671–680 of 692 posts
Earlier quoted context omitted.
Probably China will always outpay Cruise so you see migration from Waymo to Chinese companies, but not to cruise
Huh? Cruise was paying $400k+ for like 3YoE. I don't think China is paying more than that.
Earlier quoted context omitted.
The vehicle already has the inverter, using the same active rectifier hardware in the on-board charger. People want to have electricity when they are at home. And they are home a lot of the time. So I agree that it is a very niche case to want to have a backup system to provide electricity to your house during a natural disaster when you are not even there. https://gmenergy.gm.com/for-home/products/vehicle-to-home-so…
OK but you still need a grid cutoff--you absolutely can't put power into the grid when it's down. It's not a huge investment, but it's not nothing.
Earlier quoted context omitted.
If you let me pick the roads, I could easily drive 90 minutes at night with no intervention in the LA area: driving from Ventura, down the 126 to the 5 to LAX can be done basically with no interventions no problem today. It could be done with EAP two years ago too. But, FSD works well on essentially arbitrary roads and it didn't four years ago.
> driving from Ventura, down the 126 to the 5 to LAX That's a lot simpler than what Cruise showed in their video. Dealing well with pedestrians, stopped cars, cross streets, etc are a significant challenge.
Frankly, I think Tesla is going to win this. The new self-driving is remarkably good. Based on this alone, I estimate (actual) level 5 self driving in 2-3 years. I'm convinced that the lidar sensors are essentially unnecessary and the vision-only strategy is basically going to work and be much cheaper in the process.
How does vision work at night?! RGB Cameras dont see anything...
Earlier quoted context omitted.
This i what is so tricky with Self Driving. People "feel" it is almost there because most of their rides are mostly ok. However to make a system truly driverless you need to master the long tails of difficult events and FSD is nowhere even close to do that. Going from 99.9% to 99.99999% is what makes a system truly driverless and where most of the work is. Waymo is way way ahead of FSD for this.
I know essentially nothing about machine learning, nor about what approaches to ML that Tesla or Waymo are using. Is there a Metcalfe-type effect possible here? Where the better FSD becomes, the more people use it, finding more edge-cases, so the driving get better, so more people start to use it. With the end result that the learning starts to get better/faster with time, in a positive-feedback like mechanism?
Your comment/question seems to assume that somehow accurate classification of performance is possible, which is demonstrably false assumption.
First, if somehow system-internal watchdog was able to detect erroneous outputs for training, it would also be able to stop those outputs propagating to control system live, leading to zero errors. Such a watchdog requires self driving to be quantified analytically, leaving implementation entirely testable and therefore control system not passing the testsuite (i.e. exhibiting erroneous behavior, i.e. still having "edge cases") not deployable in public. Many words to say that that in practice with driving being somewhat loosely defined you need humans in the training loop.
Second, vehicle operators are not only unqualified to accurately monitor behavior of these autonomous systems, in part due to not being formally trained on safety systems in general and the system they are monitoring in particular, but on top of that incentivized to underreport erroneous output.
My prediction has been that we are going to see increased number of accidents with ADAS deployments, not less, before the number can start dwindling.
Earlier quoted context omitted.
Yes there are actual "legal" notions of driver assistance "levels," but out of those 3 companies Tesla seems to be taking it the most seriously. Elon has a high conviction in the product and has continued to pour billions of dollars in GPUs, talent, supply chains, etc to make it happen. Other manufacturers don't take it nearly as seriously, unfortunately. Sure you might not be able to legally claim that a Tesla is to…
> of those 3 companies Tesla seems to be taking it the most seriously. You call lying about it endlessly, misleading marketing, beta-testing on public roads, and not even having L3 properly in production "taking it the most seriously"?
I can't point to a single Andrej Karpathy or Chris Lattner or Jim Keller working at Mercedes, BMW, GM, etc (not to mention the not so big named people who are still very very good). And I also don't see many people crossing over from legacy auto autonomous orgs to openai, anthropic, deepmind, etc.
Other manufacturers don't have custom in car inference chips either, or spend billions in R&D for custom training silicon. This is clearly not a side project for Tesla, whereas with other manufacturers, its an obvious afterthought.
My guess is other manufacturers will just license some AV product from whoever is most successful and try to sell products like they always have - through design language and brand feel, not through breakthrough technological innovation.
So yes, I don't think any other consumer car manufacturer is taking it seriously.
Earlier quoted context omitted.
I disagree with you. Glad you worked in the industry, I am using FSD daily, it is really excellent.
Drive 100 miles in your sleep. You'd trust your partner or parent to do that for you ...
That said, based on my experience with FSD, I'd be tempted to take a nap if it let me, reinforcing my initial statement that it's very darn close.
Earlier quoted context omitted.
Have a look at V13.2 and tell me they "underdelivered". This is the most advanced AI driving system in the world. It's essentially DriveGPT, trained on absolutely massive amounts of training data on the largest GPU super cluster in the world. https://www.youtube.com/watch?v=NeNKYRWGgBc Just because Elon was optimistic in 2017 doesn't make this any less impressive.
First of all, I notice that you didn't take issue with my saying another major hardware revision is due - despite Tesla's constant promises that "we totally can do it with this one bro" for the last 4 revisions. I have been following closely. 13.2 is a major leap forward. It is VERY impressive. It's also five years late. It's got some serious issues, such as intervention rates (which, despite being massively lower, a…
As for the most advanced AI driving system, yes you could argue Waymo is more advanced, it's certainly more complex. By most advanced I mean that it's fully end to end, and this is a very new development in AI and robotics (which I work with). It's actually a much simpler architecture than other systems. But as we've seen repeatedly, what tends to win is simple models with a ton of compute and data.
As for V13.2, it won't be perfect, but it's also not the final version. It's approaching feature complete, and we've seen it handle snow covered roads at the very least. Rain hasn't really been a problem for recent versions. In my view, weather is mostly a matter of data collection at this point.
Earlier quoted context omitted.
> You are arguing the system would perform worse in contexts in which you have not seen it perform (because its designers decided to prevent it from such attempts). Should we also assume that Mercedes cars can fly better than a plane, because we have not yet seen them fly? What kind of logic is that? In other news, OpenAI has solved AGI last year, but is keeping it a secret because it's too dangerous. I have solved s…
No, but generally you should assume a system with near-100% performance within a well-defined ODD doesn't fall off to near-0% performance immediately adjacent to that ODD. Here's a better example: let's say by law Mercedes has to restrict their cars from exceeding 80mph. They have a car that can reach 80mph in 2 seconds. Is it therefore credible to claim that the car is actually incapable of driving at 81mph? Or is i…
The fact that it shuts off immediately shows that they're just copying what the car ahead does.
The system is unable to function on a pre-mapped highway on a clear day with lane markings if there is no other traffic on the road. What does that tell you?
Earlier quoted context omitted.
> What evidence do you have for your argument? The fact that their software is written only to just follow the car in front, and completely and immediately give up if there's no car in front or one cannot be seen due to it being dark or bad weather, on premapped roads. Meanwhile FSD has repeatedly shown that's it's capable of driving without a car in front. You can watch YouTube videos of FSD driving without a car in…
What makes you think that MB doesn't have the same capability, but they aren't willing to ship a half-arsed highly dangerous, people killing feature like some other companies do? Unlike any of their competitors, they assume full legal liability for their car's behavior while Drive Pilot is enabled. MB also has a luxury brand reputation to maintain. When they ship a major feature like that, it absolutely has to work 1…
Source? If someone gets killed when the system is driving, there is no legal framework for you as a driver to escape criminal penalties. Who at MB are they going to jail for it?
Can UPS indemnify their drivers from involuntary manslaughter charges?