The Wife's Test: We've had FSD since 2019 it's only been worth using for about a year now IMO. I love it. I can enjoy the day, listen to music and the car drives me where I want to go. It drives great in heavy traffic, gets me to my destination much faster than I could. Jeannine
Trolley problem is minimising accidents against people who r not inside a car. Car will rather prefer to crash on another car than crash against a person outside of car
I think the Trolley Problem in the case of autonomous vehicles is a pointless question. The AI will do its best to not get in an accident. It wont choose between who it should kill it will look for the best way to not get in an accident. If that means take option two at that point in time it will take option two. If that ends up hurting more people but was the correct decision to minimize the chance of an accident it was correct. The problem with the trolley problem is that is assumes the decision maker has knowledge of the outcome of its decision and has the time to make the decision. In the case of a car accident you are not going to have time and knowledge enough of the situation. The person or AI making the decision will do what is best with the knowledge it has. If somehow it has two choices that both lead to an accident then most intelligences, AI or human will probably try to avoid both. Or just not make a decision.
@@danharold3087 I guess cars could use BLE or something similar to detect if humans is around. And if so, they can probably find an estimated distance, and lower the speed whenever they are close. Also, they could have a proximity app that talks to cars around and reveal if they are close to the location of the car. Interesting future.
@@jsjs6751 This needs to be done in the framework of V2X/V2V, It needs to be standardized. Cars need to communicate like airplanes. Each reports a position and a vector. Where they are what direction and speed. That allows cars to arbitrate, who does what to prevent a collision or close encounter.
Tesla will upgrade HW3's if it can't make FSD work on it. My expectation is that shoehoring a working FSD into HW3 while possible is going to be a massive expense going forward version after version. I expect you will get your HW4 and/or an incentive to upgrade.
This is my understanding... The owners who purchased FSD with their V3 hardware vehicle will have it installed free. Those who purchased FSD after buying their car will miss out on the 'free' upgrade but can still have it done at a cost (undisclosed at this time). As with a lot of grey regions like this one, I expect Tesla will likely bow to pressure from owners to make it free across the board. If Tesla can afford it, it is a win-win for both Tesla and their customers.
@@jimparr01Utube Free upgrade is best for Tesla. Anybody who bought and anybody who will buy FSD whose cars have HW3 Paid or pay more than the cost of hw4. Taking out rate was 15% and potentially 50% of hw3 will join FSD family. What Tesla will not lose anything by give them a fee up. Not for subscription users and auto pilot users. They will pay $8000.
I listened to this portion of the conference call multiple times and here is my understanding: If you ordered FSD at the time you purchased the car and you have HW3, then you will receive a free self driving computer upgrade that will allow your car to operate in unsupervised FSD mode, providing that it is determined that HW3 doesn’t allow for the level of safety necessary to do unsupervised FSD. Note that no one on the call specifically said that the upgrade would be to HW4, so it could be to AI5 depending on the time frame involved when Tesla reaches a decision.
I had never considered the 'trolley' problem as it pertains to SFD until you mentioned it here. This looks like a gnarly issue - at least in the short term - and I agree that in the US especially, the drive to litigate before meditate requires some solid perspective on how to deal with such events. I like your suggestion to think of FSD as a quasi human, but there will be massive legal pushback on this - at least for a while - and at the very least by the Ambulance chasers. Great video. Thank you Sir.
As a former autopilot designer, I am curious about the architecture of the FSD computer and how it will act in the event of a HW failure. I have heard it is dual channel, but no words on expected failure rates and what happens to the vehicle when a failure occurs. Typically, a dual channel system can not be fail-active, but only fail-passive. This means it will stop driving after a single critical failure. Any info from Tesla engineers would be nice.
Good episode. Thank you. I think Optimus will be huge. I break it into (home use/factory use/farm use). I think it will do it all, but at different stages in its development. I think Boston Dynamics could compete now that they're partners with an AI. I don't know if they can make that jump to production like I know Tesla can.
I live in California (my home state) so I'm excited that I may buy a model y or model 3 at my Tesla car dealership next year. Tesla will take a hybrid approach for their robotaxi. Which is you can go to your Tesla car dealership to buy one for personal car ownership or use an app to use a robotaxi. Tesla will be unsupervised level 4 by next year. I'll go to Etsy to buy a 2025 Tesla wall calendar. Tesla's robotaxi will be you can travel the entire state like California (unsupervised) however if you go outside the state go to a next door state like Nevada you have to supervise as long as you pay attention by putting your eyes on the road with 100% hands free driving.
First cut at the Trolley problem: Provide a software switch between . Protect life of car occupants at all cost to others when injury to occupants is imminent. Or . Minimize loss of life to all individuals, both internal and external
There is a auto analysis company that works with the suppliers that claim parts for the "out of the box" mini model Y will begin in first quarter. My fingers are crossed. That's from Autoline Detroit.
@@wolfgangpreier9160 There is NO evidence that Ford has the Self Driving solved. Tesla is lot more closer to it than ANY OTHER. Waymo is a system that is on RAILS and don't deviate from where it can go. It is FENCED in.
Elon likes to say "stuff" and "stuff might happen, but sometimes after putting $50k down on a Roadster 2 you get the comment, about cherries on top and we're in the final stages of design - only 8 years late. My 24 Y was initially supposed to have 4680 cells, but that didn't happen.
"While not necessary it may become the regulatory requirement for autonomous driving" Yes it is necessary. As stated in the laws and regulations. Who told you it would not be?
Legal ‘trolly problem’ for those supervising / sat in the drivers seat! If they have a driving licence? If they don’t have a license? Will there be such a thing as a ‘supervisory’ license? Legal and regulatory enforcement in case of accidents causing property destruction and or death of human / animal. Could a seemingly abrupt action by a Tesla under FSD (in a trolly problem scenario) appear random/wrong to the supervisor (who interrupts said action and creates a different / worse outcome) legal and regulatory process likely needs to consider a human/FSD flip (without supervision / human intervention even possible) in order to clarify responsibility - an unshakable black box recorder is absolutely needed. For realtime mapping and traffic management a Starlink mini per FSD vehicle is likely mandatory as will be space x needing more bandwidth allocated to any and all brands - the sharing of live data will enable (amongst other outcomes) fewer accidents and provide for updates wherever as well as a route to black box health integrity (if not last moment data transfer).
As you say John it's a very exciting time to be alive with possibly the industrial revolution as being the only comparable time of such big transitions and changes coming. One question I have which no one else has been able to answer is that if we have this abundant labor robotic force, which will inevitably end up taking an awful lot of jobs, how will people pay for this additional abundance if they have no job to buy these houses and services, unless of course there is a universal income. The Star Trek Utopia would be lovely but knowing mankind's natural greed, I have my concerns. Thoughts?
Since it’s all neural nets, the only way to tell the car what to do in the trolley problem is through simulations. This is the rarest of edge cases in other words. Personally I hope the car prioritizes saving the driver, otherwise Tesla owners may be reluctant to purchase FSD. Maybe we can opt for FSD version 1 which will try to save the driver or version 2 which will coolly save the most humans even if it has to kill its owner.
26:44 The rules of the roads are the same across Europe. The liability always reside in the driver that didn’t respect the rules. In case of autonomous car, the company will be responsible but should be treated like any other human driver unless it is proven that this is a recurring problem of the fleet.
"The rules of the roads are the same across Europe" Mostly similar but not the same, no. "In case of autonomous car, the company will be responsible but should be treated like any other human driver unless it is proven that this is a recurring problem of the fleet." Cortrect, its a simple insurance problem. Of course the companies must insure their cars. They are not the Bundesheer.
The company will have their own insurance and legal services. No sane human being would take responsibility for an autonomous car they happened to sit in but have no control of. And no, the rules are not the same all over Europe. There are minor differences that the autonomous car must be able to handle.
So what is the situation for Australian Tesla owners who mostly have HW3? FSD has never been released here so there has never been a chance to evaluate it.
We will get V13+ once the safety data is thrown in the faces of the people ruining this great country. My guess is end of 2026 at the latest. I find enhanced autopilot very handy with flaws around Melbourne but even better on long drives to Sydney/Adelaide. I realise an approximate 2 year wait requires patience but there is zero chance we aren't getting FSD, it's too safe for Albo et al to stand in its way, please mark my words in about 2 years oi oi oi
I think Tesla is going fork FSD to HW3 and HW4. HW3 will always be 'supervised' but you will be able to buy it or subscribe. This should be at a lower cost . HW4 will be supervised, until it isn't. The cost for unsupervised will be substantially higher.
Long term forks are massively expensive. They act as a training time multiplier. It will be less expensive to build a HW4.1 layout to go into HW3 cars.
What do you think it means for Tesla FSD to have longer miles between interventions than the human driver? I don’t understand this. Can you do a video segment on this?
First to the key. First to the egg! @Tesla 'solving' FSD needs them to have their big pants on! I ask that the make kits, upgrade ALL Teslas ever made to FSD capable, for free, irrespective of the current owner’s FSD status! They need ‘fleet at scale’ to meet their own mission, maintain momentum, first to the egg - so I ask Tesla to do the right thing!
I think the issue (in the US least) is that we are nearing saturation on early adopters of EVs. There are some never EV people out there... don't even bring up FSD. It's equivalent to the "I will never buy an import" attitude that is still around from the 80s. Unless there's a mass manufacturing shift (like there was to automatic from manual trans) it will be 30 years for the hardliners to die off (literally).
As you pointed out, if there is saturation it is temporary. When people see EVs are clearly the more economical, more dependable cars many more will buy,
I don’t understand how the comparison can be made between supervised FSD crashes/mile versus human driving. It would have to be fully unsupervised to be comparable, otherwise you have two systems teamed up together, it’s normal that the numbers are better?
It drives on roads on the right side predominantly and follows the signs it knows about. That is all. That works in every LHD country. On- and off-road as long as they are roads for the underlying nevigation.
Longer miles between interventions than a human driver? My miles per intervention when I’m driving are 0. So FSD already does that. What the heck does this mean?
Do you not find it strange that California is even a consideration after Musk dedicates all over them? Odd, they pick a couple of states that typically don't have much inclement weather, especially snow, freezing rain, etc. If this software has
What Musk and you are suggesting is that the States surrender sovereignty, not already authorized by the Constitution of These united States of America, to the central government. In a federal system the States and the People are the sovereigns not the Central government.
In the question and answer session, Elon discussed the possibility of HW3 being inadequate. He's been telling his fans that HW3 is enough. It sounds like he's been too optimistic. There are no interventions for human drivers, so that statement is senseless.
To upgrade a HW3 car to HW4 would not only require a new computer, but a change to all cameras. Tesla initially said, when they introduced the HW4 computer, that there was no (easy?) upgrade path for HW3 computers, due to physical dofferences.
@@mikeoberg1 just repeating what Elon said on the call - that the cameras were sufficient and would just need to replace the (modular) computer. He specifically said the cameras wouldn't need to be replaced. Time will tell
My suspicion is that CA FSD approval will not come quickly. Does NHTSA have authority over road safety decisions in the US? I think it does and that TX won't be able to approve FSD in TX. The ACM has investigated autonomous vehicles and is negative about the prospects. I think it should be done with lidar and radar. The hardware cost has been high, but electronics gets cheaper over time. Also, in the US it should be possible to combine sensor suites with map data to have geofenced solutions. I've done enough programming on hard problems to know that the solution seems a lot easier for the first 90%. Also there are problems which computers won't solve. 3 ton vehicles driving 80 MPH is a scary proposition for highway driving. It would never be unsupervised with me as driver, assuming I can stay alert. A sleeping driver would be more dangerous than anybody's autonomous solution. If Elon hadn't dedicated himself to fascism, I wouldn't be against using supervised FSD.
We don't want it here in California until it's tested in Texas for at least a year. Elon is putting too much pressure on engineers to ship half baked products. Only real world safety data matters.
The Wife's Test: We've had FSD since 2019 it's only been worth using for about a year now IMO. I love it. I can enjoy the day, listen to music and the car drives me where I want to go. It drives great in heavy traffic, gets me to my destination much faster than I could. Jeannine
Highway or city streets? I stopped using mine
Trolley problem is minimising accidents against people who r not inside a car. Car will rather prefer to crash on another car than crash against a person outside of car
I think the Trolley Problem in the case of autonomous vehicles is a pointless question. The AI will do its best to not get in an accident. It wont choose between who it should kill it will look for the best way to not get in an accident. If that means take option two at that point in time it will take option two. If that ends up hurting more people but was the correct decision to minimize the chance of an accident it was correct.
The problem with the trolley problem is that is assumes the decision maker has knowledge of the outcome of its decision and has the time to make the decision. In the case of a car accident you are not going to have time and knowledge enough of the situation. The person or AI making the decision will do what is best with the knowledge it has. If somehow it has two choices that both lead to an accident then most intelligences, AI or human will probably try to avoid both. Or just not make a decision.
That is the challenge. Attempting to avoid any accident leads to the most casualties.
V2V part of V2X will allow every car to know where every other car is, their direction and speed. USDOT is now considering it.
@@danharold3087 but that is not the trolley problem. It's about minimizing pedestrian injuries.
@@danharold3087 I guess cars could use BLE or something similar to detect if humans is around. And if so, they can probably find an estimated distance, and lower the speed whenever they are close.
Also, they could have a proximity app that talks to cars around and reveal if they are close to the location of the car.
Interesting future.
@@jsjs6751 This needs to be done in the framework of V2X/V2V, It needs to be standardized.
Cars need to communicate like airplanes. Each reports a position and a vector. Where they are what direction and speed. That allows cars to arbitrate, who does what to prevent a collision or close encounter.
WOW ! You have obviously been thinking about this for a while. Nicely done.
Did I hear you correctly that Tesla will upgrade my Model 3's CPU Version 3 to Version 4 since I already purchased FSD?
Tesla will upgrade HW3's if it can't make FSD work on it. My expectation is that shoehoring a working FSD into HW3 while possible is going to be a massive expense going forward version after version. I expect you will get your HW4 and/or an incentive to upgrade.
This is my understanding...
The owners who purchased FSD with their V3 hardware vehicle will have it installed free. Those who purchased FSD after buying their car will miss out on the 'free' upgrade but can still have it done at a cost (undisclosed at this time).
As with a lot of grey regions like this one, I expect Tesla will likely bow to pressure from owners to make it free across the board. If Tesla can afford it, it is a win-win for both Tesla and their customers.
@@jimparr01Utube
Free upgrade is best for Tesla. Anybody who bought and anybody who will buy FSD whose cars have HW3 Paid or pay more than the cost of hw4.
Taking out rate was 15% and potentially 50% of hw3 will join FSD family.
What Tesla will not lose anything by give them a fee up. Not for subscription users and auto pilot users.
They will pay $8000.
I listened to this portion of the conference call multiple times and here is my understanding: If you ordered FSD at the time you purchased the car and you have HW3, then you will receive a free self driving computer upgrade that will allow your car to operate in unsupervised FSD mode, providing that it is determined that HW3 doesn’t allow for the level of safety necessary to do unsupervised FSD. Note that no one on the call specifically said that the upgrade would be to HW4, so it could be to AI5 depending on the time frame involved when Tesla reaches a decision.
If necessary, yes.
I had never considered the 'trolley' problem as it pertains to SFD until you mentioned it here. This looks like a gnarly issue - at least in the short term - and I agree that in the US especially, the drive to litigate before meditate requires some solid perspective on how to deal with such events. I like your suggestion to think of FSD as a quasi human, but there will be massive legal pushback on this - at least for a while - and at the very least by the Ambulance chasers.
Great video. Thank you Sir.
As a former autopilot designer, I am curious about the architecture of the FSD computer and how it will act in the event of a HW failure. I have heard it is dual channel, but no words on expected failure rates and what happens to the vehicle when a failure occurs. Typically, a dual channel system can not be fail-active, but only fail-passive. This means it will stop driving after a single critical failure. Any info from Tesla engineers would be nice.
I would say the same. Fail = stop.
Next year, you will want to go to watch a launch in Texas again, just to use unsupervised FSD!
My little brain is hurting, but a great video, so much to to take on board, but exciting stuff, love it.
Tesla AI 😍
Good episode. Thank you. I think Optimus will be huge. I break it into (home use/factory use/farm use). I think it will do it all, but at different stages in its development. I think Boston Dynamics could compete now that they're partners with an AI. I don't know if they can make that jump to production like I know Tesla can.
I live in California (my home state) so I'm excited that I may buy a model y or model 3 at my Tesla car dealership next year. Tesla will take a hybrid approach for their robotaxi. Which is you can go to your Tesla car dealership to buy one for personal car ownership or use an app to use a robotaxi. Tesla will be unsupervised level 4 by next year. I'll go to Etsy to buy a 2025 Tesla wall calendar. Tesla's robotaxi will be you can travel the entire state like California (unsupervised) however if you go outside the state go to a next door state like Nevada you have to supervise as long as you pay attention by putting your eyes on the road with 100% hands free driving.
First cut at the Trolley problem:
Provide a software switch between
. Protect life of car occupants at all cost to others when injury to occupants is imminent.
Or
. Minimize loss of life to all individuals, both internal and external
There is a auto analysis company that works with the suppliers that claim parts for the "out of the box" mini model Y will begin in first quarter. My fingers are crossed. That's from Autoline Detroit.
FSD is like FUD. I just don’t trust it while in Beta. It needs to be ready.
It will always be in BETA, sort of. Until it reaches 99.999% or better, Tesla is going to train the FSD.
You would take about 6 lifetimes to be in an at fault accident while supervising FSD 😊
That is perfectly ok. Ford will soon come out with autonomous vehicles. Really. You must believe.
@@wolfgangpreier9160 There is NO evidence that Ford has the Self Driving solved. Tesla is lot more closer to it than ANY OTHER. Waymo is a system that is on RAILS and don't deviate from where it can go. It is FENCED in.
@@wolfgangpreier9160 🤣😂😂🤣
Elon likes to say "stuff" and "stuff might happen, but sometimes after putting $50k down on a Roadster 2 you get the comment, about cherries on top and we're in the final stages of design - only 8 years late. My 24 Y was initially supposed to have 4680 cells, but that didn't happen.
I guess either you are patient or you buy VW. Much better cars, cost much more. Guaranteed fraud included.
Great guide and explanation of Tesla's path forward. Should have listened to you when you predicted the stock would drop after the we robot event.
Why doesn't Tesla put radar back in so that the car can see through fog, darkness, and sun glare to make it much safer than humans can ever be?
Texas passed their autonomus driving law in 2017. The next autonomus driving task force is in December.
Too bad FSD is incapable of 100% autonomous FSD.
Only the CT has two sreering motors upfront. While not necessary it may become the regulatory requirement for autonomous driving
"While not necessary it may become the regulatory requirement for autonomous driving" Yes it is necessary. As stated in the laws and regulations. Who told you it would not be?
Legal ‘trolly problem’ for those supervising / sat in the drivers seat! If they have a driving licence? If they don’t have a license? Will there be such a thing as a ‘supervisory’ license? Legal and regulatory enforcement in case of accidents causing property destruction and or death of human / animal.
Could a seemingly abrupt action by a Tesla under FSD (in a trolly problem scenario) appear random/wrong to the supervisor (who interrupts said action and creates a different / worse outcome) legal and regulatory process likely needs to consider a human/FSD flip (without supervision / human intervention even possible) in order to clarify responsibility - an unshakable black box recorder is absolutely needed. For realtime mapping and traffic management a Starlink mini per FSD vehicle is likely mandatory as will be space x needing more bandwidth allocated to any and all brands - the sharing of live data will enable (amongst other outcomes) fewer accidents and provide for updates wherever as well as a route to black box health integrity (if not last moment data transfer).
As you say John it's a very exciting time to be alive with possibly the industrial revolution as being the only comparable time of such big transitions and changes coming. One question I have which no one else has been able to answer is that if we have this abundant labor robotic force, which will inevitably end up taking an awful lot of jobs, how will people pay for this additional abundance if they have no job to buy these houses and services, unless of course there is a universal income. The Star Trek Utopia would be lovely but knowing mankind's natural greed, I have my concerns. Thoughts?
Since it’s all neural nets, the only way to tell the car what to do in the trolley problem is through simulations. This is the rarest of edge cases in other words. Personally I hope the car prioritizes saving the driver, otherwise Tesla owners may be reluctant to purchase FSD. Maybe we can opt for FSD version 1 which will try to save the driver or version 2 which will coolly save the most humans even if it has to kill its owner.
26:44 The rules of the roads are the same across Europe. The liability always reside in the driver that didn’t respect the rules. In case of autonomous car, the company will be responsible but should be treated like any other human driver unless it is proven that this is a recurring problem of the fleet.
"The rules of the roads are the same across Europe" Mostly similar but not the same, no.
"In case of autonomous car, the company will be responsible but should be treated like any other human driver unless it is proven that this is a recurring problem of the fleet." Cortrect, its a simple insurance problem. Of course the companies must insure their cars. They are not the Bundesheer.
The company will have their own insurance and legal services.
No sane human being would take responsibility for an autonomous car they happened to sit in but have no control of.
And no, the rules are not the same all over Europe. There are minor differences that the autonomous car must be able to handle.
So what is the situation for Australian Tesla owners who mostly have HW3? FSD has never been released here so there has never been a chance to evaluate it.
We will get V13+ once the safety data is thrown in the faces of the people ruining this great country. My guess is end of 2026 at the latest. I find enhanced autopilot very handy with flaws around Melbourne but even better on long drives to Sydney/Adelaide. I realise an approximate 2 year wait requires patience but there is zero chance we aren't getting FSD, it's too safe for Albo et al to stand in its way, please mark my words in about 2 years oi oi oi
I think Tesla is going fork FSD to HW3 and HW4. HW3 will always be 'supervised' but you will be able to buy it or subscribe. This should be at a lower cost . HW4 will be supervised, until it isn't. The cost for unsupervised will be substantially higher.
Long term forks are massively expensive. They act as a training time multiplier. It will be less expensive to build a HW4.1 layout to go into HW3 cars.
What do you think it means for Tesla FSD to have longer miles between interventions than the human driver? I don’t understand this. Can you do a video segment on this?
First to the key. First to the egg! @Tesla 'solving' FSD needs them to have their big pants on! I ask that the make kits, upgrade ALL Teslas ever made to FSD capable, for free, irrespective of the current owner’s FSD status! They need ‘fleet at scale’ to meet their own mission, maintain momentum, first to the egg - so I ask Tesla to do the right thing!
I think the issue (in the US least) is that we are nearing saturation on early adopters of EVs. There are some never EV people out there... don't even bring up FSD. It's equivalent to the "I will never buy an import" attitude that is still around from the 80s. Unless there's a mass manufacturing shift (like there was to automatic from manual trans) it will be 30 years for the hardliners to die off (literally).
In 30 years FSD will be mandatory or some form of FSD
Never car, never inside toilet, never cell phone. Yeah, never-people are either near death or self delusional.
As you pointed out, if there is saturation it is temporary. When people see EVs are clearly the more economical, more dependable cars many more will buy,
Yes, nobody will want to buy faster, safer, cheaper cars. NOBODY.
Everybody will want to buy stinking, expensive, failure prone slow poke fossil burners instead.
I don’t understand how the comparison can be made between supervised FSD crashes/mile versus human driving. It would have to be fully unsupervised to be comparable, otherwise you have two systems teamed up together, it’s normal that the numbers are better?
This comparison is nonsense and everyone should see this at a first glance. But it helps selling the story.
@@EinzigfreierName True, Humans can not be as safe as FSD, that is simply not possible.
Yes we can compare. Add up the miles driven in FSD without accidents and now add them up in a - idk - Ford Pinto? Now compare those numbers.
@@wolfgangpreier9160 That's not what they are comparing. They are comparing humans vs. humans + FSD.
I would like to know how FSD would do in others countries that are kind of poor with not the best infrastructure and road regulations. Thanks
It drives on roads on the right side predominantly and follows the signs it knows about. That is all. That works in every LHD country. On- and off-road as long as they are roads for the underlying nevigation.
Longer miles between interventions than a human driver? My miles per intervention when I’m driving are 0. So FSD already does that. What the heck does this mean?
"My miles per intervention when I’m driving are 0" Exactly. You can not win. Just give up.
❤im thinking this could be broke up in 4 parts. 30 minutes to 1 hour each. Way too much information at once. Maybe you also have Asperger's too😊
The Tesla stock is as if the Bears are taking a breather just to start their next aunslaut on the gullible small-time shareholders.
Do you not find it strange that California is even a consideration after Musk dedicates all over them? Odd, they pick a couple of states that typically don't have much inclement weather, especially snow, freezing rain, etc. If this software has
It is not California. It is Palo Alto.
What Musk and you are suggesting is that the States surrender sovereignty, not already authorized by the Constitution of These united States of America, to the central government. In a federal system the States and the People are the sovereigns not the Central government.
?
how can you just gloss over the HW3 upgrade. the only reason I came
In the question and answer session, Elon discussed the possibility of HW3 being inadequate. He's been telling his fans that HW3 is enough. It sounds like he's been too optimistic. There are no interventions for human drivers, so that statement is senseless.
just heard the hedging part, only IF you purchased FSD with the vehicle (not subscription), that's a finer caveat (price is likely
To upgrade a HW3 car to HW4 would not only require a new computer, but a change to all cameras. Tesla initially said, when they introduced the HW4 computer, that there was no (easy?) upgrade path for HW3 computers, due to physical dofferences.
@@mikeoberg1 just repeating what Elon said on the call - that the cameras were sufficient and would just need to replace the (modular) computer. He specifically said the cameras wouldn't need to be replaced. Time will tell
So annoying hearing you read from a a number of paper and repeat the same sentence 4 times.
Why not just read the paper?
My suspicion is that CA FSD approval will not come quickly. Does NHTSA have authority over road safety decisions in the US? I think it does and that TX won't be able to approve FSD in TX. The ACM has investigated autonomous vehicles and is negative about the prospects. I think it should be done with lidar and radar. The hardware cost has been high, but electronics gets cheaper over time. Also, in the US it should be possible to combine sensor suites with map data to have geofenced solutions. I've done enough programming on hard problems to know that the solution seems a lot easier for the first 90%. Also there are problems which computers won't solve. 3 ton vehicles driving 80 MPH is a scary proposition for highway driving. It would never be unsupervised with me as driver, assuming I can stay alert. A sleeping driver would be more dangerous than anybody's autonomous solution. If Elon hadn't dedicated himself to fascism, I wouldn't be against using supervised FSD.
We don't want it here in California until it's tested in Texas for at least a year. Elon is putting too much pressure on engineers to ship half baked products. Only real world safety data matters.
@@litchips Without real world testing it would be impossible to make progress at a sustainable rate.
Its not California. Its Palo Alto.