-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
FSD Beta 10.69
- Thread starter Buckminster
- Start date
Good to see the continued improvement of FSD at least according to this source.
With 10.69.3 due out soon it will be interesting to see how the dashboard reflects the improvements since .3 is supposed to include many changes.
FSD Dashboard
With 10.69.3 due out soon it will be interesting to see how the dashboard reflects the improvements since .3 is supposed to include many changes.
FSD Dashboard
AlanSubie4Life
Efficiency Obsessed Member
With this doubling* every release, and it only needing to be 10k times better (probably needs to be more, but for the sake of argument, and 100k times better makes little difference in this argument), we only have 13 more releases to go! At current release rate, that's about 2-3 years. Exponential improvement!
* Note small sample size for 10.69.2.3, potentially with @EVNow data skewing ~10% of it (I think he contributes or used to?), since he says he's changed his driving style, and maybe other drivers have taken the "let it ride" plunge as well.
mtndrew1
Active Member
I did my test route on 10.69.2.3 last Friday and it was really, really bad. I thought the test might be a fluke but my commute this afternoon was rerouted through urban streets due to freeway traffic and FSDb remained awful.
10.69.2.3 is one of the worst performers ever and has the most forced disengagements of any version I’ve tried. Just terrible. See the link below for the gory details.
FSD Beta Test Route
10.69.2.3 is one of the worst performers ever and has the most forced disengagements of any version I’ve tried. Just terrible. See the link below for the gory details.
FSD Beta Test Route
Um. Once upon a time, in the deeps of time, a coworker and I were writing diagnostics for a fairly massive piece of electronics gear. The gear had, compared to any piece of modern equipment one might care to name, dead slow read/write access times. The main CPU was speedy for its day, though. I had written a large C structure that overlaid all the registers that mattered, and we had written repetitive loops that cycled through everything with what was supposed to be on-line diagnostics. It took 10 hours to complete.With this doubling* every release, and it only needing to be 10k times better (probably needs to be more, but for the sake of argument, and 100k times better makes little difference in this argument), we only have 13 more releases to go! At current release rate, that's about 2-3 years. Exponential improvement!
* Note small sample size for 10.69.2.3, potentially with @EVNow data skewing ~10% of it (I think he contributes or used to?), since he says he's changed his driving style, and maybe other drivers have taken the "let it ride" plunge as well.
Turns out that generating actual addresses from a structure takes a fair number of assembly cycles. So, switched to cast-a-pointer once, then do pointer math. 30 minutes.
Then my coworker realized (well, it was his hardware) that there were two bits per chip, 16 bits total, so we could run the diagnostics on each chip in parallel. That, and a few other improvements, and we were down to 5 seconds. From 10 hours.
Stuff like that can happen in software. And that’s not the only example that comes to mind in my career. In general, better algorithms can make a huge difference.
Now, yeah, the Tesla SW people may have reached a major dead end. Or they might be going up the slope of diminishing returns. Or (and this is my point) a number of things may be coming with an order of magnitude step improvement in FSD-b’s capabilities. Now or later.
No question, FSD-b resembles a research project more than anything else. Tesla is pushing the state of the art. But Tesla talks a good line and they sure don’t seem like they’ve given up. So, I’m optimistic. Maybe a year, tops?
With this doubling* every release, and it only needing to be 10k times better (probably needs to be more, but for the sake of argument, and 100k times better makes little difference in this argument), we only have 13 more releases to go! At current release rate, that's about 2-3 years. Exponential improvement!
* Note small sample size for 10.69.2.3, potentially with @EVNow data skewing ~10% of it (I think he contributes or used to?), since he says he's changed his driving style, and maybe other drivers have taken the "let it ride" plunge as well.
Where is the data being pulled from?? That data doesn't match mine at all. 69.2.1, 69.2.2, 69.2.3 have almost no discernable difference to me, and all have more disengagements than 10.12.2 and 10.10.2 (almost entirely lane selection issues)
I'd like to put my own test results there!
AlanSubie4Life
Efficiency Obsessed Member
@EVNow and other converts, I think.
You can put your own test results there, I think you just have to sign up though have not looked into it.
AlanSubie4Life
Efficiency Obsessed Member
This term is a little harsh. Shorthand for what I meant which was: this is completely unscientific and there is likely strong self-selection bias. There’s nothing wrong with “converting” and modifying how you drive as long as you are careful about when to allow it. But it will skew results.
And people should expect their personal experience to not match especially at these very high intervention rates.
Sure. Above I extrapolated from these extremely noisy data with extremely limited miles driven and other major issues, that we might see 4-5 orders of magnitude improvement over 3 years, at the rate of improvement measured with this last release.
So one order of magnitude could happen in a bit over a half a year!
Last edited:
willow_hiller
Well-Known Member
Maybe a new version of Smart Summon in 10.69.3?
I say not likely. It would probably need to be done to be included in 69.3. Also in Elon talk Almost Done is a lot like Two Weeks or By the End of This Year.
willow_hiller
Well-Known Member
I say not likely. It would probably need to be done to be included in 69.3. Also in Elon talk Almost Done is a lot like Two Weeks or By the End of This Year.
This is at least the second, if not third time he's said it's coming soon. And given how often the self-imposed deadlines for 10.69.3 and these new features get missed, it seems likely to me that they'll be released together.
If you recall, in September he said they had end-of-month deadlines, and that 10.69.3 was meant to be shortly after AI Day (end of Sept as well).
EVNow
Well-Known Member
LOL. I’ve signed up but never submitted data (somehow I never got a reply or lost the mail … reminds me I should follow up).
Agree, besides I suspect Elon is referring just to Development and then internal employee testing is needed. I assume testing of smart summon would take a lot longer than typical employee testing of a FSD build.I say not likely. It would probably need to be done to be included in 69.3. Also in Elon talk Almost Done is a lot like Two Weeks or By the End of This Year.
AlanSubie4Life
Efficiency Obsessed Member
Most definitely not. They’ll save it for single stack which can’t be called that if Smart Summon does not use all of FSD’s bells and whistles.
What is going to be interesting is to see what HW2.5 people get. They (if they have EAP) were promised Smart Summon too! Are they going to get (Actually Not) Smart Summon? Or will they run the FSD stack (for parking lots) on HW2.5???
Maybe they’ll get (Very Slow) Actual Smart Summon?
Seems like it will be easier for Tesla to just upgrade those people to HW3.
Last edited:
funny enough theres a seperate thread about people getting hw3 free upgrade at their SC. I myself got lucky but it wasn't easy as i believe the trick is whoever is making the service ticket isn't paying attention and typing in the wrong code (prepaid upgrade) oppose to paying the 1k. I for one wasn't gonna stay behind with hw2.5
AlanSubie4Life
Efficiency Obsessed Member
Yeah I have seen that. Not sure what to think of that cluster.
EVNow
Well-Known Member
I've been thinking about this ...
How does FSD improve disengagements ? Obviously by making fewer mistakes .... but how does the driver know whether a mistake is going to happen or not ? What we do is - guess. The guess is based on several factors - some can be easily characterized, like we think given the speeds and trajectories of all actors on the road, an accident might happen. Or we see that FSD is very indecisive ... jittery. But it all remains .... a guesswork. May be my guess has been faulty i.e. I've been disengaging when there wasn't a need. Afterall Omar disengages quite a bit less (and AFAIK) hasn't had any accidents. So, his guess is actually better than mine ?
In other words, one way disengagement rate goes down is when the testers feel less threatened. FSD can improve disengagement rate by gaining more trust.
Looks to me that is exactly what has happened. FSDb has gained a bit more trust from me. So, though it feels like I "lowered" my standard, in reality disengagement rate went down because it gained more trust.
Now, did FSDb gain trust by improving in some ways ? I do think so ... but I don't expect my trust level to be linearly correlated to actual FSDb ability. Its probably a step function ...
Last edited:
EVNow
Well-Known Member
Here is the latest "robotaxi" report. The driver seems to be professional (atleast part time) ride-sharing driver. He has done some 16k kms (or about 10k miles) of city driving using FSD. He had a zero disengagement trip 47% of that time !
Similar threads
- Article
