Elon: FSD Beta tweets

[JulienW]

...It can certainly regress in other areas, but they have to show massive improvement for this test or they'll lose face.

Yea because this would be the FIRST time Elon said something that didn't happen exactly as he said it would.

 

[AlanSubie4Life]

Your expectations are ridiculously high…

Whoosh…

I bet $100 that it won't exceed 95% on that test (given realistic situations and 20 or more attempts), but no one countered my bet. I'm wiling to change my threshold to 90%.

I have wagered a beer, and am betting it will be less than 90% (the exact conditions agreed upon defining success and failure are covered elsewhere - it’s an extensive but logical list; the main objective is to ensure that failures aren’t counted as successes; without this protection, that is what would happen). For some reason I agreed that testing without traffic would count, but it was such a ridiculous ask, that I felt sorry for Daniel and just went with it - have to give him a chance. Notably, if the car does not make the turn when there is an opportunity, that is a failure (for each occurrence).

Daniel is very bullish on geofenced FSD. I’m not as bullish because I understand geofencing is a crutch that Tesla will never succumb to.

If it hits 90% it will be an extremely impressive result (in context).

 

[momo3605]

I traded my Model 3 Performance with FSD beta in on a Rivian R1T so 10.13 will be the the first version in a long long of versions that I'll have no first hand account of.

I rarely used FSD beta because I simply wasn't the right kind of person to do so.

I stopped reporting issues because they never responded to any questions or concerns. There was never the satisfaction of reporting a problem, and seeing the problem get fixed. Even a maps only fixes would have allowed me to use it more especially if it would stop it from putting on the turn signals in a corner.

I also rarely had the opportunity to use it because there is almost always someone behind me or someone I have to watch out for when it takes a corner. It was just too stressful to bother with.

When I debated about which vehicle to trade in for the Rivian the AP/FSD stuff didn't factor in. There just wasn't a feeling that it was really going anywhere. I'm sure it will get better as the months drag on, but it was never going to be what I wanted it to be. I wanted something that would make sure I was in the correct lane, and always interpreted an intersection correctly. It could still be L2 and provide a lot of value to me, but even NoA didn't meet my expectations.

Mostly I'll miss the road handling characteristics of a Model 3 performance driven manually. But, a lot of that was ruined by increasingly bad roads. It's to the point where I need a rally car with a full size spare.

Maybe in a few years Tesla will have a Model Y with L3 or L4 highway driving that will appeal to me.

I have both an R1T and Model 3 with FSD Beta. I am still curious where FSD is heading and with my car being worth so much more than I paid, i'm not too worried about holding it for a while longer lol. But I'm in the same boat as you, NoA does not meet my expectations, in fact i keep lane change suggestions, and auto lane change disabled because i hate how it forces you to move when you don't want to. The minute Elon/Tesla says FSD on HW3 development is done, i'll sell my model 3, it's 100% redundant lol.

 

[momo3605]

So are you saying you'll bet against him?

Will you give $100 to a charity of his choosing if 10.13 exceeds his threshold for Chucks unprotected left?

I bet $100 that it won't exceed 95% on that test (given realistic situations and 20 or more attempts), but no one countered my bet. I'm wiling to change my threshold to 90%.

This Chuck thing is fundamentally different from everything else before as not only did Elon mention it specifically, but they're testing in this location.

It can certainly regress in other areas, but they have to show massive improvement for this test or they'll lose face.

Sure lol, i'll wager a $100 charity donation that 10.13 will NOT exceed 95% success rate on Chuck's turn. Has anyone forgotten that the 10.12 release notes had like a whole page of bullet points related to ULT related improvements, yet in my opinion and in Chucks tests, it got WORSE lol

 

[Dewg]

NoA does not meet my expectations, in fact i keep lane change suggestions, and auto lane change disabled because i hate how it forces you to move when you don't want to.

I've always been curious about this one. I've read others with the same complaint. With AV/ADAS in cars, you are essentially a passenger with someone else driving. If you were a passenger and your friend, taxi driver, or Uber driver changed lanes when you wouldn't have, would you hate that/them?

 

[S4WRXTTCS]

Sure lol, i'll wager a $100 charity donation that 10.13 will NOT exceed 95% success rate on Chuck's turn. Has anyone forgotten that the 10.12 release notes had like a whole page of bullet points related to ULT related improvements, yet in my opinion and in Chucks tests, it got WORSE lol

You can't have that bet as that's exactly my bet.

 

[momo3605]

I've always been curious about this one. I've read others with the same complaint. With AV/ADAS in cars, you are essentially a passenger with someone else driving. If you were a passenger and your friend, taxi driver, or Uber driver changed lanes when you wouldn't have, would you hate that/them?

yes i get annoyed when uber drivers stay in slow lanes when traffic is moving fast in another lane.

For instance, NoA will move into the right exit lane 2+ miles before freeway exit. That means that you're going to drive through merge lanes for other exits, and gosh forbid there's another car merging, autopilot will then slam on the brakes to let other cars merge in. And it'll wobble all around the extra wide merge lanes.

Another annoyance is HOV lanes, in certain areas, if you've set it to use HOV lanes, it will force you to use HOV lanes...and alternatively if you don't have it to use HOV lanes, it'll try to kick you out of HOV lanes.

A third annoyance is that NoA never truly acknowledges the cancel option. If NoA wants to move you into a bad lane and you hit cancel...it'll just ask you again in 10 seconds. I could go on and on about the stupidity of NoA

 

[S4WRXTTCS]

I have both an R1T and Model 3 with FSD Beta. I am still curious where FSD is heading and with my car being worth so much more than I paid, i'm not too worried about holding it for a while longer lol. But I'm in the same boat as you, NoA does not meet my expectations, in fact i keep lane change suggestions, and auto lane change disabled because i hate how it forces you to move when you don't want to. The minute Elon/Tesla says FSD on HW3 development is done, i'll sell my model 3, it's 100% redundant lol.

For me the decision was between trading in my 2019 Jeep Wrangler Unlimited Moab Edition with the Power Top or the 2018 Model 3 Performance.

I didn't think it was worth keeping the Tesla out of curiosity of what would become of FSD beta. But, I really wanted to keep it for the driving dynamics and how fast it feels. Oddly the R1T doesn't feel as fast. I think it has a lot to do with how they mapped the throttle response, and even in Sport mode it feels like there is a lot of play.

Anyways the vehicle I enjoy driving the most around town is the Jeep Wrangler which is hilarious when you consider the Model 3 performance is a great drivers car, and the R1T has great reviews. Yet, for some reason I really enjoy driving the vehicle that is known to have the worst driving characteristics. Being able to open the top when its sunny certainly helps the enjoyment aspect.

For road trips I'll take the Rivian. I'm even considering getting an RTT tent for it.

Ideally I'd have a Tesla Camper Van, but there is no Tesla Camper Van.

 

[S4WRXTTCS]

Yea because this would be the FIRST time Elon said something that didn't happen exactly as he said it would.

But, there was never an A/B test before.

Now sure this doesn't come close to that whole Twitter bet.

But, Elon trapped himself into a corner by acknowledging Chucks test.

 

[S4WRXTTCS]

OT but: Running yesterday in west Atlanta (West Midtown) and passed by 7 R1Ts. Not sure what was going on since it is not the Rivian delivery center. Wish Tesla offered some of these BEAUTIFUL pant colors.

View attachment 831021View attachment 831022

Tesla absolutely needs to have more paint colors.

It's to the point where people don't know which Tesla is there Tesla.

In fact if I was buying a Tesla today I'd get a wrap so I knew which one was mine.

 

[S4WRXTTCS]

Whoosh…

I have wagered a beer, and am betting it will be less than 90% (the exact conditions agreed upon defining success and failure are covered elsewhere - it’s an extensive but logical list; the main objective is to ensure that failures aren’t counted as successes; without this protection, that is what would happen). For some reason I agreed that testing without traffic would count, but it was such a ridiculous ask, that I felt sorry for Daniel and just went with it - have to give him a chance. Notably, if the car does not make the turn when there is an opportunity, that is a failure (for each occurrence).

Daniel is very bullish on geofenced FSD. I’m not as bullish because I understand geofencing is a crutch that Tesla will never succumb to.

If it hits 90% it will be an extremely impressive result (in context).

Yeah, I saw the list and I agreed with the list. I would insist though that testing was done at random times to randomize the amount of traffic.

I'm very bullish on geofenced autonomous driving, but I'm not bullish on HW3. I don't believe geofencing is a crutch, but a necessary component due to the safety thresholds humans demand from robots.

What I'm curious about with Chucks test is whether there are very real hardware limitations that prevents it from performing well.

It could easily be a 4 way stop, and in fact there is a wide 4 way multilane stop near my work where I the results of FSD Beta felt like it simply couldn't see everything. Humans can't either, but humans can remember what they did see as they were stopping before other cars blocked their vision.

 

[2101Guy]

didnt elon say a couple years back, that Tesla was going to offer wrapping service themselves?

 

[Dewg]

yes i get annoyed when uber drivers stay in slow lanes when traffic is moving fast in another lane.

For instance, NoA will move into the right exit lane 2+ miles before freeway exit. That means that you're going to drive through merge lanes for other exits, and gosh forbid there's another car merging, autopilot will then slam on the brakes to let other cars merge in. And it'll wobble all around the extra wide merge lanes.

Another annoyance is HOV lanes, in certain areas, if you've set it to use HOV lanes, it will force you to use HOV lanes...and alternatively if you don't have it to use HOV lanes, it'll try to kick you out of HOV lanes.

A third annoyance is that NoA never truly acknowledges the cancel option. If NoA wants to move you into a bad lane and you hit cancel...it'll just ask you again in 10 seconds. I could go on and on about the stupidity of NoA

Thanks, you answered my question in the first sentence.

 

[diamond.g]

didnt elon say a couple years back, that Tesla was going to offer wrapping service themselves?

IIRC Tesla China wraps cars.

Tesla's Own Car Wrap Service Debuts In Chinese Service Centers

This is big news for Tesla since it doesn't offer its vehicles in many colors. In addition, Tesla charges for most paint colors, and it's not cheap.

insideevs.com

 

[AlanSubie4Life]

Tesla absolutely needs to have more paint colors.

In the future all Telsas will have no stamping, with stainless steel panels, and use single-piece castings, so no need for paint. It’s legacy technology which caused other automakers to go bankrupt in the past. Tesla must avoid this fate.

And then they will disrupt the wrap industry, as mentioned above.

 

[nvx1977]

In the future all Telsas will have no stamping, with stainless steel panels, and use single-piece castings, so no need for paint. It’s legacy technology which caused other automakers to go bankrupt in the past. Tesla must avoid this fate.

And then they will disrupt the wrap industry, as mentioned above.

Full Self Wrapping.

 

[Supcom]

For instance, NoA will move into the right exit lane 2+ miles before freeway exit.

Curious. My experience is that NOA moves into the right lane approx 1 mile before the exit. It will move into the second from right lane 2 miles from the exit.

I wonder if there's some other factor that varies this from one locale to another.

 

[KArnold]

Full Self Wrapping.

First, BETA Full Self Wrapping. Should be ready in about 2 weeks to the internal team.

 

[Bladerskb]

This will prove to the haters (like you) that if Tesla uses geofencing they can achieve greater than human performance.

They're probably going to raise the price after 10.13 proves that the left turn can be mastered.

"Yes. We will increase the price of FSD sometime later this year. I think probably just before we go to wide beta, wide beta is anyone who wants to use the beta software with all the caveats associated with that can use it. Then it would make sense to increase the price of FSD. The value of FSD is, I think extremely high and not well understood by most people [he's talking about you]. It is basically currently ridiculously cheap, assuming FSD materializes, which it will."

If we are talking about a geofence like Waymo or Cruise then that is obviously categorically false.
Look no further than the community tracker. They have a huge runway even for a geofenced service.

Home

FSD Community Tracker

www.teslafsdtracker.com

 

[Daniel in SD]

If we are talking about a geofence like Waymo or Cruise then that is obviously categorically false.
Look no further than the community tracker. They have a huge runway even for a geofenced service.

Poe's law - Wikipedia

en.wikipedia.org

Chuck says that he is unaware of any collisions at his infamous turn (he is aware of collisions at the nearby signalized intersection however). The idea that 90% success is remotely close to human performance is absurd. Chuck could spend 8 hours a day for the next year doing that turn and still not prove greater than human performance.