That demonstrates they created what they think is necessary to achieve L5, but that doesn't mean they have L5.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Autonomy Investor Day - April 22 at 2pm ET
- Thread starter diplomat33
- Start date
That demonstrates they created what they think is necessary to achieve L5, but that doesn't mean they have L5.
You think their new chip is more powerful than what Waymo has in the trunk? Seriously? Cheaper, yes. More power efficient? Probably. More powerful? Not a chance.
You think Tesla knows something about deep learning that Waymo does not? You think they made a breakthrough? I haven't heard a single thing yet that isn't already well understood in the machine learning community.
The value of "the fleet" and training data has been debated endlessly, but let's just say that it is clear that they are only using a fraction of the fleet data, and that part is not debatable. Waymo has a lot of money. If they wanted more training data they could get it.
The Onion has a really funny parody piece about Tesla:
https://www.theonion.com/tesla-posts-massive-first-quarter-loss-after-self-drivi-1834310890
CarlK
Active Member
The answer is NO Waymo does not. Anthony Levandowski who was the Lidar guru at Google and started the whole thing there said in a recent interview "a wiser man than me said Lidar is a crutch" and "he (Elon) is right".
People usually don't understand those who are much smarter them and often even redicule them because of their own inadequacy. In the end the wiser man is always right. One of the definition of wise is to be right most of the time. You can probably figure out what not wise is, that is if you're smart enough to figure that out.
Last edited:
Wow. You're going to use Levandowski as a source? The guy who stole from his employer and defrauded investors? OK.
Well, as long as we're engaging in personal attacks, I would be happy to submit to an IQ test if you will, and we can compare scores. A real IQ test, not some random crap you find on the Internet. Heck, I'll even put myself up against Levandowski and Musk if you can convince them. Maybe they'll beat me, but I humbly submit that I think I have a shot.
Disclaimer: I've actually taken a real IQ test, administered one-on-one by a trained clinical psychologist over the course of two full days, and I know those results. Though admittedly that was decades ago; I might not do quite as well now -- carbon-based neural nets don't always age well.
Though to paraphrase the great philosopher Alanis Morissette, I will also admit that intellectual capacity alone does not equate wisdom.
As I said, I meant demonstrable by a test drive. How are any of those demonstrable by a test drive?
Several manufacturers have done more impressive demo drives, including without Lidar, so that alone is not yet a demonstrable lead.
I acknowledged the fleet and the project that goes along with that in the next paragraph you left out of your quote. I fully acknowledge Tesla’s deployment and validation advantage in the OTA fleet.
I’ve said time and again, both here and offline, that I’m a bit of a cynical asshole. That cynical side of me says that if they had an ACTUAL lead, they’d show it. They’d make sure EVERYONE was able to see it.
Check their stock price. They closed their worst day in two years yesterday. Whether you’re the most hardcore Tesla supporter or you’re one of the people that truly wants to see Tesla cease to exist, people can all agree on one thing: this is NOT GOOD for Tesla. They need some wins.
They spent almost four hours talking about how great their hardware and software is, and how they are on the cusp of putting a million fully, 100% level 5 autonomous robo-taxis on the road. But they DON’T show it to the world and specifically forbid photos and videos of it. You need to A) get people with money excited about this and B) get the general public comfortable about this concept.
You can look at pieces of the picture and maybe think there’s a lead there, but if you look at the whole picture, thinks start to look... questionable.
I really, truly want to see Tesla make this a reality. My struggles with believing it’ll happen don’t come from the likes of the Bladers of these forums, they come straight from Tesla themselves.
@J1mbo
Besides, Tesla definitely has an advantage: They are selling a Level 5 no geofence car, feature complete by end of 2019.
That is a massive advantage.
Besides, Tesla definitely has an advantage: They are selling a Level 5 no geofence car, feature complete by end of 2019.
That is a massive advantage.
I say this as a huge Tesla fan who is totally in love with my Model 3: when it comes to actual capabilities, Tesla is not ahead in FSD right now. Tesla is behind the likes of Waymo, Cruise and Mobileye. They have cars that can do more than what Tesla can currently do, like navigate complex intersections completely hands free and more reliably. So we can stop the argument that Tesla secretly has a big lead and is just hiding it. They don't show off a big lead because they don't have a big lead. Right now, Tesla is still working on getting feature complete. What we saw in the demos is the best that Tesla has so far. They do have much better camera vision and computer hardware now (AP3) than they had before on AP2, and they have the software to do hands free driving pretty reliably on easy local roads and highways, including most easy intersections, traffic lights, stop signs, and highway transitions which they could not do before on AP2. So they have made big progress, not compared to the leaders, but progress compared to what they had before.
BUT, what Tesla does have is the ability to catch up very quickly and distribute that progress very quickly to a large number of cars already on the road.
Basically, the leaders like Waymo have:
- More comprehensive and redundant FSD hardware.
- Better software capabilities.
- More resources to devote to the problem.
What Tesla has:
- A large fleet of commercially available cars already driving on roads across the US that just need a software update to improve in their self-driving capabilities.
- The ability to grab large quantities of targeted snippets from this fleet to quickly train the machine on a particular task.
- The ability to run their software in shadow mode in a large number of their cars on the roads to test and validate an improvement very quickly.
- The ability to upload a newer version of the software to the large fleet to do real world testing and get feedback to further improve the software.
- An ever growing fleet of cars so the sample of data is increasing.
- Tesla can repeat this cycle over and over to refine the software.
I likened the comparison to a race. Waymo has 1 runner that is way head. Tesla has 100 runners that are currently behind but catching up.
Or think about it this way: Waymo gets to L4 first with a relatively small fleet of ride sharing cars. Maybe, Tesla gets to L3 but with hundreds of thousands of cars already driving on public roads all across the US now. And if and when Tesla does get to L4, it will be on hundreds of thousands of cars already on the road today.
Last edited:
Let’s stick to the facts and not make up stuff. Tesla’s entire expense structure is Autonomy so they have more.
Are you sure? Tesla has to worry about manufacturing cars, sales, gigafactory expansion, etc.. and Tesla did lose $700 million last quarter. They can't exactly afford to burn money. Waymo gets a ton of money from Google that they can pour directly into FSD with no other expenses. And by the way, Google is almost a trillion dollar company. Their market cap is $817 billion. They can just give Waymo a huge budget to work with. And Waymo does not have to worry about making a profit like Tesla does.
I’m just quoting Tesla’s CEO from the 22nd. I would imagine he knows better than you or me.
CarlK
Active Member
Lidar is a stop gap, or crutch, and will let you to do things without a full machine learning capability. Vision and large machine learning fleet that covers all tail cases is the only way to get the last few 9's.
Tesla is the only company that is fully intergreted in this. Waymo is the only other company that makes all its components but it does not make cars and will never be able to run a large fleet. No one could catch up with Tesla in years if they copy what Tesla is doing today even if they have the ability to.
They may not be ready for that yet but anyone else can do that? Also there is a likelihood that the route was just selected by the Nav.
Tesla is the only company that is fully intergreted in this. Waymo is the only other company that makes all its components but it does not make cars and will never be able to run a large fleet. No one could catch up with Tesla in years if they copy what Tesla is doing today even if they have the ability to.
They may not be ready for that yet but anyone else can do that? Also there is a likelihood that the route was just selected by the Nav.
Last edited:
The stock price is a reflection of investor sentiment, nothing else. Multiple factors are pushing it down. The cynic in me says that the timing of the Autonomy Day was an attempt to defuse the impact of the bad results.
Ignore everything that Elon said at the event.
Step back and look at the picture objectively and you see something else.
Listen to the experts talking about their domains.
Understand that this is a moonshot, and like all moonshots, it has already produced some fantastic byproducts - but may not ever achieve all of the original goals.
For example, the HW3 board could be put in consumer packaging and sold as a standalone NN compute device. 144TOPS at 75W, for a couple of hundred bucks? IMO, this could be as revolutionary as the Raspberry Pi was.
Elon can't sell a moonshot without selling you the moon. So take the million robotaxis as an example of his ambition and not as a statement of fact.
Except everyone else does this aswell. Everyone else uses Vision AND machine learning and they also have fleet of their own. there are literally companies whose only job is to collect, label and sell data to SDC companies.
Mobileye also makes all its components other than the car.
They picked the easiest highway and suburban route and STILL had disengagements.
Literally their demo consisted of like 90% highway, driving on one lane.
Sure about that?
Karpathy mentions shadow mode in the fleet during the autonomy investor presentation, 2:16:35 mark:
Kant.Ing
Member
Similar threads
