Elon: FSD Beta tweets

[diplomat33]

Right- which is why just throwing more types of expensive sensors- without any evidence you actually need them- is a poor approach to building autonomous vehicles.

I already proved that the sensors are not too expensive. And yes, there is plenty evidence you need lidar for safe and reliable AVs.

This idea that autonomous driving with cameras+radar+lidar is too costly and not even proven it can do L5 unless you add even more costly sensors is a total myth. We have plenty of evidence that cameras+radar+lidar will work to solve FSD and is not too expensive. The proof is that dozens of companies hare reliable, generalized autonomous driving that is not too expensive (and costs of sensors are coming down) and they are scaling to more and more areas. And all these systems are far more reliable than Tesla's vision-only FSD.

We already have ~$150k Waymo cars that can barely manage a pretty narrow definition of L4 driving- only in a couple specific places- and not on hard routes- if you also have humans on call as backup navigators--- imagine how much more expensive HW they'd need to throw at it to try and brute force their way to L5.

This is total BS and a complete mischaracterization. Waymo does not need to add more sensors to brute force their way to L5. The Waymo 5th Gen sensors are good enough for L5. Software is the limiting factor, not hardware. And it does not "barely manage a pretty narrow definition L4". Waymo Driver is generalized, able to handle really well a wide range of different driving scenarios in parking lots, all types of roads, urban and highway driving. In fact, Waymo Driver can drive autonomously anywhere, Waymo just has not validated safety everywhere yet.

Whereas if you solve vision, you don't need most of that expensive hardware at all.

And if humans had wings, we would not need to build expensive airplanes. The fact is that we have not solved vision. What you are saying is technically true but it relies on a total hypothetical. Why would we pursue a hypothetical like "solving vision" which is unproven when we know we can solve FSD with cameras+lidar+radar and it is not too expensive?

 

[Knightshade]

I already proved that the sensors are not too expensive.

No, you did not.

You parroted a press release from a car show about a future vehicle that:

Is not for sale at any price to the public- but has a claimed lower "cost" in the press release.
Won't even enter limited early testing until late next year.
And when it does will be testing at L4, not L5.

And yes, there is plenty evidence you need lidar for safe and reliable AVs.

Really?

Which safe and reliable L5 vehicle can I purchase today?

This vehicle does not exist, so there is no evidence, and can't be any on what is required until it does.


Further- even if there WAS one, that isn't necessarily evidence it's needed. Aircraft once flew with triple wings, but triple wings are not "needed" for heavier than air flight.

This idea that autonomous driving with cameras+radar+lidar is too costly and not even proven it can do L5 unless you add even more costly sensors is a total myth.

You keep saying that- then failing to show a cheap L5 car at any price with any combo of sensors

We have plenty of evidence that cameras+radar+lidar will work to solve FSD and is not too expensive. The proof is that dozens of companies hare reliable, generalized autonomous driving that is not too expensive (and costs of sensors are coming down) and they are scaling to more and more areas

No, we do not.

First- If they were "reliable generalized" there'd be nothing to "scale" they'd just work everywhere.

Second- Waymo is the only one I know of offering rides in even TWO cities in the US... and as mentioned by their former CEO those cars cost in the range of $130,000-$150,000.

COST, not "a consumer can buy one that cheap"

This is total BS and a complete mischaracterization. Waymo does not need to add more sensors to brute force their way to L5. The Waymo 5th Gen sensors are good enough for L5. Software is the limiting factor, not hardware.

Again this is utter nonsense.

Until you have L5 you can not know what is needed for it.

Waymo doesn't know until they achieve it. Tesla does not know until they achieve it. The fact even AMONG the folks who think they need LIDAR they take different approaches also puts the lie to your claim anybody "knows" what is needed. They're all guessing and hoping they can make whatever guess they made work eventually.


And YOU sure don't know.

And it does not "barely manage a pretty narrow definition L4". Waymo Driver is generalized, able to handle really well a wide range of different driving scenarios in parking lots, all types of roads, urban and highway driving. In fact, Waymo Driver can drive autonomously anywhere, Waymo just has not validated safety everywhere yet.

Waymo still avoids hard left turns even in validated areas. Of which there's like 2 in the whole country- and only small easy areas of those two. And even then not everywhere, and with a tiny fleet.

So again you sound more like a Waymo commercial than anyone presenting facts matching reality.

And if humans had wings, we would not need to build expensive airplanes

Except, we would still need airplanes.

Because our bones are not hollow and we'd be too heavy to fly like birds do.

I'd suggest you quit while you're behind but...too late I guess.

. The fact is that we have not solved vision. What you are saying is technically true but it relies on a total hypothetical. Why would we pursue a hypothetical like "solving vision" which is unproven when we know we can solve FSD with cameras+lidar+radar and it is not too expensive?

because nobody has actually proven you can solve FSD that way.

it's exactly as hypothetical.

If it was solved you'd have an actual generalized L5 vehicle you could buy and have it drive you anywhere a human could.

This does not exist

At any price.

 

[Yelobird]

No, you did not.

You parroted a press release from a car show about a future vehicle that:

Is not for sale at any price to the public- but has a claimed lower "cost" in the press release.
Won't even enter limited early testing until late next year.
And when it does will be testing at L4, not L5.




Really?

Which safe and reliable L5 vehicle can I purchase today?

This vehicle does not exist, so there is no evidence, and can't be any on what is required until it does.


Further- even if there WAS one, that isn't necessarily evidence it's needed. Aircraft once flew with triple wings, but triple wings are not "needed" for heavier than air flight.




You keep saying that- then failing to show a cheap L5 car at any price with any combo of sensors






No, we do not.

First- If they were "reliable generalized" there'd be nothing to "scale" they'd just work everywhere.

Second- Waymo is the only one I know of offering rides in even TWO cities in the US... and as mentioned by their former CEO those cars cost in the range of $130,000-$150,000.

COST, not "a consumer can buy one that cheap"




Again this is utter nonsense.

Until you have L5 you can not know what is needed for it.

Waymo doesn't know until they achieve it. Tesla does not know until they achieve it. The fact even AMONG the folks who think they need LIDAR they take different approaches also puts the lie to your claim anybody "knows" what is needed. They're all guessing and hoping they can make whatever guess they made work eventually.


And YOU sure don't know.




Waymo still avoids hard left turns even in validated areas. Of which there's like 2 in the whole country- and only small easy areas of those two. And even then not everywhere, and with a tiny fleet.

So again you sound more like a Waymo commercial than anyone presenting facts matching reality.






Except, we would still need airplanes.

Because our bones are not hollow and we'd be too heavy to fly like birds do.

I'd suggest you quit while you're behind but...too late I guess.





because nobody has actually proven you can solve FSD that way.

it's exactly as hypothetical.

If it was solved you'd have an actual generalized L5 vehicle you could buy and have it drive you anywhere a human could.

This does not exist

At any price.

Have to agree with all of your counter points. Furthermore I hate when individuals Assume what’s necessary or wrong just because it has never been done or someone does it differently. When something fails I am fine with calling it wrong. Until that point we should accept that it just might be a different approach.

 

[2101Guy]

Won’t be necessary; FSD is very likely solved this year, so no more versions needed beyond 11.

solved AND with vision only. Elon is "extremely confident..would be shocked if FSD is not fully completed and NYC to LA by end of this year". (c)BleachedMale

 

[S4WRXTTCS]

Have to agree with all of your counter points. Furthermore I hate when individuals Assume what’s necessary or wrong just because it has never been done or someone does it differently. When something fails I am fine with calling it wrong. Until that point we should accept that it just might be a different approach.

I just don't want people to lose half their lives trying to break out of prison with a plastic spoon.

Sure it MIGHT work, but there are better ways.

 

[Knightshade]

I just don't want people to lose half their lives trying to break out of prison with a plastic spoon.

Sure it MIGHT work, but there are better ways.

Agreed- so long as the guy with a plastic spoon AND a solid gold fork doesn't tell me he already escaped because because he only digs in a shallow cell in arizona

 

[Supcom]

Agreed- so long as the guy with a plastic spoon AND a solid gold fork doesn't tell me he already escaped because because he only digs in a shallow cell in arizona

The smart guy outsmarts everyone by picking the door lock with the spoon.

 

[Yelobird]

I just don't want people to lose half their lives trying to break out of prison with a plastic spoon.

Sure it MIGHT work, but there are better ways.

Fair thought however most kids over 5 have dug a hole in sand with a spoon to make a hole and know the outcome. How many people have solved FSD that you know of? To say one way is wrong without knowing how to solve the problem is foolish.

 

[Goose66]

Fair thought however most kids over 5 have dug a hole in sand with a spoon to make a hole and know the outcome. How many people have solved FSD that you know of? To say one way is wrong without knowing how to solve the problem is foolish.

As with any technical problem, there are a thousand ways to solve it. In my experience, the hard core engineer certainly can’t see them all, or even more than a few. She’s equipped with a hammer (whatever her particular expertise) and everything looks like a nail. If that engineer is set to the task of finding the solution, they will flounder.

That’s where you need a good project manager to pick one solution based on known technologies, availability of resources, and required timelines and drive the project forward. If you wind up missing the target (solution, cost, or time-wise) then that’s on the project manager. Elon is the project manager here and he has missed the target because of bad choices. I don’t have to know how to implement all of the technology required to solve autonomous driving in order to see his failure.

 

[Knightshade]

That’s where you need a good project manager to pick one solution based on known technologies, availability of resources, and required timelines and drive the project forward. If you wind up missing the target (solution, cost, or time-wise) then that’s on the project manager. Elon is the project manager here and he has missed the target because of bad choices. I don’t have to know how to implement all of the technology required to solve autonomous driving in order to see his failure.

Which is the same failure of literally every other company in this space, some of which have been working on this much longer, with much greater sensor/HW suites, without producing an L5 car either.

It's a hard problem.

 

[Goose66]

Which is the same failure of literally every other company in this space, some of which have been working on this much longer, with much greater sensor/HW suites, without producing an L5 car either.

It's a hard problem.

It took Elon years just to come to the realization that it is a hard problem. In 2015 he considered it solved, while in 2021 he said it is the hardest technical challenge man had faced (or something along those lines).

As far as the work of others, I’m not suggesting that anybody else has got it figured out. I’m just saying I have no more faith in what Tesla and Elon are doing to “solve” L5 driving than I have in others. It’s Elon’s delusion (and/or misrepresentation) of where FSD is and what it will become that’s the most disturbing.

 

[Matias]

It took Elon years just to come to the realization that it is a hard problem. In 2015 he considered it solved, while in 2021 he said it is the hardest technical challenge man had faced (or something along those lines).

As far as the work of others, I’m not suggesting that anybody else has got it figured out. I’m just saying I have no more faith in what Tesla and Elon are doing to “solve” L5 driving than I have in others. It’s Elon’s delusion (and/or misrepresentation) of where FSD is and what it will become that’s the most disturbing.

Yes. And with Tesla the problem is, that they have been all the time selling it. 2019 they dialed it back little, but before that it was "pending regulatory approval".

 

[diplomat33]

Which is the same failure of literally every other company in this space, some of which have been working on this much longer, with much greater sensor/HW suites, without producing an L5 car either.

You keep pointing out that nobody has L5 yet. That is a strawman. Of course, nobody has L5 yet. But these companies with 30+ sensors, like Waymo, Cruise, Zoox, Argo, Baidu, Aurora, Motional, have made a lot of progress towards that goal. They all have autonomous driving that can be deployed in some areas without supervision. That is progress. We cannot simply ignore that progress just because they have not achieved L5 yet.

As I stated before, the companies with 30+ sensors are focused on reliability first and then scaling out. Tesla is scaling out first and then working on solving reliability. Both are logical approaches. The companies like Waymo have made great progress on the reliability part as they have AVs that can drive without supervision in some areas. It is an open question how quickly they can scale. Tesla has no problem with scaling but it is an open question if they can solve the reliability issues. Obviously, nobody has L5 so we cannot say conclusively which approach will achieve L5 first. You can certainly argue that you think the Tesla approach will achieve L5 first. But when you spread misinformation like saying AVs with 30+ sensors costs $250k or saying that Waymo can barely do limited L4, you are arguing in bad faith.

But personally, I don't think we need to achieve 100% L5 because I think L4 that covers 99.9999% of the ODD of L5 would be good enough for most people. So I think L5 is an aspirational goal but not something we need to actually achieve. As long as we achieve L4 that works in 99.9999% of the ODD of L5, that should be good enough IMO.

 

[diplomat33]

Perhaps if you believed Elon, you probably thought Tesla would simply deploy a large fleet with FSD Beta that works everywhere like L5 and then just train on a ton of data and in like 6 months to a year, FSD Beta would achieve L5 reliability. The reality is that was never going to happen. Actual reliable L5 is much harder than that. Elon's L5 claims were always a fantasy. Even if you get great perception with vision-only, which that by itself takes a long time to achieve, there are still millions of edge cases that require smart decision-making and every US city is different, with different roads, signs, different road rules and different types of drivers etc... So you were never going to just train a lot of data for like a year and then poof, FSD Beta is L5.

But the fact is that not too long ago we did not even have any driverless robotaxis. Then, driverless robotaxis were basically a novelty. Now, we have driverless robotaxis in some areas that the public can use and the list is growing each year. Like I said, it is exciting progress. But it will take time to train, test and validate before deployment to make sure that the driverless robotaxis are safe in the new area where they are driving. You cannot just assume your FSD that works great in one area, will work the same everywhere, since every area is so different. Personally, I see L4 scaling over time until it is eventually very close to L5. I also don't think one single company will scale to the entire US. More likely, it will take the combined efforts of multiple companies each scaling in different areas, to eventually cover most or all US cities.

 

[2101Guy]

who are the other CEO's who are repeatedly, intentionally, and continually making false timeline/deadline claims of L5 autonomy?

I'll wait...

 

[Matias]

who are the other CEO's who are repeatedly, intentionally, and continually making false timeline/deadline claims of L5 autonomy?

I'll wait...

I think it is more important to ask who else is selling it. It is not a problem if I say I own imaginary bridge, but it is if I sell it to you.

 

[n.one.one]

I'm glad he didn't say 2 weeks.

 

[S4WRXTTCS]

It took Elon years just to come to the realization that it is a hard problem. In 2015 he considered it solved, while in 2021 he said it is the hardest technical challenge man had faced (or something along those lines).

As far as the work of others, I’m not suggesting that anybody else has got it figured out. I’m just saying I have no more faith in what Tesla and Elon are doing to “solve” L5 driving than I have in others. It’s Elon’s delusion (and/or misrepresentation) of where FSD is and what it will become that’s the most disturbing.

I think both of these things are make believe.

The Elon didn't know any better is massively charitable to him, and fails to realize that it was really used as a marketing stunt. They didn't just fail to achieve autonomous driving at any level (L3 to L5), but they also failed to have AP/NoA work at a level where the vast majority (95% or better) felt confident in using it and enjoyed it.

L5 is a complete distraction because it's a fantasy. L5 tends to come up in discussions when the pro-Tesla person wants to discredit the real progress achieved by companies in the autonomous industry. They know all the progress has been made in the L3, and L4 space.

Tesla sold science fiction to consumers for years using the threat that the price would be much higher later on, and that you need to buy in today for the low price of $12K. Of course when the customer comes to TMC we tell them not buy in because there isn't actually any promise or timeframe for autonomous driving.

 

[dramsey]

I'm glad he didn't say 2 weeks.

View attachment 834211

Man, I sure hope 10.13 is a massive improvement over 10.12. Please, please, please let it make much more intelligent choices about what lane to be in for upcoming turns!

 

[bobsmy]

Autonomy day April 2019: Elon Musk said "that the automaker will deliver a car without a steering wheel within 2 years"
July 2019:


If we make all cars with FSD package self-driving, as planned, any such Tesla should be worth $100k to $200k, as utility increases from ~12 hours/week to ~60 hours/week

Elon Musk

@elonmusk
·
Jul 16, 2019