Dirty Tesla released a 12.4.2 video:
Summary: He thinks it is much better.
Summary: He thinks it is much better.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
FSD v12.x (end to end AI)
- Thread starter Buckminster
- Start date
-
- Tags
- #fixphantombraking 12.4.x is coming autopilot chill is not chill elon is an ass elon rocks fix fsd extreme late deceleration fsd does not commit to go fsd does not maintain proper speed fsd doesn't read signs properly fsd hard and late braking fsd has decision wobble fsd jerky driving fsd late deceleration fsd sometimes may not detect train crossings fsd stops too early too quickly fsd stops too far behind line fsd stops when there is zero visibility fsd tailgating in city streets fsd very slow at stops fsd will never enable l3 with hw4
On the side bar, I guess these statistics means I say what I want to say without worrying if anyone likes it or not. I have less reactions than the number of comments I have made. Anyways, continue with the FSD discussions...
Daniel in SD
(supervised)
Elon literally said it would be a ChatGPT moment. ChatGPT4 came out March 2023.
gottagofast
2022 Model 3 Performance
Tesla keeps track of total interventions and safety disengagements for each new update compared to the last to determine whether to release it to more people, not 1 or 2 influencers saying if it’s ready or not
No new 2024.15.10 installs on TeslaFi today as of now and meanwhile the 2024.20.x branch is having a ball with so many incremental updates being pushed out in parallel. Hope we all get some love soon.
If there is no serious regressions reported by the OG testers and the influencers today then a new wave will begin at 5 PM PST. This is not an advice for stock trading.
FSDtester#1
Large Member
accurateI thought it was BruceOkay, tit's or tits?? Voice texting is not eazy!
I did tell you my first name is Forrest right?
FSDtester#1
Large Member
You're 10% accurate, If anyone thinks that Elon doesn't listen to his lapdog about how the software is working is kidding themselves...
Omar wants limited release so he can keep on posting and making youtube videos. @aronth5 nailed it yesterday.
Last edited:
FSDtester#1
Large Member
Oh and some "friends" of TMC are posting stuff on X that is posted here. Some of the comments are hysterical.
I checked on the people reposting it and they are all TSLAQ.... So just be advised, if you don't want anything you say here not ending up
on X, then don't post it. I could care less because TSLAQ are losing tons of $$ the last few weeks, and I just don't give a rip in general.
They are taking it completely out of context like we are panicking because the software is bad, when the point was we want it to go out to everyone as quickly as possible and don't want anyone slowing down the rollout. I will be responding to them shortly.
I checked on the people reposting it and they are all TSLAQ.... So just be advised, if you don't want anything you say here not ending up
on X, then don't post it. I could care less because TSLAQ are losing tons of $$ the last few weeks, and I just don't give a rip in general.
They are taking it completely out of context like we are panicking because the software is bad, when the point was we want it to go out to everyone as quickly as possible and don't want anyone slowing down the rollout. I will be responding to them shortly.
Last edited:
Tesla sends FSD to influencers early for testing, like 12.4.2 is now. They are provided with the Snapshot button (we used to have it until Tesla opened up FSD). The Snapshot Button sends data directly to Tesla.
In all likelihood Tesla takes data from the fleet (us) and anonymous/aggregates/combines into categories and then runs through algorithms looking for repeatable patters. They don't follow, know or care about any of our individual data. If so we would have the Snapshot Button.
Also suspect a large qunity of our data is mostly "junk' and dumped into a blackhole.
The only exceptions are likely high accelerometer disengagements. They probably warrant some investigating.
Last edited:
FSDtester#1
Large Member
Totally agree, I think they said they are now using 1 out of 10,000 clips to train the NN's.
At the same time, In my opinion, it doesn't help to have notable people bashing any new release.
Lets be real, how bad were some of the V10 and V11 releases. They tried to kill people if you weren't hypervigilant, but they released them anyway.
Good thing we have expert drivers like you to keep the program safe and progress moving forward!
FSDtester#1
Large Member
When is your next available time to check for update.... I bet you get it today by 8 pm.
If I win, you and Alan have to become friends IRL, If I lose you just have to become internet friends.
powertoold
Active Member
The ChatGPT moment refers to initial release of ChatGPT, which I believe was gpt3
sleepydoc
Well-Known Member
He's totally got a crush on you!I knew it, Elon wants to watch me drive via the cabin camera
FSDtester#1
Large Member
You are quite talented, Thank you!!
Do they also use simulations? It seems like there would not be many video clips of really dangerous stuff, like being in the wrong way lane, or about to drive into a brick wall. Simulations of the scenarios of actual real world crashes might let the AI see and learn how to behave in situations where very few real driver dare to tread.
Is it true that NN training has largely replaced de-bugging with training data curation? And that both curation is labor intensive, and NN training is massively computation intensive. Might this workflow make each trial iteration may take longer than just tweaking a few lines among 300,000 lines of C code?
Tesla says hardware is no longer a limiting factor for net training. At one point we were also told Tesla needed more data of optimal human driving. And then more recently Elon said the team is always happy to see intervention data from the field as they are so rare.
From the outside it's tough to know what the heck is going on other than 'who knew it would be this hard?' Answer - almost everyone in the industry!
I don't have knowledge in the area but it's been said current understanding, techniques, and tools aren't ready for net to net training. It was also a topic of discussion here when net to net was originally proposed.
AlanSubie4Life
Efficiency Obsessed Member
FSD was engaged before the right turn, while stationary. I requested the lane change from #3 to #2. Since I saw the truck coming out, I hoped maybe he would be able to cut back to #3 lane. You can see fairly clearly where I disengaged at ~9.3 seconds. Obviously I saw this problem immediately, but since traffic was clear behind me, I let it do its thing. I was not impressed. A little surprised there was no response, to be honest - I would usually expect a response to this. I guess trucks at various aspects remain an issue. Safety critical for sure.
Too bad the GoPro wasn’t running.
The wife would have been absolutely irate. This is of course immediately disqualifying. Can basically never happen for a minimum viable product.
Too bad the GoPro wasn’t running.
The wife would have been absolutely irate. This is of course immediately disqualifying. Can basically never happen for a minimum viable product.
Last edited:
mgs333
Member
That person seems like a rah-rah cheerleader, he/she seems to say every release is great. I distinctly remember him/her saying 12.4.1 was so amazing.
