Model 3 - FSD Supervised 12.4.3 - Test Loop around Santa Clarita - 2024.15.15 - 5 July 2024
Vložit
- čas přidán 4. 07. 2024
- Not the best drive unfortunately. Car would not commit with clear left turns a couple of times.
Left lane changes for an upcoming turn way too late several times.
Was in the correct lane, then left it and got back at the last possible moment sometimes putting it in a very bad spot when it didn’t need to. - Auta a dopravní prostředky
Point of FSD is to get u from point A to point B safely abiding road rules. I don't think one should expect it to have your own individual driving style..!
WOW. The evolution is amazing I can't wait to see where this tech will be in just 3-5 years
Awesome drive
Nah, this seems like perfect drive to me. Changing lanes too late, lets leave this debate for philosophers.
Nowhere NEAR a "perfect" drive.
Changing lanes late for FSD is one of my biggest annoyances. Why wait so long to get over!
There is nothing perfect about the buggy FSD software. Interventions every day, on any challenging drive. And simple drives too .
Yeah sure, multiple times the car just got stuck at intersections. Very "perfect". 🤣
@@DerekDavis213you don’t even have it on your pos lexus
Watching on my large screen TV. Almost like being there!
Thanks for your great videos.
lol
I appreciate the videos. You really should invest in an external clip-on mic. Even a cheap one would be much better than what you have now - especially with the AC blasting.
Keep up the good work.
It surely can drive. It needs a bit of confidence boost at the intersectons, and lane staging could be sooner. Speed and smoothness seem nice.
In big cities, FSD is nowhere even close to being trustworthy. Many examples right here on CZcams in LA, New York, Chicago, Toronto .
Can you turn the fan down a bit the next time? Thanks! :)
Wait what 50ºC?! You need that airco fan!
Not range anxiety but wrong lane anxiety 😅
I want FSD to work but after seeing this video with the last-second lane changes and hesitation at intersections, I cancelled my subscription. I've had v12.3.6 for awhile and it does many things well but has some of the same issues as what I'm seeing in this video (last-second lane changes, hesitation at intersections) plus some other issues (speed limits based on erroneous map data instead of posted speed limit signs, hard braking at intersections when the traffic signal changes to yellow), but I've decided I don't want to pay Tesla $99/month to test their FSD software. Tesla should pay me $99/month to test their software. When FSD can drive as confidently as I can while also driving safely, then I'll be the first to renew my subscription. Kudos to Tesla for all they've accomplished thus far with FSD. I wish them luck and will be following this channel and others looking for progress.
50°c? Are u in arizona or v3gas or something
Late deceleration forcing friction brakes still happening. The car needs to start regen earlier and use full regen much sooner if needed to slow down fast. 3:58, 5:13, 12:55
Yeah I see the 5x improvements on interventions per mile. Just in the wrong direction.
Not seeing the green dot.
Get your eyes checked.
Are you an employee? How’d you get 12.4.3?
OG tester
Chris also got it already :)
It's really starting to feel like neural nets as an autonomy solution are hitting a plateau, where emphasizing something in the training makes something else regress. The momentum seems lost compared to the first v12 wide release.
Maybe better hardware could help that, but it's hard not to lose confidence watching these test drives.
That's because the neural net is pretending to know how to drive, versus knowing how the roadways work and driving based on rules, like us humans do. Eventually they will need to inform the neural network with much more knowledge of the world, ballooning the compute required to drive as well as a competent human driver can.
Agreed. People that have been watching the evolution of FSD have *lost* *confidence* a long time ago.
@@vinny9152 Agreed. Not sure why they switched from all C++ code to all neural nets. Can rules be integrated into a neural net to get the best of both?
@@DerekDavis213Anybody with common sense can sense that. Except those cyber bulls. Herbert Ong from the CZcams channel “brighter with Herbert” is the worst. FSD 12.4x is definitely not the 5-10x improvement that Elon claimed it would be.
@@user-py7or7zi5k The problem with code is that it takes time to load and run - leading to slower decision making from FSD. Will not surprise me if Tesla resort to using LiDAR or ultrasonics. At the moment all that Tesla seems to be doing is throwing billions of dollars at compute and AI with minimal or no progress in the driving experience. What a waste of resources!
At least, show your software version man. You could have 12.4.2 and no one can tell. 😂
And seems like what is in the description is from Highway experience????? Then thats v11 stack
Your keyboard seems to be broken.
So. Many. Regressions. So disappointing Tesla!
But Elmo says 'a game changer' and 'best ever' and 'better than a human driver'. what a joke
I think this vision only training will never work. It is like teaching driving to a monkey.
That's the whole point! Get MUCH better compared to a monkey. So we can watch it drive us. Buy now and lets celebrate in 2030 mate. First beer is from me. The rest is yours because i'm sure you can pay from your profit!
It’s already working. Every self-driving software out there is a work in progress including Waymo which hit a cyclist recently. Look at the data buddy and try not to be so naive when posting comments just to say you’re posting.
human vision doesn't work, roughly 120 people per day die in car accidents just in the USA
@@rctezluh42069 Agreed. The design of FSD is fundamentally wrong, and will never reach Full Autonomy. Different hardware, and a different software engineering team are needed.
@@rctezluh42069 120 people die per day? There's 340 million people in America. 120 is a very small number.