FSD v12.x (end to end AI)

[spacecoin]

Just on the very short run. Insurance is a competitive business. Automated driving that never creates accidents will eliminate the need for automotive insurance.

For insurance claims caused by the autonomous driver, yes. There are obviously other types of claims such as break-ins and hit-and run-parking damage.

In some alternate reality where people can subscribe to unsupervised wide-ODD autonomy before 2035, the manufacturer would most likely need to cover AD-incidents and not the car owner.

 

[JB47394]

Just on the very short run. Insurance is a competitive business. Automated driving that never creates accidents will eliminate the need for automotive insurance.

It's about margins. The insurance companies will be just fine as they adjust. Trees will tip over, hail will fall, people will throw rocks, and there will still be tire blowouts, black ice, and so on. People will still want to insure their vehicles. Insurance rates will fall, but so will claims, and the insurance companies will make as much money as ever.

 

[EVNow]

Just on the very short run. Insurance is a competitive business. Automated driving that never creates accidents will eliminate the need for automotive insurance.

We are quite far away from zero accidents. Infact there will always be some accidents - just fewer.

IMHO this will all come down to insurance. If the insurance companies figure out FSD (or similar) is safer, they are going to push people into using these assists more, since most of theme they pick up the tab. It will be interesting, if a little alarming, to see when the first insurance policy comes out that REQUIRES that the car is driving most (all?) of the time.

In auto business (and insurance !) things take a long time to change. I doubt we'll see "FSD required" anytime before most of the cars on the road already have that feature. Well, in that case may be few will buy cars - preferring to hail when needed. Just like nobody buys CDs now

I mean we are just touching 8% new auto sales being EVs now ! It will be 2 decades before most of the cars sold will have FSD - and a decade or two before most of the cars on the road will be FSD. So, we are talking 30 to 40 years. Quite sure the patterns of auto ownership will also change by then.

 

[sleepydoc]

Just on the very short run. Insurance is a competitive business. Automated driving that never creates accidents will eliminate the need for automotive insurance.

I agree with the comments above.

0 accidents will not happen in the near term nor will full 100% autonomous driving. if nothing else there will be edge cases (and parking lots) where people need to take over. Insurance will be a necessary evil for the foreseeable future.

Insurance companies are in the business of risk management and they won't be dropping rates until they have data that supports dropping rates. If the accident rates are lower they'll be more than happy to charge the same rates with fewer payouts. Once there's evidence I imagine they'll start with incentives, something like "if you have an accident while using FSD we'll pay your deductible" or something like that.

 

[drtimhill]

Just on the very short run. Insurance is a competitive business. Automated driving that never creates accidents will eliminate the need for automotive insurance.

Well, in an ideal world. First off, the insurance companies will fight that one to the death. Second, don't under-estimate the ingenuity of humans make self-driving cars dangerous

 

[Supcom]

Well, in an ideal world. First off, the insurance companies will fight that one to the death. Second, don't under-estimate the ingenuity of humans make self-driving cars dangerous

So long as there is wind, gravity and thieves, car insurance will be a necessity.

 

[Kablooie]

Many seem to expect the V12 will be a redo of all the FSD software including the current environment recognition AI. I think this is mistaken. I expect that that the driving control systems will be replace with an AI version but it will still receive all the environmental data from the current AI software. Tesla needs to keep control over which elements are important and which are not, i.e. street signs would be recognized but bus bench ads would not. Without this it could be difficult or impossible to control the system because you could never tell what data the system used for decisions. Using the current AI defined elements along with GPS, map and other data, the new AI would basically just decide the steering wheel, accelerator and brake control positions which is currently done in computer code. It still would be pretty much end to end but not an opaque monolithic black box.

 

[stopcrazypp]

Many seem to expect the V12 will be a redo of all the FSD software including the current environment recognition AI. I think this is mistaken. I expect that that the driving control systems will be replace with an AI version but it will still receive all the environmental data from the current AI software. Tesla needs to keep control over which elements are important and which are not, i.e. street signs would be recognized but bus bench ads would not. Without this it could be difficult or impossible to control the system because you could never tell what data the system used for decisions. Using the current AI defined elements along with GPS, map and other data, the new AI would basically just decide the steering wheel, accelerator and brake control positions which is currently done in computer code. It still would be pretty much end to end but not an opaque monolithic black box.

This was discussed up thread. The more standard definition of end-to-end is you put in the video input to a large NN and then it spits out actions.

The other way is to change all hand coded functions to NN, but keep the same modular structure previously. In the latter, close to "everything" is done by a NN of some sort, but it's not a total black box either.

 

[kpanda17]

Elon said 2 weeks for V12
Imagine they only developed the UI of the Cybertruck on V12?
Then the two weeks inline with 11/30 CT delivery event and v12 makes sense

 

[Artful Dodger]

Elon said 2 weeks for V12
Imagine they only developed the UI of the Cybertruck on V12?
Then the two weeks inline with 11/30 CT delivery event and v12 makes sense

You don't need to imagine that FSD v12 is being developed 1st on HDW3, and that HDW4 will lag by at least 6 months. Because Elon said that too. CT will have HDW4. It won't have v12, because that won't exist yet on HDW4.

 

[sleepydoc]

You don't need to imagine that FSD v12 is being developed 1st on HDW3, and that HDW4 will lag by at least 6 months. Because Elon said that too. CT will have HDW4. It won't have v12, because that won't exist yet on HDW4.

Ah, but you forget - Elon says a lot of things that are imaginary!

 

[Artful Dodger]

Ah, but you forget - Elon says a lot of things that are imaginary!

That's what you got? Denial? Hope? Any EVIDENCE?

 

[drtimhill]

You don't need to imagine that FSD v12 is being developed 1st on HDW3, and that HDW4 will lag by at least 6 months. Because Elon said that too. CT will have HDW4. It won't have v12, because that won't exist yet on HDW4.

You can interpret Elons comment in two ways: It will take 6 months for HW4 development to catch up with HW3 ... OR ... HW4 releases will always lag behind HW3 releases by 6 months for the foreseeable future. It's not clear right now which he meant (possibly even to him).

 

[kpanda17]

how is Tesla announcing the CT and V12 at two announcements this month? Separate or together?
The two week statement, if together, must have them related.
We will know soon

 

[EVNow]

Any EVIDENCE?

Last 5 years - every year he has been saying FSD will be ready this year. What more "evidence" do you want ?

 

[rlsd]

Last 5 years - every year he has been saying FSD will be ready this year. What more "evidence" do you want ?

"We hold these truths to be self-evident"

 

[uscbucsfan]

That's what you got? Denial? Hope? Any EVIDENCE?

What reason would you buy into Elon's timelines? He even makes fun of them. The 2 weeks thing is a joke.

Especially regarding FSD in specific timelines, he's always off by months, if not years.

 

[Mardak]

Tesla / Elon Musk seems to specially treat Thanksgiving / Christmas with holiday updates. Notably for FSD Beta, 10.69 deployment went much wider Thanksgiving 2022 no longer requiring a good Safety Score. Unclear if we'll actually get something end-to-end related for this month or next, but presumably there's some pressure and expectations internally to have something special ready.

 

[Artful Dodger]

What reason would you buy into Elon's timelines? He even makes fun of them. The 2 weeks thing is a joke.

Asok Ellesamy said Dec for first release on FSD v12. Elon's statement is consistant with that. Elon having fun at work doesn't mean that work isn't getting done.

Especially regarding FSD in specific timelines, he's always off by months, if not years.

Always? You have selective memory (or bad rhetoric). One contra-example falsifies your claim. The v12 alpha demo drive was that.

 

[uscbucsfan]

Always? You have selective memory (or bad rhetoric). One contra-example falsifies your claim. The v12 alpha demo drive was that.

That wasn't a delivery of a product or release. Should we list the numerous instances that Elon's Twitter timelines were late or never delivered?

This isn't the investment thread. The people here are mostly testers and we all know the timelines are a joke. There's even instances where Elon jokes about his "2 weeks", whether it was the numerous misses of V11, the promises of level 5/finished FSD, etc. Elon is not one to trust on FSD timelines. If you do, that's fine...I'm sure you own some moon property as well.