FSD v12.x (end to end AI)

[FSDtester#1]

I don't mind honest arguments. But if someone wants to take issue with something I said, quote it instead of tagging me and then making something up. I think he's upset about an argument we had 2 years ago (during FSD 10.69) that I just had to look up, in which I said:



I've never said Omar doesn't hide things. I said he makes uncut videos in the same vein as the promotional videos made by autonomous driving firms. He's clearly a biased source and only shows the positives. But likewise we only have positive videos from companies that have autonomous driving programs without customers.

Agree with you ... I and a handful of others say dumb things all the time, and have 1 serious post out of 10. We deserve getting called out. You have been a straight shooter and excellent poster since I joined. Keep it up, your content is always spot on, both here and in the investors forum.

 

[JHCCAZ]

Unfortunately the videos so far are lacking. I know I'm in the minority but nags don't bother me. Way down on my list of priorities probably because I rarely get them. Would much rather see if there are actual improvements in v12.4.1 in areas that cause disengagements. Omar's videos are useless in that regard, Dirty Tesla focused almost entirely on nags which I did find interesting but didn't shed any light on real improvements and Rocco used FSD with a trailer so that was a fail. Will just have to wait on everyone who posts in this thread for some good feedback. And hopefully I'll get the update in the next week too. I wonder though if wide release will end up being v12.4.2?

I agree that the much anticipated influencer videos are a bit of a letdown this time.

I think AI Drivr is giving us the most good material, including the manic lane-change example.

Dirty Tesla is usually one of my favorites and he made a credible video, but he's been focusing on Cybertruck (without FSD of course) and I think he's already become a little rusty about the 12.3.6 comparison because he hasn't used it much. And I agree that the camera monitoring is obviously a key new feature to understand- but I'm not driving around Tucson without sunglasses during the day, so I don't know how much I'll be able to use it.

Rocco's test with a trailer just can't be evaluated for normal driving. Actually I'm surprised and somewhat concerned that Tesla even allows this with FSD. It clearly messes up the view to the rear and totally distorts the control dynamics. Towing a trailer or having a big kayak hanging over the windshield, or anything else like that, should probably be trained just enough to recognize it and disable full auto FSD.

I usually don't spend too much time with Omar's videos, not because I really dislike him (I don't), but because I just don't like the background music or whatever other useless audio, and also it's been clear that he was using some kind of hardware to game the nag. I tried to watch a video today where he had someone else on the phone talking about FSD while he (Omar) was getting in and out of sticky driving situations, as well is having someone in the car he was conversing with simultaneously. It was just a mess. I was okay with the videos he did with his father a few months ago.

Not having Chuck's relatively objective and consistent measure of new-release performance (because his pilot work schedule is keeping him away from home this week) has been another significant detraction from the usual excitement of FSD release videos.

The upshot is that I really don't know what to expect this time. I think general improvement but not flawless Robotaxi.

 

[sleepydoc]

https://twitter.com/x/status/1799571673181769868

watching the video, the car starts to move before the flagger even changes his sign or makes any other indication; I wonder if the car was simply stopped because it saw cars then started to proceed when the coast was clear rather than based on a cue from the flagger directing traffic

 

[kabin]

https://twitter.com/x/status/1799571673181769868

Slowing the video down shows FSD lights up the detour path and starts moving before the flagman flips the stop sign.

 

[kabin]

watching the video, the car starts to move before the flagger even changes his sign or makes any other indication; I wonder if the car was simply stopped because it saw cars then started to proceed when the coast was clear rather than based on a cue from the flagger directing traffic

I just posted the same thing.

It's sad how the piped pipers project imaginary success.

 

[enemji]

Slowing the video down shows FSD lights up the detour path and starts moving before the flagman flips the stop sign.

It is a predictive engine, and it predicted the flaggers action

 

[FSDtester#1]

 

[Matias]

He seems to have a very strict criteria for success.

 

[JulienW]

He seems to have a very strict criteria for success.

His drives are sooooooo boring and just low traffic easy suburbia. He did have ONE exciting drive back around 10.5 or so where he curbed a wheel. There used to be an influencer named Frenchie (I think) that lived in downtown Chicago and he had some good urban driving routes with REAL challenges. Chris goes into Ann Arbor some and that is a little urban with lots of VRUs.

 

[JulienW]

Also the Lunashi person must not be a fired Tesla employee since they got 12.4.1. I bet a they have a significant other that is an employee.

 

[JB47394]

I agree that the much anticipated influencer videos are a bit of a letdown this time.

Well here's a good one for you. It shows some good and some bad, but be sure to mute it. At one point, the car runs up to a construction worker with a stop sign, and it seems to want to go around until the worker steps out in front of the car, gesturing for Omar to move back into the correct lane (to allow oncoming vehicles to pass). Omar does nothing until an oncoming car appears, then he disengages.

From what I've seen in the videos so far, there has been some helpful tuning in 12.4, but no major improvements, and possibly some regressions.

 

[Todd Burch]

Well here's a good one for you. It shows some good and some bad, but be sure to mute it. At one point, the car runs up to a construction worker with a stop sign, and it seems to want to go around until the worker steps out in front of the car, gesturing for Omar to move back into the correct lane (to allow oncoming vehicles to pass). Omar does nothing until an oncoming car appears, then he disengages.

From what I've seen in the videos so far, there has been some helpful tuning in 12.4, but no major improvements, and possibly some regressions.

One thing that seems to be considerably improved is stop signs—less delay, faster creeping. That’s a big win in my book but we will need to verify ourselves!

 

[gottagofast]

I’m surprised the speed limit slowing down wasn’t fixed in this

 

[dhanson865]

He seems to have a very strict criteria for success.

I agree with using the strict criteria for the most part. The average joe isn't going to put up with almost good enough.

Put another way I don't think 12.3.x was worth paying for. I need to see dramatic improvement in 12.4.x and possibly 12.5.x before I'll pay to keep it.

 

[FSDtester#1]

His drives are sooooooo boring and just low traffic easy suburbia. He did have ONE exciting drive back around 10.5 or so where he curbed a wheel. There used to be an influencer named Frenchie (I think) that lived in downtown Chicago and he had some good urban driving routes with REAL challenges. Chris goes into Ann Arbor some and that is a little urban with lots of VRUs.

And he routinely disengages to make a report and hits the go pedal too much

 

[fasteddie7]

His drives are sooooooo boring and just low traffic easy suburbia. He did have ONE exciting drive back around 10.5 or so where he curbed a wheel. There used to be an influencer named Frenchie (I think) that lived in downtown Chicago and he had some good urban driving routes with REAL challenges. Chris goes into Ann Arbor some and that is a little urban with lots of VRUs.

I think a lot of folks look at my test route the same way, but for me it’s that FSD has trouble completing it, so I test it over and over. He definitely doesn’t have the issues I do, as there’s a lot of blind turns and to this day fsd can’t handle the forks in the road on my route.

Although based on last nights tweet, 12.4.1 isn’t the one we’re getting. And a new version will be the wide release one.

 

[SSDTester#1]

Well here's a good one for you. It shows some good and some bad, but be sure to mute it. At one point, the car runs up to a construction worker with a stop sign, and it seems to want to go around until the worker steps out in front of the car, gesturing for Omar to move back into the correct lane (to allow oncoming vehicles to pass). Omar does nothing until an oncoming car appears, then he disengages.

From what I've seen in the videos so far, there has been some helpful tuning in 12.4, but no major improvements, and possibly some regressions.

Why do you still watch his videos?

 

[TK211X]

Where the hell is 12.4.1

I’ve been peaking occasionally while busy and wow is Elon over promising. It’d be like some promise and then one week later I check back and nothing.

 

[kabin]

Agreed with his routes being boring. Sounds like he prefers 12.3.6 aside from 12.4.1 being hands free. Some issues with unprotected lefts and late lane selections.

 

[SSDTester#1]

Where the hell is 12.4.1

I’ve been peaking occasionally while busy and wow is Elon over promising. It’d be like some promise and then one week later I check back and nothing.

Two weeks is the key, you’re doing it wrong