Cheburashka
Active Member
I still don't get how Tesla even knows that the car got into an accident or do they only count airbag deployments as an accident?
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
OFFICIAL BUTTON WATCH
- Thread starter Drrobmiller
- Start date
Daniel in SD
(supervised)
\\
I think that's airbags or seat belt pretensioners.
"we count all crashes in which the crash alert indicated an airbag or other active restraint deployed. In practice, this correlates to nearly any crash at about 12 mph (20 kph) or above, depending on the crash forces generated."
I think that's airbags or seat belt pretensioners.
diplomat33
Average guy who loves autonomous vehicles
Surprising that they're almost ready to release v10 - isn't that the rewrite that moves NoA to the same stack as City Streets Beta? Of course I'm taking the 4 weeks for the button with a giant grain of salt.....
EDIT: Actually something doesn't add up here. A couple of days ago he said he was driving 9.3. Now he's saying there's no point release and they have the v10 rewrite ready to go out in 10 days? Why wouldn't he be driving v10 if it was that close.
rxlawdude
Active Member
AlanSubie4Life
Efficiency Obsessed Member
Right on time - that poll a while back I guessed September or so. Look at Elon, coming through. Four weeks best guess, right?
Back in March it was ~10 days IIRC? So how does this all scale? 10 days scales to 5 months so four weeks goes to about 1 year.
Back in March it was ~10 days IIRC? So how does this all scale? 10 days scales to 5 months so four weeks goes to about 1 year.
Shocking.
Last edited:
Elon is driving builds that haven't been released to beta testers yet. If v10 is the next build to be released in just 10 days (as he claims), I would think he'd be on it.
EDIT: Wow, two sarcastic responses in a row, from @AlanSubie4Life and @rxlawdude. Thanks for the productive conversation guys!
AlanSubie4Life
Efficiency Obsessed Member
I'm just getting a little tired of the silly timelines. Just don't announce when it's coming, and release the button when it's ready. Why waste people's time and energy?
Lexdysic
Member
I read somewhere that what is v9.3 will become v10.
I don't know- something this time tells me we might actually see it. No, not in 4 weeks, but maybe early October. Elon's gotta sense the level of hostility at this point and he knows he can't BS much more without some repercussions. Or, maybe they are genuinely ready to widen the testing. Obviously I could be wrong.
Here's a stupid question- has Tesla ever released something even quicker than Elon promised?
Matters which version of the "promise" you use. Some people will say Elon delivered on time when he promises 2 weeks, 4 weeks, next month, June for sure, no point in releasing, 4 weeks... And then he delivers 4 weeks from the last tweet which was 11 months after the first tweet.
diplomat33
Average guy who loves autonomous vehicles
I am surprised too that V10 seems to be so close. It seems that maybe Elon is in a hurry to get to the single stack that does both city and highway so that he can declare "feature complete". The idea is to get the single stack so that all the pieces are in place and then address edge cases as they come. It would also make sense to give the public the single stack since then they can get the most feedback on the "feature complete" version.
powertoold
Active Member
Elon's timelines make no sense, but they really don't matter at this point, as fsd beta has no competitors, and no other company is remotely close to Tesla.
My best guess about beta 10 is that it's going to incorporate the spacial RNN that Karpathy mentioned during AI Day. This will essentially enable fsd beta to generate an HD map of the road geometry and features as it goes. As you'dve guessed, this is going to be a quantum step change in probably two weeks.
My best guess about beta 10 is that it's going to incorporate the spacial RNN that Karpathy mentioned during AI Day. This will essentially enable fsd beta to generate an HD map of the road geometry and features as it goes. As you'dve guessed, this is going to be a quantum step change in probably two weeks.
Gee.. that sounds really familiar…. Like no point in releasing the button familiar.
powertoold
Active Member
Hopefully beta 10 will take it to Chandler-level performance+, and everyone else will be doing way'mo crying. Beta 10 must be the turning point to generalized fsd.
diplomat33
Average guy who loves autonomous vehicles
So if V10 achieves "chandler-level performance+", can we expect Tesla to remove the driver like Waymo did?
Also, Waymo has now deployed their 5th Gen FSD to the public in San Francisco that is way better than "chandler performance+".
Last edited:
linux-works
Active Member
its all r/d at this point. fancy way to say 'trial and error'. they try things, some work, some dont, they go back and iterate some more.
but the crutch is this insistence on 'data'. they are betting the farm entirely, 100% on it and I think that's risky as hell, given that its NOT proven that a data-driven approach (NN) is sufficient for this task.
if that were known and its just a matter of implementing it better and better, fine. but we dont even know if this data-centric approach is going to be maintainable and operational with the amount of cpu needed to chug it along.
brute force (which is all this is, lets be honest) only gets you so far.
I wish them luck. but I dont think its gonna work when you put all your emph on 'training NN'. totally unproven that it will get you 99.999% there, safely.
just remember, NN is just fancy statistics and probability. there is still ZERO comprehension going on. that really should worry a lot of you who blindly trust this 'tech'.
but the crutch is this insistence on 'data'. they are betting the farm entirely, 100% on it and I think that's risky as hell, given that its NOT proven that a data-driven approach (NN) is sufficient for this task.
if that were known and its just a matter of implementing it better and better, fine. but we dont even know if this data-centric approach is going to be maintainable and operational with the amount of cpu needed to chug it along.
brute force (which is all this is, lets be honest) only gets you so far.
I wish them luck. but I dont think its gonna work when you put all your emph on 'training NN'. totally unproven that it will get you 99.999% there, safely.
just remember, NN is just fancy statistics and probability. there is still ZERO comprehension going on. that really should worry a lot of you who blindly trust this 'tech'.
Lexdysic
Member
All true, but nobody is touting this as L5. Only the foolish types will “blindly trust” this supposed imminent release. (And we’ll likely see the results of their idiocy on you tube) Driver will still be required to maintain control/awareness at all times. I’d guess there will be no changes with the steering wheel nag.
But I‘ll welcome the improvements (if they indeed come), while staying ready to quickly take over if needed.
powertoold
Active Member
They don’t need to. They‘re not at the robotaxi stage yet.
Fsd buyers will be plenty satisfied with a high-performing fsd level 2 feature. There’s no rush for robotaxis until 10x+ human safety.
Similar threads
