By the way, didn’t Tesla originally discuss 10 times as safe as human back in 2016? I know they scaled back to twice then, but the narrative keeps changing so it is hard to keep up...
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
The «Full» in Full Self-Driving Capability
- Thread starter lunitiks
- Start date
By the way, didn’t Tesla originally discuss 10 times as safe as human back in 2016? I know they scaled back to twice then, but the narrative keeps changing so it is hard to keep up...
thegruf
Active Member
I'll tell you in 3 months probably; 6 months definitely
Now now, keep your pants on. Ordering page said «twice». Removing that should raise your eyebrows more than enoughBy the way, didn’t Tesla originally discuss 10 times as safe as human back in 2016? I know they scaled back to twice then, but the narrative keeps changing so it is hard to keep up...
strangecosmos
Non-Member
What Elon said at 18:50 on the ARK Invest podcast (my emphasis):
"I do think that we could in principle achieve full autonomy with the Nvidia hardware, but it's a much harder software problem and like I said, you know, like you sort of really have to try to budget your compute and do all sorts of tricks to manage how you use your compute. So it's a harder software problem. With hardware that's, you know, 2000% better, you just, you don't have to do that constant budgeting. And so the software problem is much easier. So like I think with the current hardware and a lot of effort we could get to full self-driving with maybe being like 50 to 100 percent safer than a person. But with Hardware 3, I think it's probably like a thousand percent safer than a person."
Here, I'll read it for you in the way Elon messages should be read nowadays :
1. We no longer focus on highway (past tense, so NoA is as good as it gets).
2. We will focus on intersections (future tense - so not now) - and they are hard (lots of variants) - or so we think since we don't have anything completely working yet.
3. Right NOW we are focusing on stop signs and traffic lights which is the early part of the intersections problem and even that does not yet work all that well because there are many different layouts. This somewhat works and when we see 'em we have no problems recognizing them (but don't see them or have false positives often too -> ambiguity)
strangecosmos
Non-Member
But he said "that's what, you know, we're working on right now". Present tense. And "it's working at a development level". Present tense.
I have little doubt that it's working on a development level. In my software development day job (nothing even remotely related to driving tech), when my boss asks me about a new feature, I can generally hack something together in a day or two to get an idea of how it would work. So yes, "at a development level" our system can now support that feature. Supporting that feature at a production level though is a whole different beast which takes much longer. If there's a bug in my code when it goes live, maybe we temporarily annoy people and maybe revenue drops slightly until things are sorted out.
If there's a bug in handling intersections, people will get seriously injured or killed. "A development level" is basically code for "we have some ideas on a whiteboard" as far as a feature like this goes.
Based on NN design flow, if they have any driving performed then they are past the whiteboard.
context is everything. From the context it is clear they mostly work on the "pre-tasks" of real intersection handling which are stop signs, traffic lights and other stuff they apparently did not even look into (congestion, pedestrian crossing detection, ...). There's no code needed to just cross an intersection - the car can already do this as is even in production versions after all.
@lunitiks
1. I would still recommend AP on HW3 to anyone buying a new Tesla going forward, but would certainly steer anyone to avoid getting stuck with HW2.0 sensors.
2. Regarding FSD&HW3 with v2.5 sensors, I would also still recommend it, with the caveat that it is highly unlikely to ever exceed L3, which IMHO it should achieve ~Jan.2023 if the treacherous Firetruck Super-Destruction mode [for definition see my sig-link] has finally been eliminated. Until then, however, it should still be very useful as an advanced ADAS on most roads by around mid-to-end of 2020.
3. For those like myself who pre-paid for EAP+FSD to L5 driverless robotaxi, I propose we need to be assembling our class-action lawsuit right now in order to ever get proper satisfaction, which I would envisage to be something like an exchange to whatever vehicle actually achieves the promised Tesla Network vision [and I personally was reassured by Tesla sales that not only the computer unit but also sensors would, at no extra cost, be upgraded as necessary on my car to fulfil L5 FSD].
4. Yes, Tesla lawyers really outdid themselves at understatement this time round ... unable even to bring themselves to insert "appropriately" in that quoted sentence, it is left open to the interpretation that an equally acceptable reaction is that the new FSD automatically accelerates across stop signs or steers into traffic lights, which is a supremely generous latitude for declaring the thing feature-complete by end of 2019, whereas it will by then probably be just about half started!
5. I also believe in getting no less than what I paid for and invite you and anyone else in this boat to enlist for the CALS proposed in #3.
6. LOL, yeah, as in we are [for no extra charge] assisting them to develop the actual FSD they once promised! Or maybe mtFSD will prove a more apt acronym, meaning "meandering towards FSD"?
7. Could you please outline this buyback process or link to where you have already done so?
8. Musk's answer there is such a classic, Manafort could have said it!
9. FySD = Fool yourSelf Driving or Full yourSelf Driving, perfect!
1. I would still recommend AP on HW3 to anyone buying a new Tesla going forward, but would certainly steer anyone to avoid getting stuck with HW2.0 sensors.
2. Regarding FSD&HW3 with v2.5 sensors, I would also still recommend it, with the caveat that it is highly unlikely to ever exceed L3, which IMHO it should achieve ~Jan.2023 if the treacherous Firetruck Super-Destruction mode [for definition see my sig-link] has finally been eliminated. Until then, however, it should still be very useful as an advanced ADAS on most roads by around mid-to-end of 2020.
3. For those like myself who pre-paid for EAP+FSD to L5 driverless robotaxi, I propose we need to be assembling our class-action lawsuit right now in order to ever get proper satisfaction, which I would envisage to be something like an exchange to whatever vehicle actually achieves the promised Tesla Network vision [and I personally was reassured by Tesla sales that not only the computer unit but also sensors would, at no extra cost, be upgraded as necessary on my car to fulfil L5 FSD].
4. Yes, Tesla lawyers really outdid themselves at understatement this time round ... unable even to bring themselves to insert "appropriately" in that quoted sentence, it is left open to the interpretation that an equally acceptable reaction is that the new FSD automatically accelerates across stop signs or steers into traffic lights, which is a supremely generous latitude for declaring the thing feature-complete by end of 2019, whereas it will by then probably be just about half started!
5. I also believe in getting no less than what I paid for and invite you and anyone else in this boat to enlist for the CALS proposed in #3.
6. LOL, yeah, as in we are [for no extra charge] assisting them to develop the actual FSD they once promised! Or maybe mtFSD will prove a more apt acronym, meaning "meandering towards FSD"?
7. Could you please outline this buyback process or link to where you have already done so?
8. Musk's answer there is such a classic, Manafort could have said it!
9. FySD = Fool yourSelf Driving or Full yourSelf Driving, perfect!
So when Tesla says stuff like, NOA is complete on highway. I take that as it's done and works.... sorta well. The only way going forward that will improve is from fleet data. To a degree it's out of their hands but will continue to evolve its effectiveness of the task indefinitely. But, the heavy lifting on their end is done. Time to move on to the next....
NOA on street level would follow the same idea... get the ball moving and the fleet will continue the refining push.
NOA on street level would follow the same idea... get the ball moving and the fleet will continue the refining push.
Good one.
"Fleet data will solve it eventually" is a nice placeholder for everything FSD, I guess, second best to "the shadow mode will teach the car".
To save everyone else the time/effort, I emailed support about this tweet by Elon: Elon Musk on Twitter
In that tweet he says “If anyone already paid $3k, they should get $1k back.” Despite knowing with an almost certainty what the response would be, I asked if I qualify for that refund having purchased my FSD for $3k back in 2017. To the surprise of no one, it only applies to “customers who purchased Full Self Driving after the changes were announced, but due to an error, received incorrect pricing based on their delivery date”.
The price of the vehicle fluctuating I’m (mostly) fine with. I’ll let someone else plant their flag on that issue if they so desire. This one though... this one pisses me off.
I swear despite my constant negative/pessimistic posting lately, I really want for Tesla to make me eat each and every one of my negative words by proving them all wrong. I don’t think they will, but I WANT them to.
In that tweet he says “If anyone already paid $3k, they should get $1k back.” Despite knowing with an almost certainty what the response would be, I asked if I qualify for that refund having purchased my FSD for $3k back in 2017. To the surprise of no one, it only applies to “customers who purchased Full Self Driving after the changes were announced, but due to an error, received incorrect pricing based on their delivery date”.
The price of the vehicle fluctuating I’m (mostly) fine with. I’ll let someone else plant their flag on that issue if they so desire. This one though... this one pisses me off.
I swear despite my constant negative/pessimistic posting lately, I really want for Tesla to make me eat each and every one of my negative words by proving them all wrong. I don’t think they will, but I WANT them to.
zmarty
Member
This just dawned on me... The fact that the EAP+FSD package has (at best) remained at the same 8K price with the re-introduction of the FSD package speaks volumes on the value Tesla itself places on FSD.
Do you guys really REALLY think that the software in an actual autonomous car is only worth 8K? In my opinion it should be twice or three times that price. I think this is maybe the reason why we believed these prices will go up, because the package is unbelievably cheap for what Tesla promised it will do. The problem is that this is starting to be unbelievable as in not believable, not in the other sense of the word!
Do you guys really REALLY think that the software in an actual autonomous car is only worth 8K? In my opinion it should be twice or three times that price. I think this is maybe the reason why we believed these prices will go up, because the package is unbelievably cheap for what Tesla promised it will do. The problem is that this is starting to be unbelievable as in not believable, not in the other sense of the word!
I had numerous issues both mechanical and software. The front suspension rattle, door handles (fixed) windows not adjusted and more.
Then all sw issues like autowipers, Spotify issues and the front parking ultrasonics was not working for months and that at no time delivered the promise of EAP nor FSD.
In Norway, consumer laws state that the vendor can have 3 attempts to repair the product. You might then claim a discount, a redelivery or a cancellation of the sale with increasing severity of the issue. Many non severe issues will accumulate towards legal liability as one severe.
I wrote them a polite and factual letter, citing these laws. I counted a sw update as one attempt to repair, and the sum of all repairs and SC visits increased severity. The boss of the SC then called me to a meeting and tried to negotiate a redelivery but we settled on a cancellation/buyback. After a week he came back with a buyback offer iniwhich they, on courtesy reasons without liability, buy my car as a used car. After 1,5 years and 20000km, price was only 5% lower than as new. Car had been 11% of its life at SC.
Good deal for both, and one AP2 FSD less for them to handle in the future. Good service, all polite and effective, no fighting, no expensive lawyers involved.
I suggest you address this in a dialogue instead of a lawsuit, as I believe that will be more productive.
It's hard to disagree with you. And btw congrats on parting with Tesla in a way that you're satisfied with.
But let's face it, lawsuits will ensue if Tesla doesn't act fast and proactively with the rest of their customers as well. Because this FSD debacle stinks. Here's my take on it:
When buying their cars, customers like @boonedocks, @croman and tens of thousands of others, were told that it had
They were also promised «details ... next year» about using their «self-driving Tesla for car sharing and ride hailing» - even for revenue purposes on the «Tesla Network».
Spiced with talk about auto-connecting supercharger cables.
No question about it; Tesla sold thousands of cars and collected tons of FSD money by promising a future driving experience where the car would handle everything («full self driving») in a manner that was substantially safer than a human driver.
Fueled immensely by CEO tweets and interviews.
Today, almost 2 1/2 years later, the same customers - who's got squat for their FSD money so far (except doubt and uncertainty) - are suddenly looking at Tesla dropping all wording about «Tesla Network» and self-charging, and what's worse IMO is that they’re selling new buyers a product with
After 2 1/2 years of development and testing, Tesla is no longer delivering "the" hardware "needed" for FSD "capability", but instead some hardware "capable" of FSD "capabilities" in the future. (Yes, capable of capabilities
) FSD is now a bundle of these very specific features:- NoA
- Autopark
- Enhanced Summon
- Recognize and respond to traffic lights and stop signs
- "Automatic driving on city streets"
And with that, Tesla has redefined the promise (product) that they've been selling for so long, in a meaningful way. There is no longer any promise that FSD will work at a safety level substantially greater than that of a human driver. And the features promised aren't all new features.
See, my AP1 car autoparks. Both parallel and perpendicular spaces.
- It can change lanes automatically, albeit with a nudge on the blinker stick.
- It can do "automatic driving on city streets". TACC and autosteer works on city streets, in fact it works on most streets.
My AP1 car does all those things. Not always in a safe nor comfortable way. But it's capable. I've tried it many times and it works. I love my AP1.
AP2 also does these things, no? So in principle, Tesla just has to deliver on a couple of new features (traffic lights and stop signs + some kind of "enhanced" summon), and they're feature complete.
Perhaps I'm pessimistic. But do I think I'm being realistic when I say that if Tesla doesn't fix this - fast - what they'll get in return is lawsuits complete
I wonder if there will be any recourse for owners in general. A major selling point was that the car would be upgradable to an FSD platform via a paid software update, i.e. it was FSD-capable even if you didn't buy the software right away. It is becoming more and more obvious every week that this is not the case.
Similar threads
