Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register
This site may earn commission on affiliate links.
And yet, with all that data and testing one can’t get a bit closer than ‘somewhere between 1/2 and 1 order of magnitude’.. sounds more like spit balling to me.

The estimate likely comes from simulations. 12.4 isn't going out to employees until this weekend, at which point they'll have some real-world data on it. And then once we get our hands on it, Tesla will have a precise measurement of the number of miles between disengagements.
 
The estimate likely comes from simulations. 12.4 isn't going out to employees until this weekend, at which point they'll have some real-world data on it. And then once we get our hands on it, Tesla will have a precise measurement of the number of miles between disengagements.
Which they will never tell us !

ps : Elon after all these years continues to believe at any moment now, FSD will exponentially become better quickly. That is the whole reason why every year he has claimed FSD will be achieved that year.

Up to a magnitude of improvement in miles per disengagement from 12.3 to 12.4, if true. And single-track (end-to-end on highways) by 12.5:

MPI - not MPD. We should start using this.
 
Last edited:
I cannot wait to see those glorious paid enthusiasts releasing new vids declaring “FSD IS THE GREATEST WITH NO INTERVENTIONS!” Then an actual paying owner finds out they are completely FOS. Reckoning is coming. My car has tried killing occupants over and over again since testing started in 2022.
Shills like Whole Mars have been releasing hands free no intervention drive videos for years now. Nothing new.
 
I cannot wait to see those glorious paid enthusiasts releasing new vids declaring “FSD IS THE GREATEST WITH NO INTERVENTIONS!” Then an actual paying owner finds out they are completely FOS. Reckoning is coming. My car has tried killing occupants over and over again since testing started in 2022.
Interesting as your previous post stated you just got FSD 2 weeks ago? Apparently in your imaged theory it could have happened since 2022?

“I am having the exact same experience as the OP and I just got FSD today. Terrible disappointment after seeing all of the MarsCatalog and DirtyTesla paid reviewers singing such praises. I decided to renew it for the first time in a year. Ehhh wrong answer. Wrong idea.”
 
Interesting as your previous post stated you just got FSD 2 weeks ago? Apparently in your imaged theory it could have happened since 2022?

“I am having the exact same experience as the OP and I just got FSD today. Terrible disappointment after seeing all of the MarsCatalog and DirtyTesla paid reviewers singing such praises. I decided to renew it for the first time in a year. Ehhh wrong answer. Wrong idea.”

I bet this is something less than a Columbo worthy gotcha.

Screenshot 2024-05-23 121450.png
 
  • Funny
Reactions: Wnuk
Last edited:
That sounds less like a "spanking" than it does Yann LeCun asking a question about FSD's internal architecture.

We don't know if it uses CNN "much" and neither does Yann LeCun.
You need to catch up on the full thread... Also, ConvNets, which LeCun "invented" are very much the foundation for computer vision, and Elon should know that if he's getting into a pissing contest about it.

Screenshot 2024-05-28 at 21.53.08.png
 
Last edited:
  • Like
Reactions: RabidYak and DanCar
You need to catch up on the full thread... Also, ConvNets, which LeCun "invented" are very much the foundation for computer vision, and Elon should know that if he's getting into a pissing contest about it.

The full thread is just a pissing contest. But that doesn't mean it's a "spanking" or a "slamming" or whatever other gerund the Verge or Jalopnik will inevitably pick when cobbling together an entire article about a few tweets tomorrow. Elon is obviously just saying things to egg LeCun on.

Yes, it's unlikely that FSD processes images without the concept of convolution. But it's also possible that FSD was once largely dependent on the exact kind of CNN architectures proposed by LeCun, and now with V12 they only use image kernels for things like down sampling or edge detection before passing the task onto something else like a Vision Transformer.

Should Elon acknowledge the work of academics that his businesses are built upon? Absolutely. But are you surprised in the slightest that he's not?