-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Tesla, TSLA & the Investment World: the Perpetual Investors' Roundtable
- Thread starter AudubonB
- Start date
Until you discount rides for other times. Easy fix.... Come into work early and it's free, offsets by higher prices at peak demand. Just one of many possible scenarios. The city has an interest as well, and might even kick in $ to control traffic and make better use of the existing roads (cost avoidance).
Knightshade
Well-Known Member
While I agree with that, you still have the order of operations wrong.
You don't put an L2 system on the road with a safety driver and hope "training makes it L4"
That's not what L4 is.
L4 requires specific features and capabilities FSD currently does not have- I've called them out specifically to you but since they're J3016 terms you refuse to understand what they are and why "just train more" doesn't magically create those functions.
What you do is put an L4 system (ie one that has those 3 missing things already) on the road with safety drivers. This lets you test those specifically-above-L2 items in a safe manner-- and then refine them as needed to achieve whatever your company has defined as reliable enough to remove the driver.
Which again requires at least an attempt at a complete OEDR, an attempt at a complete DDT fallback function, and a defined L4 ODD. None of which exist in FSD today.
AFTER you have those you can start doing RT work. BEFORE that you can just have a human-driven taxi with a really good ADAS system helping them out.
I've already told you the 3 major missing items. There'll be more small things they find, but they're likely things existing fleet data is already showing them (FSD curbing wheels since 12.x takes tighter turns for example).
Since you have never read it yourself, how do you know that?
I do know they'll never solve legal RTs without reading it and complying with it though.
(barring a whole revision of the laws of a bunch of places)
Sure. But you need those 3 fundamental items to exist before you can test them. In FSD today, they do not.
Because L4 is not just "really good L2". Fundamentally.
Are these regulators in the room with us now?
Seriously dude, you're factually wrong on how any of this works legally. It's been explained to you over and over and you simply ignore it.
The actual laws already exist in a bunch of states. 100% of them require compliance with J3016 for legal self driving. There's no "regulators" to satisfy- that ship has already sailed and what you need to satisfy is already on the books.
Even in the one US state that DOES require "approval"- CA- THOSE regulators ALSO use J3016 as the benchmark, not just "Tesla says it's good"
Why keep ignoring the actual laws?
Why keep refusing to answer the question of what US state you believe Tesla can deploy driverless in without J3016 compliance?
I think this is the first major divergence for AMD cars vs MCU2 cars in terms of basic look and features I can think of... remains unfortunate Tesla didn't design MCU 3 to be retrofittable like they did MCU2- seems like a lost revenue opportunity.
Last edited:
Wow. That’s completely different than our experience. 12.3.3 was competent and reliable. 12.3.4 is not a huge regression but definitely a step back. Lane selection is worse, it’s lost its ability to merge again and it will make last minute route changes including proceeding thru intersections in the wrong lane.
I think it’s a good thing that all the free trials were on 12.3.3 as many would have been freaked out by 12.3.4.
This is one that has vastly changed in recent years. Only a couple decades back a Director was very senior, typically a member of a governing board. Now it really is situational, with often fairly low level. Same with Vice President, which is meaningless in financial services, but sometimes very senior in industrial companies. Titles are situational so I never attribute significance to one unless i also know the corporate structure.I've score you 50/50 for that one--- though I think a title of "Senior Manager of Cathode Materials & Manufacturing" at a company like Tesla might be above middle.... That said- a director is absolutely an executive at every major corp I've worked for and as commonly used in the US.
At times I've made a hobby of trying to decipher titles by country and industry. I gave up and just assume that behavior will self-define the probable level, itself often wrong.
At Tesla we once had complete clarity. Now it is huge so... it really seems opaque to me.
Musskiah
DisGruntled
Wonder if this Apr 12 tweet by Rohan had anything to do with his "retirement." It seems possible especially if Tesla announces a factory in India, as it seems very likely they will, especially given the recent chip partnership.
www.google.com
Tesla’s Head of Policy Reveals “Many of His Colleagues at Tesla” Are “Surprised to Hear The EV Maker Has Decided to Build a Gigafactory in India” – Clarifies “It’s Another Totally Made-Up Media Story”
Since Elon Musk announced an upcoming India visit, Tesla’s head of policy revealed that he has had several conversations with his Tesla colleagues, who were surprised to learn that the EV maker had made a major decision associated with India.
www.google.com
Tesla Vehicle Production & Deliveries and Date for Financial Results & Webcast for First Quarter 2024
AUSTIN, Texas, April 2, 2024 – In the first quarter, we produced over 433,000 vehicles and delivered approximately 387,000 vehicles. We deployed 4,053 MWh of energy storage products in Q1, the highest quarterly deployment yet.
7 days from now is the Earnings call.
Are you sure about that? Frankly I'm more excited about FSD and RT being able to perform tasks that humans cannot, like a 25 car "train" traveling 100 mph down freeways like I90 in Eastern WA, from which vehicles in specific locations can either join or disembark with no human intervention, or Semi convoys doing something similar. It may soon be that high speed rail is no longer needed. Or traffic lights connected with a city version, allowing traffic to travel across town in a fraction of the time.
It boggles the mind when you think of what vehicles connected to each other with FSD can do. And if/when the technology gets licensed, it can be any auto manufacturer (that may take a while)
I've found it pretty close to miraculous in function, though as you point out not perfect. But... in terms of utility? I'd be OK with it if it had just come with the car, but not really seeing the practical use since I essentially have to keep my hands on the wheel and pay at least as much attention as I did without it. Also... it mainly drives like grandma.
If and when it's good enough to make RT's work, it will be worth something. As it stands, it's a luxury option of the ilk I normally don't spring for, namely those with only modest benefit. As an investor I do hope it is a profit center at $99/mo. I also think the other monthly, for Premium Connectivity, is a little discussed profit center for Tesla that is the real deal. Useful and pretty much everyone needs it and would use it.
Yep, every state requires J3016 as a defining standard, but J3016 is NOT itself a requirement. Find the doc here: SAE Levels of Driving Automation™ Refined for Clarity and International AudienceThat is not correct.
Florida specifically requires meeting numerous, specific, requirements for a self driving system, all of which use the definitions and terms taken from J3016.
FL law here:
Read down to section 3- Automated driving system.
They define numerous specific features (like dynamic driving task and operational design domain), all of which take their definitions directly from J3016 and the system must meet those requirements to be a legal automated driving system in Florida.
Further down they use another J3016- "achieve a minimal risk condition" with a pointer to where that is defined ni FL law, which is here:
Where again they use J3016 definitions for both the L3 version of DDT fallback (human must respond to an alert to take over) and the L4 version (vehicle can achieve a minimal risk condition without a human, and the J3016 definition of what minimal risk condition is.
So yes- quite a bit about how it does it. All using the specific language and definitions from J3016.
8.1 talks about this:
Also from this statement, I conclude that the current model could be put into another mode than L2 given a few feature that I'd bet are currently being validated like L3 handoff/fallback and L4 "all foreseeable conditions in the ODD" like road closed and advisory sign handling and L5 unlimited domain
8.3 is also a great section to demonstrate levels are features and 'feature sets' as opposed to improvements.
I agree, that's why I feel goin all in on RT and skipping the consumer Gen3 is a crazy decision, and I hope Elon has not decided to go this route.
Hmmm.... this Does not compute, Dr. Smith.
Unless AI Driver was making it up that U-Turns were a byproduct of more data, there was never a function created to perform this maneuver. So why wouldn't this also apply to object avoidance on the road?
This is precisely what was meant by Tesla AI having a "ChatGPT moment", and I'm quoting the CEO of NVidia this past month. The work and $ needed at Tesla will likely focus on even more compute. Possibly more than they thought since Microsoft announced that $100B investment.
Knightshade
Well-Known Member
It does if you understand it.
First, OEDR complete is more than just "avoid objects on the road"
It's also "know how to react to a stopped school bus, including in context of the sign being out or not, and if the road is divided or not."
It's also "understanding any non-standard speed sign"
It's also "understanding parking signs"
and a slew of other things.
It also ignores the OTHER two things you need to get >L2.... the DDT fallback ability (which also is more complex than just "train more"- the system has to understand WHAT failing safely even is, then know how to do it in a variety of contexts and failure states0....and a defined L4 ODD (which isn't a training thing at all-- it's a thing Tesla has to write based on the limitations of the system... this is a definition that can potentially get EXPANDED over time with training- for example being able to go from "clear weather only" to "can use in light rain" if the algos for vision through rain improve. But it has to exist (and can't until you get the other two things working first).
You realize at this point "ChatGPT moment" is more about "Average people thinking a thing is WAY better than it actually is yet" right?
ChatGPT still entirely hallucinates imaginary references and gets things wrong frequently.
Much like an L2 ADAS system- it can be very useful if you understand its limitations, but it's nowhere near ready to remove the human from the loop.
jschwefel
Tesla fan/TSLA, Model Y and Cybertruck owner.
I don’t know about sounding the alarms, but inquiring minds would like to know why the numbers are down.? Tesla is still a relatively small auto manufacturer, and it’s at least curious why growth has stalled. One would think given the size of the auto market, the quality and price of Tesla’s products, particularly the Model 3 and Y, would continue to drive sales growth. Unfortunately, that’s not happening.
Knightshade
Well-Known Member
Your system must comply with J3016 because the law says it must. I'm not sure "Compliance is because the law says so, not because SAE says so" is a useful distinction.
I can assure you I've read it more thoroughly than the two folks (not you) trying to debate it with me
In fact I've provided links to it multiple times in the last couple days in the discussion.
None of those things are being validated in any public version of the software- those things don't even exist in them.
Nor were they in evidence in the Elon version of V12 he livestreamed.
Again one can hope there's some WAY more advanced version in secret internal-only-not-known-to-anyone-even-the-employee-version versions, but I continue to find hope a poor investment strategy.
Gigapress
Trying to be less wrong
People keep saying it's a "bet the company" situation, but I think that's hyperbole. OG roadster, S and 3 were true do-or-die product launches. Nowadays, Tesla has a comfortable lead over the competition and plenty of cash, thanks to their prudent rejection of the shareholder demands to buy back stock last year to prop up the $TSLA price. They can afford to make a bold bet on autonomy and come up empty. At worst it will cost the company around $10 billion and set back the mission by a couple of years.
Gigapress
Trying to be less wrong
I think we need to bear in mind that this company is led by an autistic man who has a strong preference for literal, precise, matter-of-fact communication and a strong distaste for any form of deception, including white lies that neurotypical people find to be acceptable. The usual social games do not apply. By his own account, as a kid he learned (slowly, gradually, painfully) that people don't always say exactly what they mean only by consciously noticing behavioral patterns in characters in books and movies, as is a common experience for intelligent people with autism. We may learn more, but I really doubt Elon is playing the normal games about disgraced executives leaving to "retire" and "spend more time with family". Also, this particular executive is a monumentally successful 41-year-old who has little kids who are probably wondering why Daddy is never home. Most people don't reach this level of a corporate management hierarchy in a company as big as Tesla until they are in their late 50s or 60s. As far as I know, Drew was second only to Gwynne Shotwell in terms of how long he tolerated having Elon Musk as his boss. With every passing year, the likelihood of him quitting has been increasing. Unfortunately, this is the year.
Last edited:
Who said it was?
But ask me again next quarter as it’s still too soon to feel this latest FSD growth curve. Looks steeper so far but need a couple more updates at least.
Let’s just say we are at odds on this. It seems you didn't address my main point then threw in the LLM mess created from the ugliest data I can think of called the web.
Good luck in whatever your goal is here.
Musskiah
DisGruntled
In 2024, weeks 14 and 15 (first 2 weeks of April) are quite similar to weeks 1 and 2 - Q2 is tracking within 9% of Q1 registrations for the first 2 weeks combined. The cumulative registrations (LR line graph) 2023 vs 2024 cumulative registrations are tracking -9%, which is less than 1k registrations. As we saw in my previous post, WuWa shared in Southport Shanghai, there was an enormous number (at least 10k) of Teslas awaiting export.
Some of these are undoubtedly headed to Chile, the third largest car market in S. America that Tesla just entered into. The EV Market in Chile is projected to reach a revenue of US$149.2m in 2024 with a 4 year 17% CAGR. Just because there are less registrations in China does not mean that inventories in China are rising. In fact, there is the lowest inventory of all in China precisely because they can divert any excess inventory easily to exports. If we investigate Tesla inventories, we will find China is an order of magnitude less than the US and several times less than Europe. This has benn true for years and remained consistent in the most recent quarters. In light of inventories remaining very small, Q2 registration data that closely tracks Q1 in China is not concerning.
SOURCE: Electric Vehicles - Chile | Statista Market Forecast.
Last edited:
2018 (January) Model 3 RWD LR in California here.
What's your hardware ?
Typical mistakes it makes for me is moving over into right turn lane when planning to go straight. Couldn't try if it would actually turn as there were other cars I couldn't let it cut off.
Map showing only one of two stop signs, signaling right on first stop sign with intent to go straight and turn right at the next one.
Also cutting over opposite traffic lane when turning left, rather than staying clearly in its lane while turning. I had to take over to ensure no colission today. Maybe it wouldn't have collided but it sure did freak me and probably the other driver out.
My wife in 2020 model x (also still intel based mcu) also reported having to take over at stop sign where it was trying to take other cars right of way.
I also had interventionless 20 mile drives door to door on other routes. Most of the time very impressive and capable but regressed compared to 12.3.3 for me.
Similar threads
- Locked
- Locked
- Locked
