Knightshade
Well-Known Member
Remind me how all the paid Tesla drivers training on that turn worked out for V11?
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Elon: FSD Beta tweets
- Thread starter Terminator857
- Start date
AlanSubie4Life
Efficiency Obsessed Member
It worked out super well. For me.
Daniel in SD
(supervised)
They got pretty close to 90% success!
I still want to know why this turn is impossible to simulate.
At one point we were told the NNs knew when training data was simulated and therefore didn't improve and more recently they say simulations are very good. Maybe it's still a bit of both or the team isn't able to create good simulations?
Six months ago the narrative was we didn't need to train the end to end NN, it just knew what to do. Now they're back to overfitting that which doesn't work well.
FSD-estivus for the rest of us. Hilarious how they characterized Omar's yapper!
Knightshade
Well-Known Member
Daniel in SD
(supervised)
But it’s great for a driver assist. It makes the turn so much easier and more relaxing.
AlanSubie4Life
Efficiency Obsessed Member
Not one single time have I wanted FSD to take my unprotected left. It is definitely way more stressful than just doing it myself, even when traffic is light.But it’s great for a driver assist. It makes the turn so much easier and more relaxing.
I don’t want to be the fool who got t-boned and died while letting FSD try to take a trivial turn. High stakes.
I will say it is way way better at taking the turn than it used to be. Probably 30% success rate, largely dictated by traffic volume.
Daniel in SD
(supervised)
You are a well known enemy of The Mission.
Knightshade
Well-Known Member
But you've assured us many times it's secretly an L5 system, so it continues to be a really awful one
With respect, I don't think this is the correct summary. What we were told in the original Elon/Ashok video, and AFAIK is still true,
Was that the key to everything is lots of training.
They didn't say the NN didn't need to be trained and just knew what to do. They said they didn't explicitly program it to recognize and label objects like stop signs, traffic lights, pedestrians etc. Nor did they program it how to drive - all that ability comes out of the extensive E2E training.
In the video, when it did make a mistake, they immediately said that it would be solved by more training of similar scenarios. More of the same. So talking about needing more training today is not a case of "back to overfitting"; it's perfectly consistent with the original explanation of v12.
I should add that I haven't seen Elon deny that overfitting can occur. He said that the first selectively released version needed more training outside of California and more training in bad weather. That's essentially an admission that it was overfit for California.
They managed to find one without the giant penis that's normally on the homunculus models lol.FSD-estivus for the rest of us. Hilarious how they characterized Omar's yapper!
View attachment 1026087
That being said, I wonder what it would be in this case? Elon?
diplomat33
Average guy who loves autonomous vehicles
Good thing Elon has never overhyped anything before.
DanCar
Active Member
News flash : v12 garbage, v13 the new hotness.
Just wait till WholeMars reviews it. It'll be a real
fest and a competition to see who can heap the most superlatives on V13 should be.
WholeMars will say v12.2.1 is a piece of *sugar*. I hope he is right thoughJust wait till WholeMars reviews it. It'll be a real
EVNow
Well-Known Member
Yes - I'd put V11 in that ULT as better than average humans.
More importantly that made FSD actually be able to handle simpler ULTs that it could barely ever do earlier. Now I've > 90% success rate on all my (admittedly simpler) ULTs.
Similar threads
