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

FSD Beta 10.69

This site may earn commission on affiliate links.
Earlier I saw a conversation about FSD in rain. I got to try it (10.69.1.1) in some decent rain earlier because I had FSD active and then it started to rain. It was a rural-suburban Midwest two lane highway going about 60 with low-medium traffic and it did better than expected, especially on a decent curve. It was raining hard enough to where it wouldn’t let me reengage after disengaging (I don’t remember why I disengaged).

Not sure if this means much because from what I’ve seen FSD can be nonsensical with weather intensity required to block or allow FSD.

My 2¢ tho, I was pleasantly surprised by it’s performance in the rain
 
No new, hopefully better working beta, love for us Tampa Bay folks?
I love this build. Put it on at the course and tested yesterday. Drove home on the side roads all the way to the gate from the other side of town and the best way I could describe it is, I felt like a passenger in the left seat. Not one intervention. It even did the two roundabouts I go through like a champ.

I was pleased.
 
I didn’t say anything about them not having software revenue! Their software revenue is likely pretty significant!

Anyway not sure the specific quote you are referring to. I guess I’d be a bit surprised if it suggests FSD (the City Streets portion) is important in the foreseeable future. There is not any evidence that it is, and there is not a competitive market for any such feature at the moment.
Short term traders don’t care much about about FSD other than its headlines and their reactionary effects on other short term traders.

Long term investors see FSD the way they saw the first iPhone (which had no apps, barely any functionality, and was an expensive toy for tech-nerds).

The driving part of FSD sits on a MOUNTAIN of generalized AI infrastructure that is completely portable to other uses.

Tesla has mentioned this several times.

Killer advantages that have come out of the R&D that people buying FSD has paid for (all of which apply to generalized AI applications outside of a vehicle):

- NN architecture that can generate live 3D vector-map of the world (including printed detail) at decent framerate on a device low enough powered to walk around.

- The Data Cycle infrastructure that feeds new data to and trains these new NNs, which in itself in in its earliest incarnation is allowing insanely fast iteration on the NN system above. (This is the easiest one to sleep on, but the speed in which they replaced the object detection nets with the Occupancy net, AND the improvement in ability for this 1.0 version of the Occupancy net is incredible. And due to the Data Engine they’ve set up.)

- The worlds-first production NNPU, the Dojo v1.0 chip, which is already made for massive tiling and scaling (NN processing as a service will be a huge revenue generator, like Amazon’s various compute-as-service).

- A stable of most of the world’s newest and most of the best AI and NN engineers (and NNPU chip designers!). Remember that AI Day 1 was a hiring event. And their applications jumped insanely high. Tesla has a mega-size and mega-funded AI project and people wanting to work on AI are going there. This is the real reason Google and Apple dominated the past 20y. The engineers that were there and were paid to build cool stuff.



And this is just the stuff we know about today! AI Day 2.0 is going to be so revealing.

Data will be this century’s oil.

AI will be this century’s microchip. It will change every industry top to bottom.

Tesla may be overvalued as a car company, but anybody thinking of them as a car company is going to get burned again and again.

Tesla is a tech company. They will be a Tech giant in a decade.

Tesla is Apple in 2006.
 
Short term traders don’t care much about about FSD other than its headlines and their reactionary effects on other short term traders.

Long term investors see FSD the way they saw the first iPhone (which had no apps, barely any functionality, and was an expensive toy for tech-nerds).

The driving part of FSD sits on a MOUNTAIN of generalized AI infrastructure that is completely portable to other uses.

Tesla has mentioned this several times.

Killer advantages that have come out of the R&D that people buying FSD has paid for (all of which apply to generalized AI applications outside of a vehicle):

- NN architecture that can generate live 3D vector-map of the world (including printed detail) at decent framerate on a device low enough powered to walk around.

- The Data Cycle infrastructure that feeds new data to and trains these new NNs, which in itself in in its earliest incarnation is allowing insanely fast iteration on the NN system above. (This is the easiest one to sleep on, but the speed in which they replaced the object detection nets with the Occupancy net, AND the improvement in ability for this 1.0 version of the Occupancy net is incredible. And due to the Data Engine they’ve set up.)

- The worlds-first production NNPU, the Dojo v1.0 chip, which is already made for massive tiling and scaling (NN processing as a service will be a huge revenue generator, like Amazon’s various compute-as-service).

- A stable of most of the world’s newest and most of the best AI and NN engineers (and NNPU chip designers!). Remember that AI Day 1 was a hiring event. And their applications jumped insanely high. Tesla has a mega-size and mega-funded AI project and people wanting to work on AI are going there. This is the real reason Google and Apple dominated the past 20y. The engineers that were there and were paid to build cool stuff.



And this is just the stuff we know about today! AI Day 2.0 is going to be so revealing.

Data will be this century’s oil.

AI will be this century’s microchip. It will change every industry top to bottom.

Tesla may be overvalued as a car company, but anybody thinking of them as a car company is going to get burned again and again.

Tesla is a tech company. They will be a Tech giant in a decade.

Tesla is Apple in 2006.
Yes this is why I said regulation does not matter. It does not even matter if all L3+ driving is banned completely. (Or even some L2+!)

But anyway Tesla is not the only one working on this stuff; we’ll see if they beat people working in specific areas of expertise at their own game!

There are certainly tons of applications, many of which we can’t even think of at the moment.

(And, for the same reasons, notably, it does not matter if FSD does not work (as I said)! What if it is the most challenging application?)
 
  • Like
Reactions: FSDtester#1
Short term traders don’t care much about about FSD other than its headlines and their reactionary effects on other short term traders.

Long term investors see FSD the way they saw the first iPhone (which had no apps, barely any functionality, and was an expensive toy for tech-nerds).

The driving part of FSD sits on a MOUNTAIN of generalized AI infrastructure that is completely portable to other uses.

Tesla has mentioned this several times.

Killer advantages that have come out of the R&D that people buying FSD has paid for (all of which apply to generalized AI applications outside of a vehicle):

- NN architecture that can generate live 3D vector-map of the world (including printed detail) at decent framerate on a device low enough powered to walk around.

- The Data Cycle infrastructure that feeds new data to and trains these new NNs, which in itself in in its earliest incarnation is allowing insanely fast iteration on the NN system above. (This is the easiest one to sleep on, but the speed in which they replaced the object detection nets with the Occupancy net, AND the improvement in ability for this 1.0 version of the Occupancy net is incredible. And due to the Data Engine they’ve set up.)

- The worlds-first production NNPU, the Dojo v1.0 chip, which is already made for massive tiling and scaling (NN processing as a service will be a huge revenue generator, like Amazon’s various compute-as-service).

- A stable of most of the world’s newest and most of the best AI and NN engineers (and NNPU chip designers!). Remember that AI Day 1 was a hiring event. And their applications jumped insanely high. Tesla has a mega-size and mega-funded AI project and people wanting to work on AI are going there. This is the real reason Google and Apple dominated the past 20y. The engineers that were there and were paid to build cool stuff.



And this is just the stuff we know about today! AI Day 2.0 is going to be so revealing.

Data will be this century’s oil.

AI will be this century’s microchip. It will change every industry top to bottom.

Tesla may be overvalued as a car company, but anybody thinking of them as a car company is going to get burned again and again.

Tesla is a tech company. They will be a Tech giant in a decade.

Tesla is Apple in 2006.

Wow. All this hard work and they can't get autowipers out of beta.
 
- The worlds-first production NNPU, the Dojo v1.0 chip, which is already made for massive tiling and scaling (NN processing as a service will be a huge revenue generator, like Amazon’s various compute-as-service).
Stop spreading Elon lies and go out read a book and actually educate yourself. You have clearly been spending too much time in that bubble.
 
Last edited:
The point of his custom UI is to make it visible when he intervenes. Showing what FSD is doing would defeat the whole purpose.
We’re talking about the Tesla UI. I am sure there is a rationale for Tesla too (perhaps similar).

One I can think of is when you press the brake you disengage. If some silly person is looking at the screen during an emergency brake they might determine (if it showed manual braking) that the car was doing the braking…and relax brake pressure. This design choice reduces mode confusion, perhaps, at the margins. When you start braking, the displayed braking drops to zero.

Should reduce chance of UI confusion where the user thinks FSD is braking (when it is actually the driver).
 
Last edited:
the speed in which they replaced the object detection nets with the Occupancy net, AND the improvement in ability for this 1.0 version of the Occupancy net is incredible
Do you think the replacement started with Karpathy's leave? Although I have my doubts on how much can actually be replaced and potentially instead a merging down the line.

It's been interesting seeing what the occupancy network visualizes with 10.69 as it seems to ignore trees and light posts but sometimes renders fences that aren't even directly next to the road.

Today it correctly visualized concrete barriers for construction although previous versions correctly rendered them as road edges and pathed appropriately. As for actually new driving behavior due to occupancy network, last night it gave extra space for a tipped over shopping cart that was in a bike lane.
 
Do you think the replacement started with Karpathy's leave? Although I have my doubts on how much can actually be replaced and potentially instead a merging down the line.

No, I think Tesla has been working on general object detection for a while now. I rewatched AI Day and Ashok's recent talk again yesterday, and the NN stack with the occupancy network and what Andrej described during AI Day are exactly the same (raw input -> Regnets -> BiFPN). While this doesn't mean that Andrej helped with the occupancy network, I think the occupancy network's foundations were conceived for FSD V11, but it took them a lot longer to get it all polished enough for deployment. Elon has teased V11 release since July 2021.

Ashok (and the FSD beta release notes) mentioned that the occupancy network was previously based on single frames, and the big advancement with 10.69 was that the network was now consuming video. Elon has also been saying that Tesla was working on making their NNs do 4D video since mid-2021. I assume this was also related to the occupancy network.
 
  • Like
Reactions: GranularHail
What's left for Tesla to do:
Something that I haven't seen anyone else mention is how long it takes to slow to the posted speed after a speed limit change. I often drive on 2-lane highways that pass through small towns. The speed limit can change from 55 to 45 to 35 and even 25 in short order. FSD beta (10.12.2 still for me) takes a very long time to slow down. Sometimes I can be through the town completely before it reaches the lower speed limit. That is a very good way to get a ticket!

As a result, I typically have to disengage every time I enter one of these little towns. Once I slow the car down, then I can reengage.
 
Something that I haven't seen anyone else mention is how long it takes to slow to the posted speed after a speed limit change. I often drive on 2-lane highways that pass through small towns. The speed limit can change from 55 to 45 to 35 and even 25 in short order. FSD beta (10.12.2 still for me) takes a very long time to slow down. Sometimes I can be through the town completely before it reaches the lower speed limit. That is a very good way to get a ticket!

As a result, I typically have to disengage every time I enter one of these little towns. Once I slow the car down, then I can reengage.

Yes, I find that the delayed slowing actually mimics most human drivers, but yes, it's a ticket risk.
 
  • Informative
Reactions: pilotSteve