Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

Tesla, TSLA & the Investment World: the Perpetual Investors' Roundtable

This site may earn commission on affiliate links.
While the thread is locked, a reminder:
All have been told that requisite to participating in this thread is subscribing to Moderators’ posts.

This occurred not for some (truly bizarre) vainglory. Rather, it was so that no participant could whine “Oh! But I hadn’t seen that directive/admonition/warning…gee, the Mods should have warned me because - well, because I’m special .”


Now, here’s a poorly-kept secret: Mods know very well most participants have not so subscribed. Not having done so, however, means the writer of a deleted post first, has no standing for complaints, and second, has no standing for asking “But what did I do wrong? I don’t even know which post of mine was deleted” (yes, we even get statements like that).
What do you mean subscribe to a moderators posts?

If it's the feature I think you are talking about it just adds to the hundreds of notifications I get and it's mixed in / lost in the noise of all the likes/dislikes/informatives/helpfuls.

I see your posts as I read through the thread, if I'm a few pages behind I won't see them until I get to them.

And if you edit someone elses post after I read it, I might miss your edit and never see your post.

and FWIW I didn't even think about posting in the last few pages, it was off the rails.
 
Last edited:
Skimmed through the release notes for 10.12. The two I found interesting was:

Improved geometric accuracy and understanding of visibility by
retraining the generalized static obstacle network with improved
data from the autolabeler
and by adding 30k more videos clips.

Improved detection recall of far-away crossing objects by
regenerating the dataset with improved versions of the neural
networks used in the autolabeler
which increased data quality.

So here we see what’s going on behind the scenes. Tesla are using neural networks to generate their dataset to train neural networks for the cars. Larger neural networks than they can run in realtime in the car. Using these they generate an increasingly large dataset of labels to videos. These networks get improved by new labels for how to label. Now they have gotten to the point that they saw the benefit of regenerate the labelling on old videos to generate new better labels. This was probably a massive computational task so they don’t do this every time, most of the time they keep the old labels.

Their pipeline is getting increasingly complex and large. But they have the best AI team in the world and many optimizations is happening all along the pipeline where labelers are now mostly training the autolabeler by correcting where it is uncertain or where they find failures. As the autolabeller gets more and more accurate, they have to correct less and less often and can autolabel even more data.
 
Good post @Words of HABIT.

I know I just talked about people trying to explain a drop and never knowing why, but IMO the weakness in TSLA lately has purely been driven by macros and the Shanghai closure. Investors now know Q2's production/deliveries won't be as good as Q1 (through no fault of Tesla), and are just front-running that. I think they'll jump back in once that rumor becomes news.

On the macroeconomic front, there's been discussion of shipping costs coming down. I think that, coupled with the Shanghai reopening, put us very close to a bottom now. I expect (hope) that the next numbers on inflation start looking positive. The minute they do, I think people may start flooding back into the market in droves.

And you don't want to be the guy that missed out on that because of a silly Tweet.

PS: I'm not sure if anything above is correct. But it makes me feel better. And frankly, this decade's been not-so-great so far, so I could use the pick-me-up!
I think most of us could use a pick-me-up right now.
I recommend "Countdown" on Netflix, the Inspiration4 mission documentary. I just watched the 1st 3 episodes (of 5).
Human, intriguing, inspiring, uplifting. (Very little Elon, for those who are annoyed with him right now). Get a taste of a future where so much more will be in reach. For all of us.
Do yourself a favor and pull it up, sit down with your beverage of choice, and take in wonder of it all.
 
Back after the original CT reveal Tesla said they'd be doing a design to handle trucks with trailers, pull through stalls so you could place the vehicle where it needed to be and leave the trailer on--- that was a few years ago now but I haven't seen them actually do it in any of the new chargers I've been to anyway, only thing I've seen is slightly wider stalls that might have CT in mind.

So at least at launch of CT there's gonna be a massive % of SC location where you'll need to remove the trailer, then pull into a stall to charge, then hook the trailer back up.
Some sort of extension from the charge port to the right front corner of the vehicle would allow PULLING INTO the charging stall to address the trailer problem, albeit in an awkward, traffic-blocking manner...
 
  • Like
Reactions: GOVA and Nocturnal
...If by 2024 FSD isn’t ready, at least for Loop usage, then robotaxi production can be postponed easily enough. Tesla is agile and flexible enough in their plans to accommodate that. Or they could just quietly build up the fleet and keep it in inventory in anticipation of the most profound single bit flip in the history of computers.
How can FSD not already be ready for loop usage when it hardly has to do anything more than drive straight??? What's it gonna do, make a wrong turn in the tunnel???
 
How can FSD not already be ready for loop usage when it hardly has to do anything more than drive straight??? What's it gonna do, make a wrong turn in the tunnel???
I am very much in the Loop will be easy for FSD camp, although the biggest question is whether the local government will allow it by 2024 because no timeline has been provided by officials.

If we want to talk more we need to take it to either the Boring Co thread or Engineering thread per mod direction.
 
What do you mean subscribe to a moderators posts?

If it's the feature I think you are talking about it just adds to the hundreds of notifications I get and it's mixed in / lost in the noise of all the likes/dislikes/informatives/helpfuls.

I see your posts as I read through the thread, if I'm a few pages behind I won't see them until I get to them.

And if you edit someone elses post after I read it, I might miss your edit and never see your post.

and FWIW I didn't even think about posting in the last few pages, it was off the rails.
If you go into your profile settings you can disable all the annoying reaction notifications.
 
If you go into your profile settings you can disable all the annoying reaction notifications.
I actually find them useful for opening threads between my home and work PCs / figuring out where I am in a fast moving thread.

If I turn off all notifications other than moderators I'd also miss it when a thread I've subscribed to had new posts.
 
  • Like
Reactions: Gigapress
Even though I know a bunch about NNs and how much or how little this may or may not improve my unique experience with my Tesla....the proof will be in these three areas:

1. Unprotected 55 MPH left, obtuse downhill obstructed oncoming traffic T-intersection (current behavior is "Needs to be rethought")
2. Several round-a-bouts that consistently exhibit erratic steering wheel movements (current behavior is "Needs work")
3. Random false positives (ghosts) which result in unknown abrupt slow downs. (current behavior is "Needs work")

I don't think any of these will be 'great', but hopefully better. The possibility of them being 'great' is when we get a unified vector space.

My definitions
Great = my wife will allow FSD to be on.
Good = When the follow car does not think I'm a bad driver
Needs work = Erratic, inconsistent behavior that annoys good drivers, but still safe
Needs to be rethought = Safety issues...
 
Improved detection recall of far-away crossing objects by regenerating the dataset with improved versions of the neural networks used in the autolabeler which increased data quality.​

… This was probably a massive computational task so they don’t do this every time, most of the time they keep the old labels.
I'm a little bit surprised that Tesla might not have been continuously relabeling with even better ground truth data as while the autolabeler has the benefit of seeing across time, it still needs to work with the current (offline/larger) neural network's prediction accuracy and confidence. A simple example is say the neural network can predict with high accuracy an object 100 meters away, so seeing into the future 10 meters can train the network to have higher accuracy at 110 meters. Relabeling with the new network could then result in better labels for objects that are 120 meters away (of course with diminishing returns). That's probably what allowed for this improvement for "far-away crossing objects," and I would expect this type of iterative improvement to allow predicting many things such as intersection extents and lane assignments from even further away for better lane selection, etc.

As you pointed out, this is probably quite expensive to reprocess existing videos that already have pretty good labels, so compute resources were probably better allocated towards improving predictions that were still wrong even at close distances. This might indicate Tesla believes the quality of neural network predictions are good enough to now focus on extending the range, which should help with comfort and smoothness as seeing further allows for better planning with fewer last-second adjustments.

This could also mean Tesla has more compute available now perhaps with Dojo improvements?
 
  • Like
Reactions: petit_bateau
The Finnish Air Force uses a Swastika. Is the Finnish government controlled by Nazis?

View attachment 805948

FYI: "The swastika was officially taken into use after an order by Commander-in-Chief C. G. E. Mannerheim on 18 March 1918. The FAF changed its aircraft insignia, which resembled the unrelated swastika of the Third Reich, after 1944, due to an Allied Control Commission decree[5] prohibiting Fascist organizations.[5] It nevertheless continues to feature in some unit emblems, unit flags and decorations, including on uniforms. In 2020 the BBC reported that the FAF had "quietly stopped" using the symbol in the emblem of the Air Force Command."

 
This from an email received moments ago speaks volumes for anyone getting skeered over the SP decline

AnalystRatings.png

Perhaps knowing that there is migration from Sell to Hold and that there has been a steady increase in the Buy and Outperform columns lately will offer some solace to everyone who misses their opportunity to sell at the bottom. 😉

HODL!