets or what the clueless support reps say. You know how many times I've been told something I know for a fact is incorrect by Tesla support reps and even service center employees? How many times have Elon's tweets turned out to be BS? None of that matters in the slightest. If they can deliver on HW2, you will not get HW3, period.

 

[emmz0r]

There is a very good chance that FSD as currently defined (after the recent re-definition) can be supported on HW2/2.5. This is the "danger" for people who click the order button today, because they are buying what is currently described on the order page, not what the old order page described. If Tesla can deliver that on HW2/2.5, they are not going to upgrade people to HW3 -- even if they can do it better on HW3.

To me, it's as simple as that, no matter what Elon tweets or what the clueless support reps say. You know how many times I've been told something I know for a fact is incorrect by Tesla support reps and even service center employees? How many times have Elon's tweets turned out to be BS? None of that matters in the slightest. If they can deliver on HW2, you will not get HW3, period.

They can't deliver it on HW2, it's too intense with failover etc. He has promised HW3 over and over to all people who bought FSD, and it has been consistent. He also said it in earnings call.

 

[rnortman]

They can't deliver it on HW2, it's too intense with failover etc.

Have you read the new description of FSD? It is only promised to be an L2 driver assistance feature, i.e., requiring supervision all the time. What aspect of it exactly do you think cannot be done on HW2/2.5?

 

[BigTonyTones]

Have you read the new description of FSD? It is only promised to be an L2 driver assistance feature, i.e., requiring supervision all the time. What aspect of it exactly do you think cannot be done on HW2/2.5?

Exactly my thinking.

 

[emmz0r]

Have you read the new description of FSD? It is only promised to be an L2 driver assistance feature, i.e., requiring supervision all the time. What aspect of it exactly do you think cannot be done on HW2/2.5?

The additional computation for traffic lights roundabouts on top of an already stressed HW2. In addition they already have a separate neural model for HW3 and they are doing testing on Retrofitted cars already.

 

[BigTonyTones]

The additional computation for traffic lights roundabouts on top of an already stressed HW2. In addition they already have a separate neural model for HW3 and they are doing testing on Retrofitted cars already.

Not trying to sound rude,was this information posted anywhere? I'd love to read it to help reassure myself lol

 

[emmz0r]

Not trying to sound rude,was this information posted anywhere? I'd love to read it to help reassure myself lol

Plenty of discussion here: Autonomous Vehicles
Tesla autopilot HW3

It needs to handle a lot more objects and contexts at a much higher rate

 

[rnortman]

The additional computation for traffic lights roundabouts on top of an already stressed HW2. In addition they already have a separate neural model for HW3 and they are doing testing on Retrofitted cars already.

Doing a bad job of recognizing traffic lights and stop signs is not too much of an additional strain frankly. When I say "bad job" I mean you have something less than the 99.999% reliability you'd want for an unsupervised system, like maybe 98% of red traffic lights recognized in good lighting conditions (plus some annoying level of false positives on non-traffic-lights or recognizing a green light as red, and poor performance in various lighting conditions). They can also limit this to low speed operation -- you don't need to process 30 frames per second to detect traffic lights. Probably 2fps is good enough for this at low speeds (<45mph maybe). Roundabouts similarly can be navigated at a crawl.

But that is actually completely beside the point. Do you know why? The current order page does not promise roundabouts, nor does it promise reliable and fully autonomous handling of traffic light-controlled intersections! Don't believe me? Here is the description:

• Recognize and respond to traffic lights and stop signs.
• Automatic driving on city streets.

"Recognize and respond to traffic lights and stop signs" could mean that it comes to a stop when it recognizes one. And as a supervised system, it makes no guarantee it will recognize them all the time; that remains your job. That description does not promise that the car will start driving again when its turn has come. I think Tesla will certainly try to do that, but they are not promising it. They're not even promising that it will come to a stop! "Respond to" could just mean an audible warning to the driver to take over. This description is so non-specific it's a joke. This is all they're committing to, though of course they will try to do better.

So where are the roundabouts in there? That second bullet does not say all city streets, and it doesn't even mention intersections at all. If it can drive on some city streets automatically (which honestly it already does), then that bullet is covered. If they can navigate some intersections with human supervision, they are very comfortably able to check that box off and call it delivered. And I believe that they will be able to navigate quite a lot of intersections -- nowhere near all, but quite a lot -- with driver supervision at all times.

I believe all of the above can be done on HW2/2.5. Saint Elon himself has been saying so for 2+ years now.

 

[iTech]

The additional computation for traffic lights roundabouts on top of an already stressed HW2. In addition they already have a separate neural model for HW3 and they are doing testing on Retrofitted cars already.

HW2/2.5 is not maxed out yet.

 

[emmz0r]

HW2/2.5 is not maxed out yet.

I never said it was maxed out, but I'm sure will be when the rest is added in.
You need more accurate models to process stuff, so you don't get "dancing" lines/edges etc. HW2 has a different and simpler model than HW3.

Doing a bad job of recognizing traffic lights and stop signs is not too much of an additional strain frankly. When I say "bad job" I mean you have something less than the 99.999% reliability you'd want for an unsupervised system, like maybe 98% of red traffic lights recognized in good lighting conditions (plus some annoying level of false positives on non-traffic-lights or recognizing a green light as red, and poor performance in various lighting conditions). They can also limit this to low speed operation -- you don't need to process 30 frames per second to detect traffic lights. Probably 2fps is good enough for this at low speeds (<45mph maybe). Roundabouts similarly can be navigated at a crawl.

I believe all of the above can be done on HW2/2.5. Saint Elon himself has been saying so for 2+ years now.

You forget that you have a zillion things going on in a roundabout and surface city streets. But you're human so I don't expect you to understand since you do it automatically being a human.

You need to react in a split second in the traffic, 2 Hz is simply not good enough. You need more than that to anticipate vectors properly to provide a smooth response. Or maybe you enjoy ghost braking
Sure you can do everything on HW2, but it will be faaaaaaaaaaaar from optimal.

Automatic city street driving means automatic, not semi-automatic where you need to confirm.

 

[boonedocks]

HW2/2.5 is not maxed out yet.

between the way my 2.0 car and 2.5 cars act I would definitely say my HW2.0 can’t handle much more.

 

[rnortman]

You forget that you have a zillion things going on in a roundabout and surface city streets. But you're human so I don't expect you to understand since you do it automatically being a human.

This is rather condescending. I am well aware how complex driving on city streets is, and in fact I have no confidence in Tesla's ability to pull it off at L3+ even with HW3. Also, as I pointed out, nowhere does the order page say anything about roundabouts or the level of reliability to expect from the system, nor any meaningful details about the level of capability -- do you care to respond to those points?

You need to react in a split second in the traffic, 2 Hz is simply not good enough. You need more than that to anticipate vectors properly to provide a smooth response.

I said traffic light and stop sign recognition could run at 2Hz, not all of the AP functions. Definitely the vehicle, pedestrian, and lane detection needs to run at higher frequency, as it already does on HW2. Once a traffic light has been recognized and localized you may want to run the red/green/yellow detection at a higher frequency, but you will know where to look at that point so it is much less work. Trafflic lights and stop signs can even take advantage of map data so that the car doesn't need to be looking all the time nor in its entire field of view.

Sure you can do everything on HW2, but it will be faaaaaaaaaaaar from optimal.

Nothing on the order page mentions an optimal experience. It is my assumption that even on HW3 it will be faaaaaaaaaar from optimal. (I have two fewer "a"s than you had, representing the level of improvement I expect from HW3.)

Automatic city street driving means automatic, not semi-automatic where you need to confirm.

Does it? Where do they define "automatic"? You are reading in your own definition. Also, I never said anything about confirmation, but about supervision. Confirmation is irrelevant to the discussion. They specifically state that they are only promising it will be a supervised system. Supervised doesn't mean you need to confirm, but it does mean you need to pay attention and be ready to take over at any moment. Note that they do not use the term "autonomous" which would have a stronger implication.

 

[emmz0r]

This is rather condescending. I am well aware how complex driving on city streets is, and in fact I have no confidence in Tesla's ability to pull it off at L3+ even with HW3. Also, as I pointed out, nowhere does the order page say anything about roundabouts or the level of reliability to expect from the system, nor any meaningful details about the level of capability -- do you care to respond to those points?

Read up on the threads in Autonomous Driving.

It didn't mention roundabouts, but then you would have a problem with city driving altogether. Then it's no longer automatic.
If you don't expect it to happen on HW3, how do you expect it to work on HW2?

I said traffic light and stop sign recognition could run at 2Hz, not all of the AP functions. Definitely the vehicle, pedestrian, and lane detection needs to run at higher frequency, as it already does on HW2. Once a traffic light has been recognized and localized you may want to run the red/green/yellow detection at a higher frequency, but you will know where to look at that point so it is much less work. Trafflic lights and stop signs can even take advantage of map data so that the car doesn't need to be looking all the time nor in its entire field of view.

Yeah, that's for the actual lights, but traffic lights are a new class of objects that it has to maintain. Thus have it in the model, thus requring more space space and processing.

Nothing on the order page mentions an optimal experience. It is my assumption that even on HW3 it will be faaaaaaaaaar from optimal. (I have two fewer "a"s than you had, representing the level of improvement I expect from HW3.)

It does mention "automatic" which means not confirming, deactivating / activating etc.

Does it? Where do they define "automatic"? You are reading in your own definition. Also, I never said anything about confirmation, but about supervision. Confirmation is irrelevant to the discussion. They specifically state that they are only promising it will be a supervised system. Supervised doesn't mean you need to confirm, but it does mean you need to pay attention and be ready to take over at any moment. Note that they do not use the term "autonomous" which would have a stronger implication.

Confirmation is very relevant to this discussion, which separates automatic from manual operation. NoA is not automatic for instance.

This means it will actually try to handle everything end-to-end.
Tesla's HW3 upgrade next year will usher in Full Self-Driving, improved Neural Net computing abilities

“The team is incredibly excited about the upcoming upgrade for the Autopilot computer. This upgrade allows us to not just run the current neural networks faster, but more importantly, it would allow us to deploy much larger computational and more expansive networks to the fleet. The reason this is important is that it is a common finding in the industry that as you make the networks bigger, the accuracy of their prediction increases with the added capacity. Now we’re currently at a place where we’ve trained large networks that work very well, but are not able to deploy them to the fleet due to computational constraints. So all of these will change with the next iteration of the hardware, and it’s a massive step improvement in computing capability, and the team is incredibly excited to get these networks out there.”

So to put it simply. Yeah of course it's theoretically possible to cram everything into HW2, but it will probably not be a smooth experience. Just as you would not want to play DOOM 3 on a Voodoo 2 card.

 

[rnortman]

Read up on the threads in Autonomous Driving.

I don't read every thread but I read quite a few; care to link to exactly what you're talking about? Or were you still just being condescending and telling me that I need to generally read up on autonomous driving because I don't know what I'm talking about? If so, directing me to the Autonomous Driving thread here is lulz.

It didn't mention roundabouts, but then you would have a problem with city driving altogether. Then it's no longer automatic.
If you don't expect it to happen on HW3, how do you expect it to work on HW2?

You still don't get what I'm saying. I expect what they are currently promising to work on HW3 and maybe on HW2/2.5. I do not ever expect L3+ city driving to happen on either one of them, but they are not currently promising L3+ even on the highway, let alone local roads. I am not referring to Elon's tweetspew -- I'm talking about the actual order description on the order page.

Yeah, that's for the actual lights, but traffic lights are a new class of objects that it has to maintain. Thus have it in the model, thus requring more space space and processing.

And I believe that HW2/2.5 might have enough capacity to do a bad job (<99.999%) at recognizing and "responding to" (which, again, is very vague) traffic lights and stop signs. Frankly this is not that hard. This has been demonstrated many years ago on lesser hardware.

It does mention "automatic" which means not confirming, deactivating / activating etc.

Confirmation is very relevant to this discussion, which separates automatic from manual operation. NoA is not automatic for instance.

Yeah. And? Confirmation or not, if it requires supervision, then they are are free to deliver something which is not reliable enough to operate unsupervised. What matters to how hard this is has little to do with confirmation and everything to do with supervision.

This means it will actually try to handle everything end-to-end.
Tesla's HW3 upgrade next year will usher in Full Self-Driving, improved Neural Net computing abilities

Does it? Where does it say that on the order page? I don't care about Teslarati articles; you're not buying what's in the Teslarati article, you're buying what's on the order page.

 

[BinaryField]

Do these semantics even matter? Nothing about the new FSD description guarantees that you're going to get a useful system, even if the stated goals are 100% achieved to buyers' imaginations (which I don't believe will happen at all in the medium-term). One caveat about the system requiring supervision is that they don't actually have to worry about integrating with your local traffic culture. Look at Navigate on Autopilot. You can technically say that it achieves its promise of on-ramp to off-ramp driving because the car has some logic to do so. In reality, it's completely useless. If I use it in my area, not only will it barely get to where it needs to go, but I will also be rear-ended because of its erratic behaviour.

My interpretation of all of this FSD business is that Tesla is mitigating risk for themselves in delivering any sort of product. They wrote a slick new description of FSD and have no tech demos or developer roadmap. They even tried to fire-sell it in the recent pricing debacle. If they actually knew how to put such a revolutionary product together in short order, shouldn't they be charging the same or more, not less? Shouldn't they be promising the same features or more, not less? Shouldn't they be trumpeting their progress everywhere, not offering shadowy implication and innuendo? I see this as basic sense, but maybe I am wrong.