Or bad...Right now it doesn't appear that investors really buy into the idea of Robotaxi or FSD. Not that it's causation, but the more Elon speaks about Robotaxi/FSD, the more the stock seems to drop.Very good for Tesla investors. Very bad for shorties.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
FSD v12.x (end to end AI)
- Thread starter Buckminster
- Start date
-
- Tags
- #fixphantombraking autopilot chill is not chill elon is an ass fix fsd extreme late deceleration fsd does not commit to go fsd does not maintain proper speed fsd hard and late braking fsd jerky driving fsd late deceleration fsd stops too early too quickly fsd stops too far behind line fsd tailgating in city streets
Again, I didn't call FSDb what you say. I called Elon's statements it, because they are.
It's not likely going to take five years for humanity to reach AGI. If we're lucky it's happening within the next 50 years. Vision-only will likely never happen as hardware costs for better sensors are dropping like a rock, so only extremely stupid people will keep insisting on vision-only in 3 years. If Tesla wanted to increase performance of FSDb by a few orders of magnitude, they'd add a better radar and a Lidar. But selling cars with a 15% profit is more important to Tesla than solving autonomy, so they won't.
Almost all current ride-hailing companies make a profit. They are all geo-fenced. You can't walk up to a cab in NYC and tell it to drive you to LA.
Tesla didn't make a profit for the first ten or so years, and most SaaS startups take 3-5 years to make a profit so your statement is completely nonsensical for a potentially revolutionary product like autonomous cars.
Also a completely nonsensical statement. It can happen in 10, 20 or 50 years or in 200 years.
It's a moonshot. It takes time. They've been going for 15 years. Waymo was arguably further ahead than Tesla is today in back in 2016.
This is like saying an LLM:s only needs work in poetry. LLM:s needs fundamental scientific breakthroughs to never lie or hallucinate. In a similar manner that computer vision needs fundamental scientific breakthroughs to be reliable enough in all conditions, everywhere.
If Tesla ever deploys FSD on vision-only (hw3/hw4) without a driver, even in the Las Vegas Convention Center tunnel I'll delete my account here so you don't have to be upset. Until then, see you around.
Last edited:
Please show me some financial statements from all these robotaxi companies that makes a profit..or are you referring to ride hailing like uber?(which also doesn't make a profit). You said ride hailing which is not robotaxi so not sure if you are throwing a strawman here.
90% of the global $200B revenue are concentrated to metropolitan centers. I'd guess 0.1% of the rides are between states/cities, if they even are allowed to.
Here's what you said, which I said was complete nonsense, since it it complete nonsense:
Do you think existing local ("geofenced") taxi businesses are (1) making a loss because they are local to an area (2) would be less profitable if they didn't have to pay wages to the driver.
All taxi business are geo-restricted by design because the cars need cleaning and service at the depot regularly. A taxi business is still a taxi business even if you remove the drivers.
I'd guess the initial extra cost for a robotaxi is ROI:ed in less than one year compared to a driver-taxi.
Can you make an argument for why "robotaxis can never make a profit" if they are geofenced to one city + airport? Or if they have a Lidar?
Would you say it's primarily because of the geofence or the lidar?
Last edited:
Changing topic a bit.
Since they are working on what they call v12 with some major reworking of code, I’d expect
1. v11 updates will slow to a trickle, with very little major happening.
2. They will again be looking at their fsd computer and the way the chip is configured. One can optimize or be general purpose. The original hw3.0 was designed to process lots of 2d camera images. With the change to the “bucket of points” some of the silicon went unused. Hopefully hw4.0 anticipated the v12 rewrite.
3. Maybe it’ll be rewritten for hw4 in mind, which will be the first time we’ll get to see a divergence in capability between older and newer cars.
Since they are working on what they call v12 with some major reworking of code, I’d expect
1. v11 updates will slow to a trickle, with very little major happening.
2. They will again be looking at their fsd computer and the way the chip is configured. One can optimize or be general purpose. The original hw3.0 was designed to process lots of 2d camera images. With the change to the “bucket of points” some of the silicon went unused. Hopefully hw4.0 anticipated the v12 rewrite.
3. Maybe it’ll be rewritten for hw4 in mind, which will be the first time we’ll get to see a divergence in capability between older and newer cars.
KArnold
Active Member
Pssst, hate to break it to you but it's also the first and that isn't going all that well.
Does he just surround himself with "yes men"?
How far can one move the goal posts? This must be a world-record attempt.
"This year" being 2019 and "work" being "robotaxi". How's Manhattan going? The typical 30 min downtown drive in most areas has 10 interventions and a few disengagements. You need to do perhaps 10 000 intervention free drives in a row for it to be robotaxi-ready. That's going to be completed around never on the Full-Self Driving Capable hardware (hw3/hw4).
Elon 2023: "I think we may have figured out some aspects of AGI. The car has a mind." LMAO.
Imaging the Tesla stock if Elon was honest in that or any other presentation he's ever made:
"Yeah, I think we'll get to a really good drivers assist system within five years".
Last edited:
diplomat33
Average guy who loves autonomous vehicles
This must be one of the most ridiculous things Elon has said. In fact, I think it is a good example of the "pseudo-science" you were talking about before.
AGI means that the same intelligence is able to perform many diverse tasks. For example, humans have generalized intelligence because the SAME human brain is able to learn many completely different tasks from riding a bicycle, driving a car, reading a book, writing music, dancing, swimming, building an object, solving a math problem, playing chess, giving a speech, arguing a case in court, talking to a friend on a topic, etc). These are all very different tasks that require different forms of intelligence yet the human brain is able to do all of them. That's AGI. Driving is not AGI, it is a specific task, a subset of AGI. Tesla FSD is designed to do the very specific task of driving and nothing else, so it is specialized intelligence, not AGI. So no, Tesla has not figured out aspects of AGI. Also, the term "mind" usually refers to human intellect. The FSD computer performs tasks and follows rules inside the neural nets but it is not thinking on its own and it is not capable of abstract thinking like humans can. So I am not sure I would call it a mind. It is a computer.
Yep. More and more sensational as things fail to improve. It's getting to be a faith-based Brother Love's Traveling Salvation Show. No status on what's going wrong and instead focus on our endless ideas and faith, hallelujah!
clydeiii
Member
First off, we don't know what v12 alpha is like so it's hard for us to judge.
But I have long believed that for us to actually get to "real" FSD, we'd need AGI, since driving itself requires a bunch of different intelligences (vision, planning, reaction to enviornment). I've also thought that a multi-modal LLM hooked up to a bunch of different inputs isn't too far away from what a brain looks like. The image/video input turns the images/videos into higher-level concepts, which the language component can then reason about, which informs the planning/driving model on what the car should be doing any any given second. Those actions are then fed back into the vision system to produce further actions. That's...very close to AGI. Elon didn't say "we have AGI", he said "some aspects of AGI"...seems reasonable.
Funny how it's always 6-12 month off... "intelligence" and "has a mind" are both ridiculous claims.
You cannot be serious....
Wikipedia:
"an AGI could learn to accomplish any intellectual task that human beings or animals can perform. Alternatively, AGI has been defined as an autonomous system that surpasses human capabilities in the majority of economically valuable tasks".
LLM:s are not intelligent by any definition:
Last edited:
People probably said similar things with GPT, and there's many definitions of AGI, so "figured out some aspects of AGI" is probably highlighting the "General" over what he usually talks about AI.
In the context of GPT, people were impressed by the ability to take different concepts and generate seemingly novel outputs without being explicitly trained on those tasks. So the ability to do zero-shot learning just based on the prompt could be some aspect of AGI as it has enough general understanding of language.
So similarly in the context of driving, I would guess early V12 training might be showing ability to handle a (large?) variety of situations it wasn't explicitly trained on because it might have some general understanding of roads and objects and how they might behave.
In either case, there are also plenty of people who say neither are anywhere close to AGI, and they can be correct as well.
Let's redefine all words until Elon is right, shall we?
Literally everyone agrees that AGI is probably decades away. Everything else is just marketing. Like "sparks of AGI".
Now Elon said "the car has a mind". So let me hear you argue his case on that one.
diplomat33
Average guy who loves autonomous vehicles
I still think Elon should avoid using terms like AGI or "mind". It's hype and marketing but it is misleading and creates false expectations.
I think he should avoid some other words such as "Level 5", "robotaxi", "autonomy" when selling a driver assist system.
Instead he doubles down on humanoid robots. The man is either delusional or has zero regard for the truth and has no shame what so ever.
Last edited:
clydeiii
Member
Has a mind probably means has a basic understanding of the world. You can see sparks of this in LLMs. Seems reasonable for this to be in the neutral nets of FSD too.
"A mental model of the world is an equivalent to, if not a consciousness at least a mind. So we're getting the building blocks of AGI here". That sounds great, but it completely false. This is a prime example of pseudoscientific marketing bull sh!t. Pseudoscience - Wikipedia
He goes on to "explain" that we don't understand that FSD is so smart because it cannot talk to us... According to this video FSD will write poetry apparently when it has an LLM attached to it and entertain us on trips. How can anyone take this clown seriously?
The car doesn't have a "mental model of the world" in the same way as it cannot think, reason, is not self-aware and shouldn't be described in these terms.
Indicators of possible pseudoscience include:
- Use of vague, exaggerated or untestable claims
- Improper collection of evidence
- Lack of openness to testing by other experts
- Absence of progress
- Personalization of issues
- Use of misleading language
Last edited:
Similar threads
- Article
