Elon: FSD Beta tweets

[Knightshade]

I haven’t read the details from verygreen if the software can survive a reboot then the parameters can as well. It all depends how the system is set up.

If you haven't read the facts proving what I said was accurate why'd you disagree with my post informing you of those facts?

The space that survives a reboot is essentially CRC checked and can not be changed without a FULL update of the firmware. See tweet below for example.

https://twitter.com/x/status/1320972539808653312

 

[AlanSubie4Life]

I’ve never seen any change in behavior for any release. Always seems the same. It’ll make different errors, but it is all the same, well within normal statistical variation.

Lots of randomness involved.

 

[EVNow]

I’ve never seen any change in behavior for any release. Always seems the same. It’ll make different errors, but it is all the same, well within normal statistical variation.

Lots of randomness involved.

Are you serious ? You have seen no change at all in the last 1+ year ?

Yes, lot of randomness is involved and many things people think are because of a particular release could be just randomness. Thats why repeatability is important before making judgements.

 

[DanCar]

Are you serious ? You have seen no change at all in the last 1+ year ?

Yes, lot of randomness is involved and many things people think are because of a particular release could be just randomness. Thats why repeatability is important before making judgements.

He is talking about changes in behavior for a single release.

 

[sleepydoc]

If you haven't read the facts proving what I said was accurate why'd you disagree with my post informing you of those facts?

The space that survives a reboot is essentially CRC checked and can not be changed without a FULL update of the firmware. See tweet below for example.

https://twitter.com/x/status/1320972539808653312

My original post should have said “in general if the software can survive a reboot…”

That is still true. Even with CRC checking I t’s possible to have multiple files with separate CRC codes. If the parameters are relatively small it’s also possible that the software periodically checks and downloads them so they would be refreshed after a reboot.

 

[swedge]

Has anyone received 11.3.3 right after checking manually for software updates?

I am wondering if the manual checking is broken. I checked yesterday 3/26 at 7:56 am. Today, 3/27 at 11 am it checked again but then said it is up to date as of 3/26 at 7:56, same result as 26 hours ago. Prior checks updated the “as of” time stamp. Any ideas?

 

[AlanSubie4Life]

Are you serious ? You have seen no change at all in the last 1+ year ?

Yes, lot of randomness is involved and many things people think are because of a particular release could be just randomness. Thats why repeatability is important before making judgements.

No, obviously not what I mean, look at the context of my response please.

 

[EVNow]

He is talking about changes in behavior for a single release.

Even for a single release - there have been obvious changes. For eg. the CULT release definitely made a lot of changes to ULT - that was easily noticed.

 

[EVNow]

No, obviously not what I mean, look at the context of my response please.

Thats what I thought ... since there were no quotes - context was kind of missing. I guess I've to go back and read.

 

[AlanSubie4Life]

Even for a single release - there have been obvious changes. For eg. the CULT release definitely made a lot of changes to ULT - that was easily noticed.

I would say that those changes required a software update. Have you been reading what the discussion is here?

We are discussing whether behavior changes can occur after downloading software before getting a new software update.

I’ve never seen that.

In general, of course each software release results in changes. And generally things have improved, a lot.

 

[Knightshade]

My original post should have said “in general if the software can survive a reboot…”

That is still true. Even with CRC checking I t’s possible to have multiple files with separate CRC codes. If the parameters are relatively small it’s also possible that the software periodically checks and downloads them so they would be refreshed after a reboot.

I mean, lots of things are possible.

But none of the ones you suggest are anything Tesla actually has in their code.

The driving code is one single DM-verity checked blob. You can not make changes that survive a reboot without a full firmware update.

There are certainly bizarre kludgy workarounds like telling it "Every time you reboot, download these new values from Tesla that could change at any time online" but that would be flatly dumb compared to just changing how the code works to allow for dynamic software changes without a full-firmware CRC check passing.

Tesla hasn't done that however. Presumably because they do not find value in changing firmware behavior without changing firmware versions.



All of which brings us back to the original fact that behavior does not change between firmware releases other than:

Map updates
or
Wishful thinking and/or people not realizing different conditions like traffic, lighting, etc produce different behavior.


Because the driving code/NNs/weights are NOT changing without a full firmware update.

 

[Dewg]

What people are experiencing are just variations in lighting, angle of attack, position of other objects in perception range, and those variables running through the NNs. It's why the car can feel different taking the same route even at the same time of day.

 

[Supcom]

I would say that those changes required a software update. Have you been reading what the discussion is here?

We are discussing whether behavior changes can occur after downloading software before getting a new software update.

I’ve never seen that.

In general, of course each software release results in changes. And generally things have improved, a lot.

I agree. Any sense of improvement day-to-day is likely confirmation bias.

 

[gsmith123]

How do car settings like Autopilot follow distance get stored? Those stick around through a reboot IIRC. I’m imagining similar settings, except not exposed through the UI. Does the analysis green and others have done thoroughly rule that out?

 

[Knightshade]

How do car settings like Autopilot follow distance get stored? Those stick around through a reboot IIRC. I’m imagining similar settings, except not exposed through the UI. Does the analysis green and others have done thoroughly rule that out?

@verygreen can chime in if he'd like-- I THINK those are stored on the media computer-- not the driving computer.

And at least during the HW2.x-3 transition it was a bit of a controversy that all this personal info (including things like login info for streaming services) was stored unencrypted for anyone who dumpster (or junkyard) dived the old MCUs to pull off at will.

But then nobody can make your car behave badly in actual driving if they change your saved seat position like they could if they could hack neural network weights so it makes at least SOME sense that stuff would be less secure.

 

[sleepydoc]

But then nobody can make your car behave badly in actual driving if they change your saved seat position like they could if they could hack neural network weights so it makes at least SOME sense that stuff would be less secure.

Unless the seat moves so far forward that my knees get caught on my ears. Then my driving gets really bad. And you don’t want to know what happens when my radio station gets changed!

 

[jeremymc7]

Has anyone received 11.3.3 right after checking manually for software updates?

I am wondering if the manual checking is broken. I checked yesterday 3/26 at 7:56 am. Today, 3/27 at 11 am it checked again but then said it is up to date as of 3/26 at 7:56, same result as 26 hours ago. Prior checks updated the “as of” time stamp. Any ideas?

I’ve been manually checking daily. It won’t actually look until 24 hrs have past from last manual attempt.

Last check was 4:28pm yesterday. I tried again today at 3:30pm and it didn’t try. I checked again at 4:30pm and checked and then updated the last day & time.

The release is barely going out so far. It’s going to be a while for most of us. I’m still on 69 myself.

 

[DanCar]

Has anyone received 11.3.3 right after checking manually for software updates?

I am wondering if the manual checking is broken. I checked yesterday 3/26 at 7:56 am. Today, 3/27 at 11 am it checked again but then said it is up to date as of 3/26 at 7:56, same result as 26 hours ago. Prior checks updated the “as of” time stamp. Any ideas?

Mine was stale as of Mar 21. I had to force check like 8 times before it updated and told me I was up to date. In other words: if "as of" time doesn't update, check again many times until it updates.

 

[gsmith123]

The release is barely going out so far.

Isn't at almost 30% of beta testers? Tesla Update FSD Beta 11.3.3 (2022.45.12) Statistics & Roll Out says 28%. That's really not that slow for being three days in. We'll see if there's another push in the next 4-5 hours or so...

 

[kriskros91]

NNs can't be updated without a firmware update. Maps can be though.

I suspect .3 is going wide-ish because it's the first version that both solves the FSD NHTSA recall AND is "good enough" to go wide.

I got it late last night FWIW, haven't gotten to drive on it yet.

Main thing I'm curious about is if the highway stack now being on the fsdb code means it drops back to basic AP less often in moderate rain or not... (which it did just yesterday in moderate rain, dropping from NOA to basic AP and forcing me to do my own exiting of the highway like some kind of peasant)

On local roads previously fsdb seemed better about still mostly working in moderate rain compared to how legacy NoA handles it (though heavy rain still killed it like it does noa)

I drove today for the first time on 11.3.3 and the entire drive home for 30 mins was in moderate rain. No disengagements though I did get the FSD may be degraded message. One time I actually disengaged but reengaged soon after and it switched from AP to FSD surprisingly.

Performance was really good though on V11. Some very noticable improvements. Makes driving much more comfortable. Highway driving is also much more relaxing and human like. Exiting behavior is bad though and I sort of miss having lane confirmations so that I can make sure the car gets to the correct exit the way I want it. But I don't really drive in a way to maximize "zero disengagements". I typically drive in a way to maximize comfort and not being annoyed by FSD. So this lack of control is a bit disconcerting in highway driving. I think I just need to play a more active role if I think it will miss an exit.