I had this reply as a draft from yesterday, replying to your post Yesterday at 8:15 AM . Didn't see the new post until the page refreshed.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
All US Cars capable of FSD will be enabled for one month trial this week
- Thread starter Ron97062
- Start date
I had this reply as a draft from yesterday, replying to your post Yesterday at 8:15 AM . Didn't see the new post until the page refreshed.
Half of the cars aren't even in US/Canada to use FSD anyway. Almost all getting 2024.8.9 were already on some 2024.8.x, and that mostly went to those outside of North America. But indeed there's some vehicles capable of FSD that need to wait for Tesla to update 12.x to be on 2024.8.10 or newer to actually deliver on the free trial. These US/Canada 2024.8.x vehicles seems to be the next largest group still waiting on FSD Supervised, so presumably Tesla is well aware and working on the update and validation for deployment.
Got a source for this? At least 60% of the cars I know of in Seattle are on 8.X. Teslafi shows the highest density of 8.X is in the USA. I see no such data indicating that 8.X is a mostly non-USA version. 8.9 is being rolled out hard in the USA right now.
Why was Tesla not "well aware" 2 weeks ago when they said "All US cars that are capable of FSD will be enabled for a one month trial this week"? I mean why would they say such a specific thing if they knew is was impossible?
I have two words for you: Elon. Musk. He has a very consistent history of making promises he can't/doesn't deliver on.
Meh they probably enabled it on a specific fork of software that you haven't gotten yet.. still technically enabled that week
That is one of the more creative pro-Musk interpretations I have ever heard.
"Tesla has written the software, so your car is "enabled" for it. We're just testing and validating it, and not putting it on your car"
I mean in that case, my car was "enabled" for FSD in October 2016.
Why call out only US cars if Elon is such a precise, pedantic communicator? I mean, if he didn't actually mean all people in the US would be able to actually start a free trial "this week" but instead just means the SW exists somewhere and needs to be released, that is true for every car no matter what country.
I really doubt a court of law would agree that "All US cars were Enabled for a free trial the week of March 25th" given 50% of owners in the USA have not been able to actually take advantage of a free trial a week later.
You likely didn't have FSD enabled in 2016 unless you paid for the lifetime which doesn't sound like you did or you'd have it now still
Anyway.. You should definitely sue again
I’ve seen nothing MYLR 2022 nor email to suggest Supervised FSD. Everything I’ve read suggests at least a large 40% won’t see it for a long time just like V. 11, and this was simply a marketing ploy to spur those on the fence into buying one considering their overstock problem.
TeslaFi general audience is biased towards USA as across all software versions, about 60% of its users are in US/Canada. Focusing on 2024.8.x, it does show United States is the most common country, but as a percentage, it's "only" 20% with the rest spread across 65+ other countries. Part of the reason for the lower percentage is that many are already on 2024.3.x with the free trial, but other older versions like 2023.44.30.8 are around 90% US/Canada presumably Tesla intentionally not pushing these vehicles to 2024.8.x.
Perhaps your audience of cars in Seattle is quite biased in a different way as looking at TeslaFi Washington data, around 60% already have 12.x while 15% are on 2024.8.x. If they're mostly older S/X vehicles that you know of, that could explain the higher-than-average on 2024.8.x, and hopefully Tesla gets all of you 12.x soon.
2017 Model X, 2018 Model 3, 2022 Model 3, 2023 Model 3 are all on 8.X.
Can you share how to look at TeslaFi data down to the state level? If you can do that, you can easily see what exact percentage in the USA is on various versions. What % of the USA is on 8.X vs 12.X?
One would note that 60% is still 40% away from "all US cars."
Only Tesla has those numbers, but of those who are on TeslaFi showing up as United States with HW3/4 on the top 10 common versions (includes nearly 95% of TeslaFi vehicles):
65% have some FSD 12.x
16% are on 2024.8.x
10% are on some other 2023.44.x
8% are on 2024.2.x
There are still pending installs of the various 12.x, but that would at most increase the percentage to 70% (if they were coming from non-12.x versions which isn't always true). It's not clear why there's still roughly 15% vehicles on older versions that could be pushed to 2024.3.10 but haven't yet. Potentially these older version vehicles haven't manually selected the software screen to check for updates, but maybe if they did, 12.3.3 could show up?
If you really want to pick on the "All," there's vehicles like the new Model 3 and Cybertruck that aren't getting 12.x yet either.
How do you get TeslaFi to give you this geographic data? I cannot find it anywhere on their site.
From a given version page like 2024.8.7, the bottom table has individual vehicle entries that indicate the Country, so you can count those up if you have specific other details to check like Model. Towards the middle of the page has "Current Installs" with a map where just below that, you can expand "+All" to see per-country numbers split by HW too. Currently this version shows 7,180 for all countries and 1,299 for US (independent of HW).
BitJam
Active Member
Tesla was already giving new owners one month of free FSD. Giving everyone else (or half/most of everyone else) a free month probably didn't sway any buyers who were on the fence.
I guess you missed the news that Elon said their FSD training was no longer compute constrained which means they need more data to make the best use of their existing multi-billion dollar servers and all the new ones that are on the way.
This seems like an entirely different situation than with v11. Almost the opposite. With v11 Tesla benefited from restricting the release. More testers meant more liability but not more training since they were compute constrained.
Another big related difference is they have rebuilt the inference engines in the cars to be end to end neural nets, removing 300K+ lines of code. This means the training is mostly limited by data and compute, not by things like human labeling. In the past Tesla had over 2,000 human labelers which was another training bottle-neck.
The urgency Tesla has to release v12 to people on the 2024.8 branch is probably related to how quickly they can continue to ramp up their training compute. Personally, I hope they are also waiting to get some of their outdated maps in rural America fixed. Where I live no version of FSD can drive me to the grocery store because the 2023 Tesla map is many years out of date.
The amount of data and compute involved in giving everyone (or most of everyone) free FSD is mind boggling. This could be one of the biggest computation projects in history. The idea that Tesla did this to momentarily raise the stock price or reduce inventory or anything like that seems silly.
Tesla is a big tech company. And with big tech companies it is almost always all about the data, especially when they give you things for free.
Everything I've read and seen for years says Tesla is serious about getting FSD to actually work. Even if they are wrong or delusional, that is their plan. Spending many billions of dollars to beef up their training computers was part of that plan. So was a wide release of FSD v12. Compared to this goal, the stock price or inventories in the spring of 2024 are meaningless.
zoomer0056
Active Member
I agree, the month free was not a marketing strategy. It's for getting more data input for system development. It is mind boggling, for even just this month, how much data is going into the system. I am happy to be a part of this.
Last edited:
At the risk of upsetting some people (that isn't my intention), how statistically relevant is Teslfi? From what I can tell they have MAYBE 3-4,000 active users. Tesla has sold about 5 million cars worldwide and about half of that in the US.
I would imagine that most Teslafi users are people who have had their cars for more than a couple of years (my speculation only - nothing factual to back it up).
So, I question the statistical relevancy of the Teslafi numbers.
I would imagine that most Teslafi users are people who have had their cars for more than a couple of years (my speculation only - nothing factual to back it up).
So, I question the statistical relevancy of the Teslafi numbers.
1) As you say, "Elon said" they were no longer compute constrained.
2) Elon is a known liar. See "All US cars getting FSD trial this week." See "Cybertruck can tow 911 over quarter mile faster than 911." See "funding secured." "All cars have all HW needed for FSD." "That guy is a pedo" "Maps for self driving are a very bad idea" Etc.
So, why trust Elon on statement #1, when there is a much larger set of statements from Elon that are easily proven false? Dismissing Elon's "this week" statement and then turning around and using his own statements to explain why "this week" wasn't marketing hype is peak dissonance.
Let's also not forget that Elon has specifically said that Tesla is "worth basically zero" without FSD, and that Elon is specifically paid when the stock hits specific short term goals, he's not paid on long term performance. Elon is a master market manipulator, and does everything to manage stock price, because that's how he gets paid. He has massive reasons to hype FSD progress, and history shows he has no issue making massive mis-statements that benefit him and his companies' short term prospects.
So it's perfectly logical to decide that the sudden push to bring FSD into the spotlight in April 2024 right before very bad quarter results was more about money than it was about technical readiness. Because for sure we know one thing, which is that Tesla was NOT technically ready to do what Elon said, which is put FSD on every US car two weeks ago. But yeah, I'm sure all the other things Elon says is true, and he's highly informed on Tesla's technical capabilities and progress.
Right, which is why FSD is $12K and they are only giving a one month trial to 50% of the owners.
The other hilarious irony is that if you think giving a month of FSD away will massively increase the data they get, it means that almost nobody up to this point has been subscribing to FSD. They've been selling FSD for 80 months now. One month of 50% of the cars having it should not dramatically help data collection unless only a few percent of cars have it otherwise. Wouldn't they be much, much better off to have a $3K fire sale on FSD and get lots more cars permanently set up with FSD instead of half the cars for 4 weeks? Wouldn't they be able to collect massive amounts of data even on normal AP or hand driving? Wouldn't they be better off spreading out the time exposure of the fleet instead of collecting all their data in one specific month?
This is not the move of a company that values data more than the actual income the sales of the product provide and the bump the hype gives to the stock.
How much data has your car been uploading after every drive?
About 10% of TeslaFi users have AP HW4, which has only been shipping for a year. There's even 50 cybertrucks in there. So the user base is not really statistically biased to older users.
But the simple answer is that they have 20K cars reporting their SW version, and that's the best data we have, and that's a large enough number of vehicles that it probably reasonably represents the larger status of all Teslas in the world.
You have no choice as a user what software you get, so I don't obviously see a reason that the self-selection of TeslaFi users would impact what SW the cars have or any other self-selection biases. Technically, Tesla could know that you're reporting to TeslaFi but it's hard to believe they would send your car different SW due to that.
I think the only bias in TeslaFi data is that probably almost all of them have their Software Update set to Advanced so would be more likely to be on a newer version. So possibly there is a higher percentage of cars on 2024.8.X than there is in the overall fleet. Otherwise I think it is probably a fairly good data sample of the fleet.
Similar threads
