But this does ring true, and the come-to-Jesus moment on Tesla’s valuation will be when people realize this is the case. If they had a legitimate shot at being the first true players in self-driving, their valuation (which currently prices them on par with something like the sum of the top three automakers...) isn’t insane. But if it really is just lane-keep assist with intelligent cruise control, I’m pretty sure the Mercedes S Class had that... back before 2010?
Tesla is the only participant that has a true path to self driving technology. Legacy car makers are Goliaths set in the past. They're stuck to the annual release cycle and can't even over the air updates and a touch screen on par with an iPad from 8 years ago.
Legacy automakers have failed to deliver on electric cars for decades. Musk talked about this. There was nothing special about the time he entered the market with Tesla. Any one of the auto makers could (and still theoretically can) develop an appealing electric car. They're simpler, cleaner and more efficient. But not a single one has been able to deliver a reasonable product. They're subsidizing the price of electric and losing 7-10k per [0] and still cannot produce a compelling product.
The legacy car companies are basically blackberry, as someone has finally shown there exists a way to create and produce a compelling car outside of the legacy participants.
Tesla actually has a path forward on self driving. They have the infrastructure, real driver data, and millions of cars on the road that can accept over the air updates. Waymo has a few Taxis in a few cities.
Tesla has just over a million cars on the road. Waymo has approximately 100,000. 100000 vehicles that have proven the ability to reach L4 autonomy, unlike tesla's.
And waymos approach isn't too be a car manufacturer, but an Uber competitor. So fewer vehicles is fine.
> They're subsidizing the price of electric and losing 7-10k per [0]
In my opinion, the problem is Tesla and Elon refuse to acknowledge the value of LIDAR. A fused sensor approach is the only way we're going to get full autonomy. I was in this field in 2007 on the DARPA Urban challenge so maybe the state of the art indicates otherwise now however, from the NHTSA investigations into crashes from Tesla's autopilot all of them would have been fixed with even basic LIDAR.
Lidar adds complexity and domain that's not available to humans. Humans drive without lidar, so presumably we can train a neural net to drive without it as well.
Lidar is almost like adding a hand crafted dimension to the decision making process. It makes the system more explainable, but doesn't necessarily mean it will help achieve better results. I look at ML in games like chess and go. Originally, they had a set of hand crafted features, like the concept of "eyes" in go. But later iterations removed those and the network performed better. I think lidar is the same thing where it'll add unnecessary complexity and ultimately hurt in learning.
Cost. Tesla has to sell cars to price-sensitive consumers. Adding a LIDAR would likely increase the price significantly. Given their business model, they basically have to argue that LIDAR is not necessary.
I agree that when Tesla made their first promises of delivering FSD HW (just add SW a little bit later) reasonably priced solid state lidars were not available. So a story had to be made 'humans have eyes and they can drive, lidar is not needed'. Super hard to back now when they sold FSD to so many customers still waiting for the SW update.
I think this right here is the issue. They've backed themselves into the corner of 'We Don't need LIDAR'. When we were scrapping together on a 75k budget we bought budget lidar with a 180 degree linear sweep and it cost ~5k, 14 years ago, from SICK and it still was better than a modern 2020 model Tesla is at detecting a white van crossing the road in broad daylight.
I’d argue Waymo is not a competitor and offers no ADAS. They operate autonomous vehicles in confined areas afaik. Driving in traffic, interacting with humans is a different story.
Well full automatic driving is more advanced than driver assistance, and Tesla is claiming that the current system will be scaled to full autonomous driving which is what the original article is about.
The waymo car is the only actively-used car with Lidar anyway as far as I’m aware, so if you exclude Waymo then your statement is self fulfilling no matter how good or bad Tesla’s ADAS is!
The point i wanted to express is that although LIDAR could be an improvement it is not per se the „way to go“ with ADAS today. Please consider that some, eg Audi, offer ADAS using lidar and it is not nearly as capeable as Teslas system.
Costs matter.
I am very certain that Waymo has no autonomous vehicle today that mist people would be ok to go around their neighbourhood driverless. „Full autonomy“- socially acceptable- does not exist today. Whether it can be done, with or without Lidar, remains to be seen.
> You attack legacy automakers but then go on to discuss Waymo
Paragraphs 1-3 discuss legacy automakers. Only paragraph 4 references Waymo.
> You talk about OTA updates as if that’s some huge technological moat?
Of course they're not a technological moat. These auto makers are hundreds of thousands of employees. They're likely an organizational nightmare and stuck to the way things have been done. The fact that they don't have over the air updates or have to iterate on an annual cadence is evidence of something deeply flawed in their organizational structure. And I'm not confident that they can overcome it.
> You talk about OTA updates as if that’s some huge technological moat? Hell, cars have been receiving very frequent software updates for over a decade.
I think OTAs are a moat but highlight an important incentive difference. Traditional automakers want to sell you next year's model of the car. Tesla seems like they want your car to get better over time.
Perhaps it is organizational cruft or the dealer network but consider even the simple case of adaptive cruise control systems in cars today that have radar + cameras. Most of those systems bounce between lanes and can't even stay centered well. This is something that could be a software fix but you'll have to wait for the next N models when they redesign the car to see it.
Infotainment systems are another example - they are almost universally crappy and it is rare to get updates OTA. Shipping a car with capable hardware and slowly improving software capability imo is a smarter long-term approach.
>Tesla is the only participant that has a true path to self driving technology.
They dont even ship cars with Stereo front facing cameras.
Eight cameras, twelve hypersonic sensors, one forward facing radar, GPS, and still no stereo vision resulting in cars plowing full speed into stationary obstacles ....
Things like this are not binary "can do/cannot do". Sure some kind of estimate could be made. But it is challenging when the scene is not static, the speed of the camera varies, lack of texture, motion blur... stereo is not without its problems of course. Lidars seem to be good in lot of the places where visible light cameras are struggling.
It's possible to extract depth information from a monocular lens. There's several ways to do this, and Tesla has talked publicly that they've implemented at least one approach.
One is depth-from-defocus combined with "overlap" hierarchy information that can be used to build a depth-map. This is the technique Tesla talked about. Interestingly, instead of manually coding up this algorithm, they simply trained a neural net to do it! The trick was to use the radar data as the "ground truth", and train the NN to predict it based on the optical input. It works quite well, and there are videos floating about showing how a Tesla can build up an accurate 3D model of the world around it as it drives along.
There are other approaches as well. For example, the way birds gather depth information is they move their head. One eye takes two "snapshots" of the scene in rapid succession from slightly different positions, producing a stereo pair. A moving car is constantly shifting the camera positions. Note that you don't have to be moving the cameras side-to-side, moving along the road can be enough, as the shift of the focal plane can be used as an additional channel of information for the depth-from-defocus method mentioned above. This is what some mirror-less cameras do: they shift focus back-and-forth a bit to gain more information about the scene. A moving car has the same effect.
Tesla has recently implemented some variant of this latter technique, they also talked about this publicly. Essentially, the first version of the self-driving hardware didn't have the computing power to analyse multiple frames together like this. The current version has the compute power to include the last few frames as well as inputs to the NN, which gives it not just more temporal stability but also the opportunity to extract depth information from motion. (However, I'm not sure if they've actually used motion in this specific way.)
> One is depth-from-defocus combined with "overlap" hierarchy information that can be used to build a depth-map. This is the technique Tesla talked about. Interestingly, instead of manually coding up this algorithm, they simply trained a neural net to do it! The trick was to use the radar data as the "ground truth", and train the NN to predict it based on the optical input. It works quite well, and there are videos floating about showing how a Tesla can build up an accurate 3D model of the world around it as it drives along.
Wow, that scares the shit out of me. I hope this is not serious. How could they possibly understand the neural net and how it will behave under novel scenarios? This is the opposite of what you want in a safety critical system.
The problem with gradually going from drivers assistance to self driving is there will be some period, possibly a very long period, where the cars almost never need intervention but aren't safe enough to actually not need it. There will be a lot of accidents if it's released widely during this period (yes much more than human driver accidents. aside: there seem to be a lot of Tesla fans who think humans are much worse drivers than they are). This is what Google saw a decade ago when their self driving cars started getting good and their drivers started being irresponsible, so they decided to start from the idea the driver couldn't be a crutch and the car always had to be able to do something safe. Getting this level of robustness is why it seems to some that they've made no progress in 10 years.
What are you basing this on? You can just say 'despite what Tesla fans believe' as an argument. One datapoint from google 10 years ago is not really convincing.
Specially as car also gets better at measuring human attention. CommaAI pilot is already doing quite well on that. Tesla is working on it internally as well.
And also, we can do things like assess the situation and adjust necessary attention. If you are by yourself driving threw the desert, you can relax the condition. If you are in a city the car keeps you honest.
That this will clearly cause more accidents then human by themselves is not at all clear.
Do you have personal experience with commaAI's pilot? I don't so basing all on videos alone. In a video to promote the bleeding edge of their development with Lex Fridman Hotz was driving the car to a specific part of a freeway that had a tilt and he was able to let go off the wheel for a brief moment. In another part the car was driving by itself in an urban setting and the highlight was that it was able to stop by itself at an intersection. This is light years removed from anything resembling a self driving system. But again, if you have more insight than the few videos I am basing this on please share. Would be interesting to hear more.
Im thinking the tesla valuation is more than the car company, the car company is just making the economy of scale battery production profitable/or just funded, while the greater goal is the decentralized power generation and storage network of the solar roof + power wall. If we think of them as a energy infrastructure company + subsidiary car company/freight company (tesla semi) then I can see the valuation making some sense.
Tesla doesnt need perfect automation to handle cheap, automated, short run freight lines, leveraging clean cheap electricity, and low maintenance trucks(no ICE, less drive train maintenance, no driver time at wheel restrictions)
Yup. Tesla has built a real profitable company with a decent car and has brought a real shift in thinking about the cars of the future. If there is ever a real self driving car then I wont own it but rather rent it like Uber.
Waymo on other hand sounds like a cool engineering project that is not going anywhere and Google will shut it down in next 10 years.
I watched a finance YouTube podcast where they observed the price of Tesla stock can best be explained by "Tesla is a religion." I think this is true.
Tesla is a religion to people so it's not clear that the stock will have a come to Jesus moment since religions don't need to deal with facts.
People want the stock for religious reasons and will buy the price up.
Even in this thread you have someone saying "Elon always succeeds" which is obviously a religious idea. It is impossible for someone who believes this to have a come to Jesus moment.
Tesla is doing a good job after all. Given the discussion going on here the stock price needs no religious interpretation. Shareholders are simply betting on (semi) autonomy, climate friendly energy storage, next gen vehicles. Doesnt seem mad to me.
> At this point it’s best not to bet against Elon.
Maybe in a "should you short Tesla stock" or "will a Californian jury rule against him" sense.
But on the other hand, if you bet a friend on how many Tesla Robotaxis would be on the road by the end of 2020 you could be sitting on a nice bottle of whisky by now.
Betting one friend on a small timing detail is fun and all, but that not really what people mean when people talk about 'not betting against Elon'. We are talking about big things.
Elon said they will capture Apollo size booster directly from the air on ocean space ports. And the majority of people were like 'mh, sound reasonable will probably work in a few years'.
Think about laterally any other person or government saying that people would be like 'yeah, sure maybe in a decade plus and 20 billion or more'.
It’s not betting against Elon, I don’t think anyone thinks he will fail outright. But company valuations are shades of grey. So it’s betting against everyone else who owns TSLA stock under the argument that they will never be able to justify these valuations.
This clearly is not a position that Bloomberg as a publication is taking. It’s a report on interindustry shittalk - not the highest form of reporting but not even comparable to the story you’re referencing.
I am a huge fan of what Waymo has accomplished. And I think in the beginning Tesla's FSD Beta was so bad that it was reckless to release to the testers.
BUT. Look at the very latest version of the Tesla FSD Beta ("beta 10") as of now. It is literally doing the full self-driving task with very few disengagements now. And the improvements over the course of the last few months have been dramatic.
At this point it is not really safe without watching closely and has a fair way to go before it gets to the high 9s, especially in arbitrary situations as they intend.
But if you look at the progress they have made and how many fewer disengagements there are now, I do not think it is reasonable to assume that they can't get to the finish line. They look like their trajectory is actually to get there.
Lets be honest here. If Tesla had invested the waste majority of their efforts on 1 city, not having to deal with all the problems of broken line markings, crazy signs, signs in different language, driving threw fog at night, driving mountains roads, driving country side in China and so on.
Could the Tesla software do what Waymo does? Specially if you map out the whole city? I would argue it could.
I agree that Tesla is quite far away, but in terms of a general solution for waste majority of the world, Tesla is the closest.
I believe so. Because if it was one city then they could iron out all the edge cases there. And even with all the places it's deployed now, it is starting to get close.
But personally I think that Waymo is going to be a bit safer because it has the LIDAR which even if you don't think that matters, it can definitely see farther because of its position above the car.
And I think that Waymo will be deployed in several cities by the end of 2021 or 2022 at least. And I think it's great to have competition. Cruise will also probably be able to deploy.
But Tesla probably will force them to build systems that require less stringent mapping so they can add cities quickly.
nobody's autopilot will achieve full autonomy in the near future. Full autonomy actually meaning, you could rip the steering wheel out of the car and have it drive through any city you wish instead of "it drives through grid streets in American suburbs correctly like, 95% of the time"
The entire current approach has no correctness guarantees, ML systems are horrible at dealing with tail-risk, there's a human communication element to driving and so on. It's a ridiculously hard problem, honestly AGI hard.
I much more suspect that before we get autonomous driving we'll get more 'rail like' street systems, geofencing etc.. that limit how drivers and cars can behave, that's kind of how the whole hyperloop thing works ironically enough. Obviously if we wanted that we could just build better metro systems to begin with which already drive themselves.
Planes have centralized Air Traffic Control that dictates every move from take off until landing of all planes within their zone. The planes only have to follow the ATC orders and not worry about other planes except for very rare emergencies. Even in those, ATC usually gives them a large block away from other planes to sort things out.
Cars in comparison can start at any moment, from any location and want to go anywhere they like without knowing what other cars are doing, or -worse- the road conditions (potholes, ice etc).
Putting blame is also easier, because legally all pilots agree to be recorded while flying since they're doing business. It'll be much harder I guess to allow recording inside cars.
Could atc ramp scale in its ability to handle a denser set of planes? Could a network of planes host a decentralized atc that manages the traffic, similar to a decentralized road traffic control that would help manage automated cars through a 5g conmection in a city.
ATC isn't only about managing traffic. Each region's government has their own rules and they can choose to enforce new rules at anytime (e.g. https://www.youtube.com/watch?v=jaTv1vAC3W8). How would you enforce a 9/11 like shutdown in a decentralized system? Or if the Internet goes down?
The closest thing to that are collision avoidance systems that instruct pilots to go up/down by talking to one another and agreeing on a plan for a particular situation.
The most interesting thing about driving is that it requires a general intelligence to do it. Sure 99.99% of the time, you stay in lane, obey lights and don't ram things. But for a few seconds every hour you actually need to think.
This is why I'm bearish on fully autonomous cars, at least without major changes to infrastructure.
59 comments
[ 4.5 ms ] story [ 161 ms ] threadBut this does ring true, and the come-to-Jesus moment on Tesla’s valuation will be when people realize this is the case. If they had a legitimate shot at being the first true players in self-driving, their valuation (which currently prices them on par with something like the sum of the top three automakers...) isn’t insane. But if it really is just lane-keep assist with intelligent cruise control, I’m pretty sure the Mercedes S Class had that... back before 2010?
They are now worth more than the sum of every automaker in the world.
This Visual is dated but is still relevant.
https://imgur.com/iSHYca5
With that said, I still wish them great success.
Legacy automakers have failed to deliver on electric cars for decades. Musk talked about this. There was nothing special about the time he entered the market with Tesla. Any one of the auto makers could (and still theoretically can) develop an appealing electric car. They're simpler, cleaner and more efficient. But not a single one has been able to deliver a reasonable product. They're subsidizing the price of electric and losing 7-10k per [0] and still cannot produce a compelling product.
The legacy car companies are basically blackberry, as someone has finally shown there exists a way to create and produce a compelling car outside of the legacy participants.
Tesla actually has a path forward on self driving. They have the infrastructure, real driver data, and millions of cars on the road that can accept over the air updates. Waymo has a few Taxis in a few cities.
[0] https://insideevs.com/news/336213/moodys-says-automakers-los...
And waymos approach isn't too be a car manufacturer, but an Uber competitor. So fewer vehicles is fine.
> They're subsidizing the price of electric and losing 7-10k per [0]
To be fair, so is tesla.
Lidar is almost like adding a hand crafted dimension to the decision making process. It makes the system more explainable, but doesn't necessarily mean it will help achieve better results. I look at ML in games like chess and go. Originally, they had a set of hand crafted features, like the concept of "eyes" in go. But later iterations removed those and the network performed better. I think lidar is the same thing where it'll add unnecessary complexity and ultimately hurt in learning.
The waymo car is the only actively-used car with Lidar anyway as far as I’m aware, so if you exclude Waymo then your statement is self fulfilling no matter how good or bad Tesla’s ADAS is!
Costs matter.
I am very certain that Waymo has no autonomous vehicle today that mist people would be ok to go around their neighbourhood driverless. „Full autonomy“- socially acceptable- does not exist today. Whether it can be done, with or without Lidar, remains to be seen.
You can’t make a statement without qualifying this. You attack legacy automakers but then go on to discuss Waymo. I would bet money you’re long TSLA.
You talk about OTA updates as if that’s some huge technological moat? Hell, cars have been receiving very frequent software updates for over a decade.
Paragraphs 1-3 discuss legacy automakers. Only paragraph 4 references Waymo.
> You talk about OTA updates as if that’s some huge technological moat?
Of course they're not a technological moat. These auto makers are hundreds of thousands of employees. They're likely an organizational nightmare and stuck to the way things have been done. The fact that they don't have over the air updates or have to iterate on an annual cadence is evidence of something deeply flawed in their organizational structure. And I'm not confident that they can overcome it.
I think OTAs are a moat but highlight an important incentive difference. Traditional automakers want to sell you next year's model of the car. Tesla seems like they want your car to get better over time.
Perhaps it is organizational cruft or the dealer network but consider even the simple case of adaptive cruise control systems in cars today that have radar + cameras. Most of those systems bounce between lanes and can't even stay centered well. This is something that could be a software fix but you'll have to wait for the next N models when they redesign the car to see it.
Infotainment systems are another example - they are almost universally crappy and it is rare to get updates OTA. Shipping a car with capable hardware and slowly improving software capability imo is a smarter long-term approach.
Time will tell though.
They dont even ship cars with Stereo front facing cameras. Eight cameras, twelve hypersonic sensors, one forward facing radar, GPS, and still no stereo vision resulting in cars plowing full speed into stationary obstacles ....
One is depth-from-defocus combined with "overlap" hierarchy information that can be used to build a depth-map. This is the technique Tesla talked about. Interestingly, instead of manually coding up this algorithm, they simply trained a neural net to do it! The trick was to use the radar data as the "ground truth", and train the NN to predict it based on the optical input. It works quite well, and there are videos floating about showing how a Tesla can build up an accurate 3D model of the world around it as it drives along.
There are other approaches as well. For example, the way birds gather depth information is they move their head. One eye takes two "snapshots" of the scene in rapid succession from slightly different positions, producing a stereo pair. A moving car is constantly shifting the camera positions. Note that you don't have to be moving the cameras side-to-side, moving along the road can be enough, as the shift of the focal plane can be used as an additional channel of information for the depth-from-defocus method mentioned above. This is what some mirror-less cameras do: they shift focus back-and-forth a bit to gain more information about the scene. A moving car has the same effect.
Tesla has recently implemented some variant of this latter technique, they also talked about this publicly. Essentially, the first version of the self-driving hardware didn't have the computing power to analyse multiple frames together like this. The current version has the compute power to include the last few frames as well as inputs to the NN, which gives it not just more temporal stability but also the opportunity to extract depth information from motion. (However, I'm not sure if they've actually used motion in this specific way.)
Wow, that scares the shit out of me. I hope this is not serious. How could they possibly understand the neural net and how it will behave under novel scenarios? This is the opposite of what you want in a safety critical system.
The Renault Zoe was the best selling BEV in Europe in 2020:
https://insideevs.com/news/467101/renault-bev-sales-december...
Tesla has gone from being the top BEV maker in Europe to now being third. Soon Tesla will become fourth in Europe after Hyundai overtakes them.
Different markets like different cars. The Audi e-tron was the best selling BEV in Norway in 2020 and is currently the best seller in 2021:
https://ev-sales.blogspot.com/2021/01/norway-december-2020.h...
https://elbilstatistikk.no/
What are you basing this on? You can just say 'despite what Tesla fans believe' as an argument. One datapoint from google 10 years ago is not really convincing.
Specially as car also gets better at measuring human attention. CommaAI pilot is already doing quite well on that. Tesla is working on it internally as well.
And also, we can do things like assess the situation and adjust necessary attention. If you are by yourself driving threw the desert, you can relax the condition. If you are in a city the car keeps you honest.
That this will clearly cause more accidents then human by themselves is not at all clear.
Tesla doesnt need perfect automation to handle cheap, automated, short run freight lines, leveraging clean cheap electricity, and low maintenance trucks(no ICE, less drive train maintenance, no driver time at wheel restrictions)
Waymo and co are simply having a gigantic money burning party and will continue.
Tesla can work on this and support all their efforts as a value at to the current products.
Waymo on other hand sounds like a cool engineering project that is not going anywhere and Google will shut it down in next 10 years.
Tesla is a religion to people so it's not clear that the stock will have a come to Jesus moment since religions don't need to deal with facts.
People want the stock for religious reasons and will buy the price up.
Even in this thread you have someone saying "Elon always succeeds" which is obviously a religious idea. It is impossible for someone who believes this to have a come to Jesus moment.
I’d bet his timelines are off, that’s always been the case. But his army of engineers know what they’re doing.
Maybe in a "should you short Tesla stock" or "will a Californian jury rule against him" sense.
But on the other hand, if you bet a friend on how many Tesla Robotaxis would be on the road by the end of 2020 you could be sitting on a nice bottle of whisky by now.
Elon said they will capture Apollo size booster directly from the air on ocean space ports. And the majority of people were like 'mh, sound reasonable will probably work in a few years'.
Think about laterally any other person or government saying that people would be like 'yeah, sure maybe in a decade plus and 20 billion or more'.
Elon said coronavirus would be finished in the US by April, 2020. How'd that work out for him?
Also, that is not something that Elon or his companies would do so it doesn't really speak to my point.
If that is the best you can do for THE LAST 10 years its actually embracing.
Tesla is far from the only EV manufacturer.
Sure, if you were dumb enough to believe that last time, you might consider Bloomberg a reputable source this time, but otherwise why would you?
This clearly is not a position that Bloomberg as a publication is taking. It’s a report on interindustry shittalk - not the highest form of reporting but not even comparable to the story you’re referencing.
BUT. Look at the very latest version of the Tesla FSD Beta ("beta 10") as of now. It is literally doing the full self-driving task with very few disengagements now. And the improvements over the course of the last few months have been dramatic.
At this point it is not really safe without watching closely and has a fair way to go before it gets to the high 9s, especially in arbitrary situations as they intend.
But if you look at the progress they have made and how many fewer disengagements there are now, I do not think it is reasonable to assume that they can't get to the finish line. They look like their trajectory is actually to get there.
Could the Tesla software do what Waymo does? Specially if you map out the whole city? I would argue it could.
I agree that Tesla is quite far away, but in terms of a general solution for waste majority of the world, Tesla is the closest.
But personally I think that Waymo is going to be a bit safer because it has the LIDAR which even if you don't think that matters, it can definitely see farther because of its position above the car.
And I think that Waymo will be deployed in several cities by the end of 2021 or 2022 at least. And I think it's great to have competition. Cruise will also probably be able to deploy.
But Tesla probably will force them to build systems that require less stringent mapping so they can add cities quickly.
The entire current approach has no correctness guarantees, ML systems are horrible at dealing with tail-risk, there's a human communication element to driving and so on. It's a ridiculously hard problem, honestly AGI hard.
I much more suspect that before we get autonomous driving we'll get more 'rail like' street systems, geofencing etc.. that limit how drivers and cars can behave, that's kind of how the whole hyperloop thing works ironically enough. Obviously if we wanted that we could just build better metro systems to begin with which already drive themselves.
Cars in comparison can start at any moment, from any location and want to go anywhere they like without knowing what other cars are doing, or -worse- the road conditions (potholes, ice etc).
Putting blame is also easier, because legally all pilots agree to be recorded while flying since they're doing business. It'll be much harder I guess to allow recording inside cars.
The closest thing to that are collision avoidance systems that instruct pilots to go up/down by talking to one another and agreeing on a plan for a particular situation.
There is always a pilot communicating with ATC, providing data, receiving and validating instructions and then entering them to the flight computer.
Also an interesting perspective from Jim Keller, that autonomy is easier than you might believe:
https://youtu.be/Nb2tebYAaOA
I think it was in this link but not 100% sure... I watched a bunch of his interviews.
Generally speaking, I think GPT3 shows the power of raw computer in ML: https://www.gwern.net/newsletter/2020/05
This is why I'm bearish on fully autonomous cars, at least without major changes to infrastructure.