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.
I'm taking a road trip from Los Angeles to Phoenix next week and it would be so much nicer to have the update just for the 85 AP speed limit.
The more frustrating part about not receiving all FSD beta updates for me is more about all the features on the stable branch that I haven't received.
 
  • Like
Reactions: beachmiles
Any examples of that ? People keep accusing influencers of "editing" videos, even when they post raw 1x videos - but hype obviously professionally edited Waymo, Cruise & MobilEye videos.

Here is an interesting video about "click bait titles" by a physics channel I watch. Youtubers experiment with multiple titles and use the one that gets promoted by YouTube most. Blame YouTube algorithms.

It’s not like Tesla and Charlatan used a professionally edited video to defraud customers 🙄
 
  • Funny
Reactions: Ramphex
...
Prior to widening the Beta, it’s probably fair to state that most videos were edited or showing curated runs the testers knew went well and likely where routes were built to avoid problem areas and portray the system as more proficient than it was.
...
I started following the YouTube FSD beta community well before the wide access / safety score program, and I would disagree pretty strongly with the claim that they were deliberately skewed or filtered at that time. I'm speaking of Dirty Tesla, Frenchie, Chuck Cook, Rocco Speranza and AI DRIVR (the last seems somewhat less active now but still posts occasionally).

On the contrary, I think that all of the above conntributors, along with some of the newer ones, have been remarkably straightforward and go out of their way to present the problems that they find. This is as it should be, And I can't really understand others watching the same videos and thinking that they are "shilling". There's no doubt that far more of each video is spent documenting and commenting on failures compared to any gushing over the car working correctly, which of course it does most of the time but is far less interesting. Yes, there is some excitement and gushing when the car performs well in a scenario that it previously failed consistently - Absolutely nothing wrong with that and in fact I wouldn't want it any other way. It adds to the interest and fun, and we need to know the editorial opinion as well as the news reporting of the FSD progress

I also give a lot of credit to Tesla for allowing this to play out In public. You're not going to find this with any of the other players, except probably comma.ai which is an inherently hobbiest/enthusiast product. A possible exception to this praise of Tesla's openness might be the case of AI Addict, who posted some pretty damning FSD beta failures while being an actual employee of Tesla. I have to reserve judgment on that one because we don't know all the background story, including the actual terms and conditions that he was operating under.

And the openness continues in general, even though it provides plenty of ammo for detractors. That guy Dan O'Dowd paid to create and publicize a highly negative and highly distorted video using this YouTuber content - that has to be the prime example of misleading edits to further an agenda, followed by wild and statistically senseless assertions that belie his claimed technical expertise. Compared to people like him, the FSD beta testers on YouTube are incredibly qualified and even-handed.
 
I'm just trying to say that individuals can not really accurately estimate how well FSD is doing overall.
While I agree with the entire post, I cherry picked this statement for a reason...

This is kind of a re-post on my part, but it has been a while since I've talked about it. With that disclaimer out of the way...

FSD Beta is nigh on unusable where I live. HOWEVER, if I drive 135 miles south on I-15 to Las Vegas, it behaves completely differently. I have far fewer interventions and disconnects using FSD Beta in 'Vegas than I do here.

People tend to think the experience they have with FSD Beta in their local area is an accurate representation of the current overall state of FSD. This is not correct; it will do better in certain geographies than it will in others.

Peeps should keep this in mind, as I know that if I used my own experience using it locally as my only source of judging how well/poorly it works, I must say I'd be pretty disgruntled over all.

As it stands, I remain cautiously optimistic. I think Elon and the boys are going to get us there, but IMO we are at least one hardware (and camera suite) iteration away.

IMO, Dojo will tell us a lot. Once they're able to get a Dojo training suite up and running, it'll really let us know how far the current hardware, cameras, and software are going to take us.

Edit: For those that don't know, Hardware 3.0 has two redundant FSD computers. They were supposed to run the same code simultaneously, and develop independent decisions, comparing their output against each other for redundancy. Unfortunately, they ran out of compute, and are now using the second FSD computer as additional computing horsepower rather than its intended redundancy use.
 
Last edited:
Now, if someone is really trying to present FSD beta versions unrealistically, they could certainly drive routes multiple times until they get a good, or bad, result, and show only the one that best serves their purpose.
Anyone wondering how Waymo, Cruise, MobilEye and other AV companies can put out flawless videos - here is your answer. Even when they post unedited "raw" videos - they can carefully select the video to post. Moreover those billion dollar companies obviously have the resources needed to do multiple runs and post the best one. Unlike say WholeMars / Omar who posts so many videos, its virtually impossible for him to do multiple takes before posting every video.
 
  • Like
Reactions: FSDtester#1
People tend to think the experience they have with FSD Beta in their local area is an accurate representation of the current overall state of FSD. This is not correct; it will do better in certain geographies than it will in others.

Here is the crowd sourced data ... though I doubt in reality the variation is as much as in this report.



1653957675521.png
 
While I agree with the entire post, I cherry picked this statement for a reason...

This is kind of a re-post on my part, but it has been a while since I've talked about it. With that disclaimer out of the way...

FSD Beta is nigh on unusable where I live. HOWEVER, if I drive 135 miles south on I-15 to Las Vegas, it behaves completely differently. I have far fewer interventions and disconnects using FSD Beta in 'Vegas than I do here.

People tend to think the experience they have with FSD Beta in their local area is an accurate representation of the current overall state of FSD. This is not correct; it will do better in certain geographies than it will in others.

Peeps should keep this in mind, as I know that if I used my own experience using it locally as my only source of judging how well/poorly it works, I must say I'd be pretty disgruntled over all.

As it stands, I remain cautiously optimistic. I think Elon and the boys are going to get us there, but IMO we are at least one hardware (and camera suite) iteration away.

IMO, Dojo will tell us a lot. Once they're able to get a Dojo training suite up and running, it'll really let us know how far the current hardware, cameras, and software are going to take us.

Edit: For those that don't know, Hardware 3.0 has two redundant FSD computers. They were supposed to run the same code simultaneously, and develop independent decisions, comparing their output against each other for redundancy. Unfortunately, they ran out of compute, and are now using the second FSD computer as additional computing horsepower rather than its intended redundancy use.
What proof do you have that the second CPU is being used because of lack of computing power. I want a real source related to Tesla, thats not just someone on twitter. Because the only place people talk about there not being enough computational power is bullshitters on twitter.

the entire purpose of the second computer is shadow mode. Not to have 2 computers deciding the same thing. thats not how computers work.
 
Define: Shill
a person who publicizes or praises something or someone for reasons of self-interest, personal profit, or friendship or loyalty.

I don’t see what’s so “disrespectful” about it.
Example…

I got banned from the tesla subbreddit because I called someone unplugging multiple peoples cars while charging at a J1772 set of chargers “ghetto trash.” I was banned because it was considered racist since they weren’t white.

It’s not about the words dictionary definition. Definitions change from what they are In a dictionary. These people are doing something they love, and you’re finding derogatory words to put them down and group them together.

It’s not about what YOU consider disrespectful.
 
  • Like
Reactions: Yelobird
The bottom line is, we need those people. For ever Dan O’Dowd, we need 5 whole mars to maintain a good public perception on FSD and self driving, or public fear will destroy any possibility of it being accepted.

And if you don’t know who Dan O’Dowd is, good
 
What proof do you have that the second CPU is being used because of lack of computing power. I want a real source related to Tesla, thats not just someone on twitter. Because the only place people talk about there not being enough computational power is bullshitters on twitter.
Tesla has covered both of these topics publicly in their various "x day" presentations. They are definitely worth watching in their entirety.

In the one video where they explained Hardware 3.0, they specifically said that they both ran the same code independently and simultaneously, for output comparison/redundancy reasons. I believe this presentation was "Autonomy Investor Day"?

And it was in the "AI day" that they discussed using the second FSD computer for additional compute.

Edit" Here's a timestamped video of where they explain they are now using the second FSD computer for additional computing power in the "AI Day" presentation:

Another Edit: Here is a timestamped video of Tesla's "Autonomy Investor Day" where they explain that both FSD computers run as I described in my previous post: they both run the same code, achieve an "answer," and compare the answers against each other

the entire purpose of the second computer is shadow mode. Not to have 2 computers deciding the same thing. thats not how computers work.
Please see the above video at the timestamp.
 
Last edited:
Tesla has covered both of these topics publicly in their various "x day" presentations. They are definitely worth watching in their entirety.

In the one video where they explained Hardware 3.0, they specifically said that they both ran the same code independently and simultaneously, for output comparison reasons.
That’s exactly what I’m saying. But both weren’t used to make 1 decision. Both weren’t used to control the car. The second one was used to analyze instances and simulations of what the other computer had just done.

Shadow mode still exists. That’s what the second computer is for. Not to double the processing power of what’s driving AP. It’s 2 identical computers. You don’t just combine processing power of the 2 into 1 process.
 
That’s exactly what I’m saying. But both weren’t used to make 1 decision. Both weren’t used to control the car. The second one was used to analyze instances and simulations of what the other computer had just done.

Shadow mode still exists. That’s what the second computer is for. Not to double the processing power of what’s driving AP. It’s 2 identical computers. You don’t just combine processing power of the 2 into 1 process.
Please see the additional edits I made, especially the Tesla video showing that the initial plan for HW 3.0 was to have each computer running the same code and comparing their outputs.
 
  • Like
Reactions: FSDtester#1