OFFICIAL BUTTON WATCH

[drtimhill]

Think about it. You have 50k cars, which gives you about 1,500,000 miles per day which is more than enough to find whatever needle in the haystack your data engine is looking for. Actually its too much data. Remember the bottlenecks all exist elsewhere. Those 1.5 Million miles of day, you are not even using nor will you keep.

Do you have any data to support this? One problem with just measuring data quantity is not allowing for repetition. The bulk of that driving data is going to be the same trips repeated over and over (commute, shopping etc). Sure, some of this will help with different road and weather conditions etc, but the fringe cases are going to require a lot of data before they are captured as well.

 

[rxlawdude]

Like I said, big data is not the big advantage that Tesla fans claim it is, because you want quality over quantity.

Again, without qualification you are declaring that as fact. Please provide authoritative citation, or clarify this is your own biased opinion.

 

[rxlawdude]

I am actually looking forward to the Button a lot. I know it will require driver supervision and there will be interventions but I am eager to try FSD Beta for myself. And having "end to end" would be nice. No more disengaging all the time, because I need to make a turn. Also, I am really curious how FSD Beta handles the roads on the college campus where I work. There are some interesting edge cases. I might even do a video of it.

Why do I get the feeling you'll take great effort to ensure your videos will focus on the edge cases that fail?

 

[Daniel in SD]

Why do I get the feeling you'll take great effort to ensure your videos will focus on the edge cases that fail?

Isn't that the point of a beta?

 

[rxlawdude]

Isn't that the point of a beta?

If the purpose is to denigrate rather than produce constructive input, no. I have seen nothing that indicates the individual has any aspirations other than to find failures in FSD and to flaunt even the most trivial of them.

 

[diplomat33]

Again, without qualification you are declaring that as fact. Please provide authoritative citation, or clarify this is your own biased opinion.

Yes, it is my opinion. And I actually disagreed with bladerskb and said that the data is not worthless and that it has helped Tesla train their NN. I don't get any credit for that? That was also my opinion. I guess I don't have to put a disclaimer in front of my opinion when it is pro-Tesla, only when it is "anti-Tesla"?

Why do I get the feeling you'll take great effort to ensure your videos will focus on the edge cases that fail?

No I won't.

EDIT: it is quite telling that in my entire post saying that I am excited for the button, that V10 might include reverse summon, and I hope it does, and I really look forward to trying it, what you got out of it is that I just want to make videos to trash it. And I just said that my campus roads have interesting edge cases. I did not say that I think FSD Beta will fail. I hope it does not fail as I drive on those roads everyday for work.

 

[diplomat33]

If the purpose is to denigrate rather than produce constructive input, no. I have seen nothing that indicates the individual has any aspirations other than to find failures in FSD and to flaunt even the most trivial of them.

You need look harder then. Yes, I have pointed out some failures because I think we should be honest about FSD Beta. I am interested in providing constructive input. I don't think we should put on rose colored glasses. But I have no interest in finding failures just for the sake of trashing FSD Beta. I purchased FSD on my Model 3 in 2019. Believe me, I want it to work.

 

[rxlawdude]

You need look harder then. Yes, I have pointed out some failures because I think we should be honest about FSD Beta. I am interested in constructive input. I don't think we should put on rose colored glasses. But I have no interest in finding failures just for the sake of trashing FSD Beta. I purchased FSD on my Model 3 in 2019. Believe me, I want it to work.

Yet you minimize Waymo failures. Hmm.

 

[diplomat33]

Yet you minimize Waymo failures. Hmm.

No, I don't think I minimize Waymo failures. I've admitted failures and shortcomings. But I don't highlight and magnify every failure as "proof" that Waymo will fail like some people do.

 

[tpatana]

Nah i disagree. It is essentially worthless because at some point after 10k-50k cars, you start getting diminishing returns on data and the bottleneck becomes something else.
This is why not every Tesla car gets a campaign plus also Tesla trains with hundreds of millions of images and not hundreds of billions.

Think about it. You have 50k cars, which gives you about 1,500,000 miles per day which is more than enough to find whatever needle in the haystack your data engine is looking for. Actually its too much data. Remember the bottlenecks all exist elsewhere. Those 1.5 Million miles of day, you are not even using nor will you keep.

If I was making the system, the car would flag each situation with reliability-score, meaning how sure it was to know what to do, both before it happened and looking back.

All the data where car was almost 100% sure to do well and it did go well, would be filtered out by the big data and NN. It would analyze especially those situations where it thought to be certain but it turned out to go wrong, and those situations where it didn't have confidence. That will narrow down that huge amount of data to somewhat more ingestible for big data.

 

[qdeathstar]

Well, I don’t quite care where you want to drive or not the point is that Tesla sold cars outside of the USA.

Certainly not to me….

 

[Microterf]

Back on track. Still no button today

 

[drtimhill]

Like I said, big data is not the big advantage that Tesla fans claim it is, because you want quality over quantity. I think that is similar to what you are saying.

I'm curious as to how you would measure the "quality" of this data. In fact, as I noted elsewhere, I think in this case there is an argument for quantity. How could you possibly know what data was necessary and what fringe cases you have without amassing a large dataset? By their very nature, many of the fringe cases are unknowable until you stumble on them in that dataset, so yes, I think a large dataset is both advantageous and necessary. In other words, we are not looking for a needle in a haystack, we are looking for anything that isnt hay.

 

[drtimhill]

All the data where car was almost 100% sure to do well and it did go well, would be filtered out by the big data and NN. It would analyze especially those situations where it thought to be certain but it turned out to go wrong, and those situations where it didn't have confidence. That will narrow down that huge amount of data to somewhat more ingestible for big data.

That' more or less what the FSD beta program is for (and is doing), except that the filtering mechanism is the human driver. In fact, that is important, because otherwise how would the car (or any software) figure out "what didn't go well?". Apart from the obvious (the airbags didn't deploy), you have to involve humans otherwise you are not training a car to drive well, you are training the car to do whatever the "what went well" software filter says is good driving.

 

[caligula666]

Certainly not to me….

Oh, sorry, I didn’t realize it was a thread about you, but now that you say it…

 

[AlanSubie4Life]

So I think there is a chance it could also include reverse summon.

Now that’s the enthusiastic and optimistic @diplomat33 I remember from the heady days of 2019!

I would estimate zero chance of reverse summon in the early releases of The Button, should the release ever occur.

 

[Bladerskb]

We don't know that.

I've worked in companies whose products where used by Billions (yes "B"). There were always surprises and edge cases to be found. There was no such thing "too much data". You just have to be good at filtering.

This is completely not true. It’s a huge fallacy that what’s keeping you from solving a problem is some edge cases that exists out there that you will only find if you had x million whatever and as soon as you get that data. Everything is solved. It’s completely nonsense. All the edge cases that happens with driving. Happens very very very often. Anyone else that says anything is selling you hogwash PR. Your system can’t disengage 10 times per mile and you are claiming you need more edge case data. Your system can’t even do the most basic task of driving.

It’s the same way when people complain about regulators stopping them when their system is terrible. The boogeyman myth.

 

[Bladerskb]

Actually, we do know that Tesla could use *more* data. Even Elon talks about having to simulate certain crashes because there aren't enough examples of them.

We know that Teslas crash about once every 3-4 million miles, according to Tesla. There's around 500k+ of the hw3 cars in the USA, so there's only 500k * 20 miles / 3,000,000 miles = 3.33 crashes a day, which aren't that many if you want a lot of crash data.

This is 100% not true and is something only sheeps would believe. If a system has a safety disengagement per mile for example but you are complaining about hidden edge cases.

Its the same as Elon claiming for 7 years that what’s gonna stop them is regulation when their system is beyond terrible.

 

[Bladerskb]

The cool thing about Tesla's data advantage is that it doesn't disappear when they retrain their NNs.

They don’t have a data advantage, it’s a debunked myth.

Their data pipeline has undergone massive improvements over the last 4 years. Every time they modify the architecture, they "retrain" using that same data they've collected, plus whatever new data they have (and potentially minus some data, if they want).

They collect 0.01% of data from their fleet. Everything else is thrown away because there’s too much of it. Even there are TOO MUCH of what they want.

And they're working on improving their data set constantly.

"Retrain" in the context of NNs isn't "go back to square one" nor does it mean "rewrite" or anything like that. It's simply a normal part of the development of ML systems. Their data pipeline is kind of like a factory, their labeled datasets are like the raw materials, and the NNs that get deployed to the cars are like the finished product.

Every time they "retrain", that just means they tweaked the pipeline and have to shovel all the labeled data though it again. I think at AI day they said it takes a few days with their (thousands of GPUs) cluster to retrain the perception NNs.

And finally... visual perception doesn't drive cars. While necessary, it isn't sufficient. There are challenges that Tesla is still ironing out, that aren't solved by just having the best dataset.

But I don't think you're giving Tesla enough credit. FSD seems to be very solid for something that many experts claimed to be impossible (or at least many decades away) 6 years ago. They've made a ton of progress.

1) First of all Tesla’s FSD is not very solid. The safety disengagement per mile isn’t good.
2) SDC experts don’t claim that safe camera only FSD is impossible, they have always said it would take over a decade and they have been 100% right. Tesla fans want to pat yourself in the back over something no sdc expert ever said. While what they HAVE said has been 100% right.

3) There were several camera only FSD systems before Tesla. Are you giving credit to those companies?

 

[diplomat33]

Now that’s the enthusiastic and optimistic @diplomat33 I remember from the heady days of 2019!

I would estimate zero chance of reverse summon in the early releases of The Button, should the release ever occur.

Well, we saw the new vision based Auto Park on the refresh model S earlier this month. So it is conceivable that they might be working on reverse summon in V10. But I'm probably too optimistic.

Tesla Autopark now uses a 'vision-based' approach

Tesla’s Autopark is making headlines this morning as the Full Self-Driving suite capability has transitioned to a “vision-based” approach for operation. Tesla hacker @greentheonly tested the new operation of Autopark and posted the results to his Twitter feed, where he regularly updates owners...

www.teslarati.com