Elon: FSD Beta tweets

[AlanSubie4Life]

My point was you didn't' see any path to it, and I see no reason they don't have such a path--- if they should take it is a separate thing.

Yeah I thought that phrasing might be the issue.

I’m actually not convinced there is actually a path in a meaningful ODD, but I certainly don’t see any path to it for us.

 

[Knightshade]

Yeah I thought that phrasing might be the issue.

I’m actually not convinced there is actually a path in a meaningful ODD, but I certainly don’t see any path to it for us.

From my experience, and again like 95% of my highway miles since late 2018 are on AP then FSD, the ODD could be "controlled access divided highways in good weather up to light rain" and that'd be pretty meaningful, and require nothing out of Tesla other than 2 improvements from where we were pre-FSDb even:

Fixing the hitting stationary objects problem (which FSDb seems likely to be well capable of in highway situations)
and
OEDR for road debris- that's the only tough one, and we've seen SOME hints of this in FSDb already


Again it likely doesn't make business sense for them to do it unless forced by competition-- but on the technical side it seems quite doable.

 

[AlanSubie4Life]

OEDR for road debris- that's the only tough one, and we've seen SOME hints of this in FSDb already

but on the technical side it seems quite doable.

Not convinced. There is a long way between hints and actual reliable capability.

It also has to dodge animals, etc. Things not all humans may be able to do. Possum dodging at 20-25mph doesn’t count or hint at capability.

Very very tough problems. Even in a very limited ODD in optimal conditions.

Weird stuff happens all the time.

Anyway time will tell.

 

[jeewee3000]

Not convinced. There is a long way between hints and actual reliable capability.

It also has to dodge animals, etc. Things not all humans may be able to do. Possum dodging at 20-25mph doesn’t count or hint at capability.

Very very tough problems. Even in a very limited ODD in optimal conditions.

Weird stuff happens all the time.

Anyway time will tell.

Autonomous vehicles won't ever be perfect, nor should they.

As long as the accident rate is better than humans overall we're good.

Elon stated many times that HW3 will probably max out at 2x better than human. HW4 5x, HW5 10x and so on.

Which will be monetizable? Depends on the whishes of the regulators.

 

[Knightshade]

Autonomous vehicles won't ever be perfect, nor should they.

As long as the accident rate is better than humans overall we're good.

Elon stated many times that HW3 will probably max out at 2x better than human. HW4 5x, HW5 10x and so on.

Which will be monetizable? Depends on the whishes of the regulators.

TBF Elon stated a lot of things about FSD that turned out less than accurate.... I mean originally he said HW2 was enough for self-driving. Then HW2.5 was. Now he's still claiming HW3 is but it seems very unlikely to be true.

NOBODY knows how much HW is enough until someone actually delivers it.

 

[diplomat33]

Autonomous vehicles won't ever be perfect, nor should they.

As long as the accident rate is better than humans overall we're good.

Elon stated many times that HW3 will probably max out at 2x better than human. HW4 5x, HW5 10x and so on.

Which will be monetizable? Depends on the whishes of the regulators.

I would take those safety rates with a huge grain of salt because we have no idea where they come from. Elon has not shared any data or safety methodology to show how he arrived at those numbers. For all we know, they are just Elon's guesses. He probably pulled them out of thin air.

 

[jeewee3000]

I would take those safety rates with a huge grain of salt because we have no idea where they come from. Elon has not shared any data or safety methodology to show how he arrived at those numbers. For all we know, they are just Elon's guesses. He probably pulled them out of thin air.

Probably. But the gist of it rings true. Every next hardware set will be more capable.

Unless you believe camera placement caps performance.

 

[Knightshade]

Probably. But the gist of it rings true. Every next hardware set will be more capable.

Unless you believe camera placement caps performance.

The most problematic gist though is that HW3 is capable of safer than a human, without a human driving, at all

Which very much remains a fact not in evidence at this time.

 

[Bouba]

As long as a Tesla is safer than a Jaguar all is well

WATCH: "Rogue" Jaguar mounts Porsche before smashing into Tesla

FOOTAGE shows the shocking moment a “rogue” Jaguar mounted a Porsche Boxster before rear-ending a Tesla.

www.bournemouthecho.co.uk

 

[willow_hiller]

The most problematic gist though is that HW3 is capable of safer than a human, without a human driving, at all

Which very much remains a fact not in evidence at this time.

I do wish we had more transparent data on FSD Beta's performance with each successive approach. Even something as basic as median miles per disengagement would give us a good sense of whether the improvement over time is exponential, linear, or logarithmic. If it's logarithmic, that would give us an indication of where the ceiling is without a big change in hardware.

My first FSD Beta version was 10.69.1, so I've only experienced 3 major revisions. In retrospect, the improvements from v10 to v11 felt very marginal, and would probably show logarithmic growth. v11 to v12 feels much more substantial, but still hard to quantify in the aggregate.

 

[jakimo]

For us daily highway HOV road warriors, AutoPilot is orders of magnitude better than FSD B.

I'm curious, have you tried FSDb 12.3 with Auto-speed correct turned on? I've never been more impressed with the level of improvement, smoothness, and reliability compared to any other update

I thought I'd miss the TACC, and I realized after updating maybe four days ago.. that it was primarily because I couldn't enjoy the previous FSDb experiences enough to use it often. The current update is so much better, that I barely miss TACC.. so now I've been jumping in and out of FSDb on my own drives, jumping out just so I can do some fast things on the road. With my wife and kids, it's the FIRST time they enjoy FSDb (where I don't make fast moves at all) from door to door with ZERO ISSUE

I can't say enough how much better it's been while using it here in the Colorado Springs area

PS - I love to drive, this is just fun for me. So I also could never imagine BUYING FSD from the start.. I'd rather hit the on button for 30 days at a time, never needing to let it auto-bill me. It's enjoyable to see the progress now and then, but I usually let it lapse.. then re-engage the FSDb program when I want to see how things have improved

 

[AlanSubie4Life]

As long as the accident rate is better than humans overall we're good.

Sure. But nowhere near that unsupervised anywhere, yet.

Every next hardware set will be more capable.

I was only talking about what I think is the very low likelihood of better than Level 2 with current hardware, to be clear.

 

[OxBrew]

As long as the accident rate is better than humans overall we're good.

Define better. Better than 100% of humans? The best humans go decades with no crashes.

Better than the commonly quoted 50%?

That might cut some % of crashes, once we get 100% adoption, without hard data, it feels like a few % improvement, maybe.

Average humans crash every 200k miles. Good drivers crash every 1M miles.

You're going to ask the best human drivers to ride around in Robo taxis that crash 5 times as much as they do? F--- that, not happening.

The whole Autonomous vehicle value proposition is dubious at best.

Current state of Tesla FSD is about .5% of where it needs to be: 1M miles per crash, before we can make any claims about societal benefit.

latest v11 FSDb on a 2024 Y with 300 miles on it tried to crash about 6 times per block for me yesterday.

What the hell are we even talking about this for? The whole thing is a sad joke. No idea how it has not been shut down.

 

[boonedocks]

I'm curious, have you tried FSDb 12.3 with Auto-speed correct turned on? I've never been more impressed with the level of improvement, smoothness, and reliability compared to any other update

I thought I'd miss the TACC, and I realized after updating maybe four days ago.. that it was primarily because I couldn't enjoy the previous FSDb experiences enough to use it often. The current update is so much better, that I barely miss TACC.. so now I've been jumping in and out of FSDb on my own drives, jumping out just so I can do some fast things on the road. With my wife and kids, it's the FIRST time they enjoy FSDb (where I don't make fast moves at all) from door to door with ZERO ISSUE

I can't say enough how much better it's been while using it here in the Colorado Springs area

PS - I love to drive, this is just fun for me. So I also could never imagine BUYING FSD from the start.. I'd rather hit the on button for 30 days at a time, never needing to let it auto-bill me. It's enjoyable to see the progress now and then, but I usually let it lapse.. then re-engage the FSDb program when I want to see how things have improved

Our ‘17 Model X got updated to the 2024.X.X FSD beta 11.X branch so I won’t be able to try 12.3.X version until they do a full on “catch up” for all of those cars.

I won’t re-enroll in to the monthly subscription until I see some MAJOR improvements.

Until then regular’ol AutiPilot is so much easier and comfortable to use on long hauls.

Some hidden “secrets” that you get with it too that the FSD Beta takes away

 

[jeewee3000]

Define better. Better than 100% of humans? The best humans go decades with no crashes.

I did define it: humans overall. In other words, better than the average crash/injury/death statistics.

 

[AlanSubie4Life]

I did define it: humans overall. In other words, better than the average crash/injury/death statistics.

I’m not saying that is the correct metric (see above for counter argument), but in any case it is nowhere remotely near that currently. Seems very likely impossible with current hardware even in a limited ODD.

Just nowhere close, unfortunately.

 

[aronth5]

Are others at the same place I am now with V12 where I'm a bit surprised to have a disengagement on a drive? Granted most drives aren't in an urban area but with every other version of FSD I often had multiple disengagements per drive.

 

[AlanSubie4Life]

Are others at the same place I am now with V12 where I'm a bit surprised to have a disengagement on a drive? Granted most drives aren't in an urban area but with every other version of FSD I often had multiple disengagements per drive.

I have disengagements on most drives with occasional critical safety disengagements. One zero disengagement drive.

Definitely fewer than prior version though. I engage it now when leaving the house (for fun; it’s definitely easier and safer to just drive myself) and wait for it to fail, rather than not bothering to try.

 

[AlanSubie4Life]

https://twitter.com/x/status/1771409645468529047

“Probably” the key word here.

 

[KArnold]

“Probably” the key word here.

So.....he doesn't know? Sounds like our current presidential administration.