Elon: "Feature complete for full self driving this year"

[mongo]

@mongo We all believe what we believe of course. But would Occam’s Razor really be ”maturity of the system” with Autopilot 2+?

I personally can’t believe an AP update really started in public on the 15th and was halted due to an exotic corner case being reported back.

The idea that it was all wrapped up, as good as it can be, ready for the real world... and then something was reported back in the first day or two of a consumer release and the release halted... no, that does not ring likely to me.

More likely reason for me is that the immaturity of the system became visible in some other testing and / or the the public release did not start at all yet.

Ah I see I misunderstood your post, I was talking maturity of the deployment system, not the specific SW.
That said, in the realm of "simpler solutions are more likely to be correct than complex ones.", software delays are usually caused by problem finding, problem finding is usually the result of testing, testing coverage is based on number of units under test and testing time.
So all sw release times are either based on an assumption of tests passing or the knowledge that all tests possible have already passed. Since AP deals with a universe of infinite possibilities, one can never say all possible input sets have been validated.
Many bugs in our modules were not discovered until the units went to durability test where tens of them ran 24/7. And that was with a known environment...

A date is an expected point in the release process that shifts based on the validation process progression. The validation process itself is staged from AP team to internal testers to external beta/ early access to fleet.

So, yeah, I would also not expect a full fleet predeployment before March 15 with configuration flag flipped at midnight.

 

[electronblue]

@mongo

Everything you say in general is of course true. But again I return to the specifics of Elon Musk announcing and explaining Tesla timelines, in which context I fear everything you said is irrelevant.

I do not believe the general software development reasoning offered by yourself of @diplomat33 are relevant here. In other words, I don’t think it is likely it happened like Musk said/implied it did or for the reasons he gave.

Not because software isn’t hard, but because Musk’s announcements on Twitter have not historically been accurate... for other reasons. The latter is more important to my point than the former. Software development is hard, but even if it were easy Musk would still likely be wrong on this one. That’s my point and that’s why I’m not readily excusing him for generic software-development related reasons.

 

[electronblue]

Basically, I believe Musk likely feels it is in his interest to make the Twitter announcements that he does for other reasons... instead of those timelines being based on realistic (or even aspirational optimistic) estimations. Basically it looks like he is playing for time much of the time.

That’s why I feel any usual logic about the difficulty of software development estimation does probably not apply here.

 

[mongo]

Basically, I believe Musk likely feels it is in his interest to make the Twitter announcements that he does for other reasons... instead of those timelines being based on realistic (or even aspirational optimistic) estimations. Basically it looks like he is playing for time much of the time.

That’s why I feel any usual logic about the difficulty of software development estimation does probably not apply here.

I can understand where you are coming from.

Without deadlines, nothing gets finished. Without something already being finished, no deadline can be accurate.
Is two weeks before the end of a quarter any less accurate than any other date ?? Shrugg.
Does Tesla have an affinity for big October announcements? Yes.
Has Elon stated his timelines are optimistic? Yes.

 

[electronblue]

I can understand where you are coming from.

Without deadlines, nothing gets finished. Without something already being finished, no deadline can be accurate.
Is two weeks before the end of a quarter any less accurate than any other date ?? Shrugg.
Does Tesla have an affinity for big October announcements? Yes.
Has Elon stated his timelines are optimistic? Yes.

I don’t think Elon’s timelines are optimistic.

 

[mongo]

I don’t think Elon’s timelines are optimistic.

I assume you mean, not optimistic for the base reason of being optimistic?
Otherwise, you think he is pessimistic...

 

[electronblue]

I assume you mean, not optimistic for the base reason of being optimistic?
Otherwise, you think he is pessimistic...

I mean I don’t think his public timelines are driven by optimism at all.

Nor on the other hand do I think they influenced to any significant degree by the difficulty of estimating software development timelines.

I don’t even think the public timelines are meant to motivate the team. They have their internal deadlines irrespective of what Elon says (that much is clear).

I feel (fear?) his reasons for public timelines are likely elsewhere. How benign or bad, I haven’t really made up my mind yet.

 

[Bladerskb]

I assume you mean, not optimistic for the base reason of being optimistic?
Otherwise, you think he is pessimistic...

No his timeline are 100% lies and fraudulent. This is i irrefutable.
His published timeline has no connection to the internal timeline that the Autopilot team has.
The software team DOESN'T aim for his timeline, they aim for the timeline of the team.
Each project have a roadmap and a set timeline not the nonsense elon tweets.
Knowing that Tesla is agile, they are most likely doing two weeks springs or something.
Where each developer is assigned a ticket with a certain velocity points (based on hours/days it takes to complete the ticket).
Most people have no idea how software development works so they believe they BS that Elon spews.
This has been proven time and time again.

 

[mongo]

No his timeline are 100% lies and fraudulent. This is refutable.
His published timeline has no connection to the internal timeline that the Autopilot team has.
The software team DOESN'T aim for his timeline, they aim for the timeline of the team.
Each project have a roadmap and a set timeline not the nonsense elon tweets.
Knowing that Tesla is agile, they are most likely doing two weeks springs or something.
Where each developer is assigned a ticket with a certain velocity points (based on hours/days it takes to complete the ticket).
Most people have no idea how software development works so they believe they BS that Elon spews.
This has been proven time and time again.

Yup, I refute that.

Like for instance, that tweet says after 10 million miles of safe driving, they'll turn of confirm. Well, if it still not enough 9s of safety, then guess what? No miles accrued toward the 10 million total. They can't apply miles from a different version to the current one...

And yeah I know what Agile is (like it is a 'sprint'), though how you force that onto a NN type system is unclear to me given the ability of new training data to impact ALL previously implemented features... You can have sprint for labeling and creating test cases for a feature, but that doesn't mean the new NN will be usable.

I am still waiting for you to explain how the pre-ME split AP timelines were examples of bad estimating...

 

[electronblue]

Yup, I refute that.

Like for instance, that tweet says after 10 million miles of safe driving, they'll turn of confirm. Well, if it still not enough 9s of safety, then guess what? No miles accrued toward the 10 million total. They can't apply miles from a different version to the current one...

And yeah I know what Agile is (like it is a 'sprint'), though how you force that onto a NN type system is unclear to me given the ability of new training data to impact ALL previously implemented features... You can have sprint for labeling and creating test cases for a feature, but that doesn't mean the new NN will be usable.

I am still waiting for you to explain how the pre-ME split AP timelines were examples of bad estimating...

Ask yourself this: Why is Elon constantly wrong about software feature timelines and why he rarely if ever retracts or corrects specifically? Why in many cases by months, years, or sometimes we never hear about a promised feature (with a timeline!) ever again?

I don’t think he is aiming for accuracy, that much is certain, because he’d revise timelines much more actively if that was the case.

So he is giving out timelines for some other reason. I suggested earlier a possibility: he is playing for time (and/or for PR effect), which seems possible, and something which would make all your suggestions for reasons irrelevant — just as an example of this thinking.

 

[diplomat33]

What is your issue with this tweet? I don't see anything false about it.

 

[Bladerskb]

What is your issue with this tweet? I don't see anything false about it.

He saying they need 10 million miles (3 days) to remove confirmation in his tweet. Here we are 6 months later.

 

[Bladerskb]

And yeah I know what Agile is (like it is a 'sprint'), though how you force that onto a NN type system is unclear to me given the ability of new training data to impact ALL previously implemented features... You can have sprint for labeling and creating test cases for a feature, but that doesn't mean the new NN will be usable.

Unless you been a software engineer for several large and small companies then you don't plain and simple.

 

[Bladerskb]

I am still waiting for you to explain how the pre-ME split AP timelines were examples of bad estimating...

Mobileye has nothing to do with any elon timeline. AP2 development began sometime in 2015.

 

[electronblue]

He saying they need 10 million miles (3 days) to remove confirmation in his tweet. Here we are 6 months later.

Well technically he is not saying that. He is saying 10 million miles after the system is deemed safe, not 10 million miles from that point of time 6 months ago.

So, basically, never...

 

[diplomat33]

He saying they need 10 million miles (3 days) to remove confirmation in his tweet. Here we are 6 months later.

Uh. He said "10 million miles or so". You are forgetting the "or so" part which grammatically means that the "10 million miles" is vague. It could be 10 million miles or more. So Musk never said that they would remove confirmation when they hit exactly 10 million miles and not a single mile more.

 

[BigD0g]

What is your issue with this tweet? I don't see anything false about it.

If i'm tapping the indicator it's not unconfirmed is it?

 

[electronblue]

Uh. He said "10 million miles or so". You are forgetting the "or so" part which grammatically means that the "10 million miles" is vague. It could be 10 million miles or more. So Musk never said that they would remove confirmation when they hit exactly 10 million miles and not a single mile more.

That would not be a good defence for this. 10 million miles is a few days so any reasonable ”or so” would not be more than a few more days. 10 million miles or so can’t for example mean 10 billion miles and be still right.

But luckily for this rhetoric there is another better defence in my post #675.

 

[BigD0g]

Also, given the number of cars now, 10M miles is what 24hrs if that?

 

[diplomat33]

That would not be a good defence for this. 10 million miles is a few days so any reasonable ”or so” would not be more than a few more days. 10 million miles or so can’t for example mean 10 billion miles and be still right.

But luckily for this rhetoric there is another better defence in my post #675.

Yes, actually, your reading was correct. Musk was saying that after 10 million miles or so, when it was deemed safe to do so, they would remove the stalk confirmation. And that is exactly what Tesla is doing. So, again, Musk has been plenty wrong in a lot of tweets but there is nothing wrong with this particular tweet.