Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

Elon: "Feature complete for full self driving this year"

This site may earn commission on affiliate links.
I'm not sure I agree actually. What's the greenhouse gas impact of the BFR vs slower means of transport? I know airplanes aren't the most environmentally friendly but rockets historically have been quite terrible, not only on greenhouse gas emissions but release of other pollutants. The only reason it's not a big deal is because there are so few rocket launches... except now there are more and more. I think it's getting to be a noticeable contributor.

That is a concern. I was just looking at it from the perspective of the huge potential. Any tech, like the BFR, that can shorten travel times around the globe to less than 1 hour, has huge potential and implications for our civilization. Basically, it would bring us one step closer to becoming a planetary civilization!! So it's pretty huge!
 
  • Like
Reactions: pilotSteve
I'm not sure I agree actually. What's the greenhouse gas impact of the BFR vs slower means of transport? I know airplanes aren't the most environmentally friendly but rockets historically have been quite terrible, not only on greenhouse gas emissions but release of other pollutants. The only reason it's not a big deal is because there are so few rocket launches... except now there are more and more. I think it's getting to be a noticeable contributor.

You may be thinking of solid rocket boosters in term of other pollutants. Shuttle main engine only produced water. BFR is methane and O2, producing H2O and CO2. If SpaceX went green by generating their own methane (like they need to do on Mars), then BFR would be carbon neutral. Once ships are heading to the moon or Mars, that would definitely carbon negative (if self generated, since the CO2 would not make it back to Earth).
Methane is a worse greenhouse gas than CO2, so if the engines run rich, they could could cause a net increase in GHG.
 
I would also point out that the world needs visionaries who ask "what if?" and push the envelope. Even if their future predictions don't come true right on schedule, they push everyone to innovate and do better. I want to see hyperloop become a reality someday. I want to see BFR routinely take passengers around the world in 30 minutes as easily as a today's commercial fights. I want to see a burgeoning Mars colony. So, if Musk helps makes that happen a bit sooner then that is a win. It's better to be daring and bold and get there late than to not even try at all.

I'm not sure I agree actually. What's the greenhouse gas impact of the BFR vs slower means of transport?

Both of you are so accurate. While the BFR vs slower transport that rnortman points out could be negative - The importance that diplomat33 states of being a visionary is very key to our continued development as a society. I am not sure about all, but I am almost 55 - and when I was a kid watching Start Trek, Star Wars, TNG, etc. etc. etc. and then when we created the first space Shuttle - Wow that had me going - something to look forward to, something to strive for. Those thoughts, those actions, attempting to reach the Stars - That boys and girls is what drives us as people and explorers, something to wake up to and think of the possibilities. Rather than the day to day drudge, watching the news of who killed whom, who the the F'ing Kardashington's has slept with today, what type of crap Tweet our Commander in Chief has posted today - that is not it, and we as a people should not be going down that path as it is the most important thing in the world.
As it is not.

Sorry for the spew....
 
I have a slightly different approach to this one. I am wondering when the video was made if they thought they were close to completing FSD - just needing to complete and verify some of their development. However also remember all of the big shake ups that were happening in the Auto Pilot department within Tesla. (Mind you my memory is not very good, so please take this with a grain of salt) But what I 'believe' I remembered was a lot of us were complaining about how EAP/FSD was not even on par with AP1 (before Karpathy) then the one head guy from the Auto Pilot team quit, and Karpathy took his place. If memory serves me, it took several long months but under Karpathy's direction the NN version emerged, which started getting us towards that AP1 parity and finally to where we are today.
So the way I am 'now' looking at the original FSD video we are discussing, is I am (or I am wanting to believe) that Musk was under the impression they were getting close to that cross-country demo, but as they kept peeling back the covers, and (assuming kept adding /if, else, next loops) into their original EAP/FSD code - they just could not get over that hump to actually make it work right, and pretty much had to start from scratch. Which is why we still don't have FSD or even automatic NOA.
Just my $0.02

Your memory is correct, and it's only a few details that you might not be aware of.

Tesla didn't have anything back in 2016.

They were stuck between a rock and hard place.

The rock being MobileEye pretty much fired them as a customer after the Florida fatality. MobileEye claims Tesla went too far with the Autopilot system despite knowing exactly what Tesla was doing for months before the accident. They just conveniently ignored it until it became bad for their image. So MobileEye made the decision not to provide Tesla with anything beyond what Tesla was already using.

Then Tesla had the brilliant idea of using the HW1 chip from MobileEye alongside the NVidia Drive PX2 that they had slated for AP2. This royally pissed off MobileEye, and MobileEye told them no chips for you.

This meant when AP2 was introduced that Tesla had NOTHING.

This is why the development took so long, and the unwillingness Tesla had in partnering with anyone.

The self driving video they posted used NVidia Drive Software

Something Tesla didn't want to use for real as that would lock them into NVidia, and Tesla didn't want that.

So two things are essentially true.

Tesla did an amazing job of creating their own ADAS system in only roughly 2.5 years. That's pretty darn impressive if you look at it from the perspective that no other car company had ever done that.

The people who bought AP2 were the ones who got shafted. They were the hard place, and they were the ones to give. They either got nothing or alpha level crap for at least a year and half. They still get nothing as AP2 can't do Sentry Mode or Dashcam. The AP2 owners are the walking wounded. They probably have PTSD from false braking events.

Nothing really came together until Karpathy took the reins.

It's now Karpathy or bust.
It's now HW3 or bust (HW2.5 simply isn't fast enough, and to get the best performance you really need to make sure everything you do is supported by the tools NVidia provides to accelerate neural networks. Things like TensorRT).

If it wasn't for Karpathy I probably would have kept my AP1 car until the Porsche Taycan became available. I liked what he was doing to fix the ship, and so I decided to go one more time around with HW2.5/HW3

I'd prefer Ian Goodfellow because he'd make everything out of adversarial networks. :p
 
Last edited:
Tesla did an amazing job of creating their own ADAS system in only roughly 2.5 years. That's pretty darn impressive if you look at it from the perspective that no other car company had ever done that.

It's now Karpathy or bust.

It's these 2 things that give me hope for "FSD". Our cars are literally the proof of what Karpathy can do with AP2 and it is impressive indeed. And now Karpathy is being given an AP3 chip that is faster and custom designed to handle even more NN. So yeah, I want to see what he can do with that. Maybe it won't be L4 autonomy, but I expect Karpathy will achieve some pretty cool stuff with AP3.

So, it's not "In Elon We Trust", it's "In Karpathy We Trust". :D
 
It's these 2 things that give me hope for "FSD". Our cars are literally the proof of what Karpathy can do with AP2 and it is impressive indeed. And now Karpathy is being given an AP3 chip that is faster and custom designed to handle even more NN. So yeah, I want to see what he can do with that. Maybe it won't be L4 autonomy, but I expect Karpathy will achieve some pretty cool stuff with AP3.

So, it's not "In Elon We Trust", it's "In Karpathy We Trust". :D
In Karpathy, Elon Trusts.
(But verify;))
 
In Karpathy, Elon Trusts.
(But verify;))

Yeah, Elon's tweets depend on Karpathy being successful.

Actually, maybe we focus too much on Musk when it comes to FSD. Yes, he is the CEO so the buck stops with him. And yes, when CEO tweets, it is going to garner a lot of scrutiny. But when it comes to FSD development, Karpathy is the one who is actually in charge, not Musk. He is the Director of Artificial Intelligence and Autopilot Vision at Tesla. And Karpathy is an expert in this area, unlike Musk. Karpathy has an undergraduate degree in computer science, a masters degree in physically simulated figures and a PhD in intersection of natural language processing and computer vision, and deep learning models. He also specializes in deep learning, image recognition and understanding. And we know that his method has produced real successes. We've seen concrete improvements in EAP through software updates that were the direct result of his work. In just 2 years, using machine learning, Karpathy transformed EAP from crap to NOA. So I am optimistic, not because of Musk, but because of Karpathy. Give an AI expert like Karpathy hardware like AP3 and I think we will definitely see some cool stuff.
 
Yeah, Elon's tweets depend on Karpathy being successful.

Actually, maybe we focus too much on Musk when it comes to FSD. Yes, he is the CEO so the buck stops with him. And yes, when CEO tweets, it is going to garner a lot of scrutiny. But when it comes to FSD development, Karpathy is the one who is actually in charge, not Musk. He is the Director of Artificial Intelligence and Autopilot Vision at Tesla. And Karpathy is an expert in this area, unlike Musk. Karpathy has an undergraduate degree in computer science, a masters degree in physically simulated figures and a PhD in intersection of natural language processing and computer vision, and deep learning models. He also specializes in deep learning, image recognition and understanding. And we know that his method has produced real successes. We've seen concrete improvements in EAP through software updates that were the direct result of his work. In just 2 years, using machine learning, Karpathy transformed EAP from crap to NOA. So I am optimistic, not because of Musk, but because of Karpathy. Give an AI expert like Karpathy hardware like AP3 and I think we will definitely see some cool stuff.

The problem is Musk undermines his work.
 
It occurred to me: the problematic tweets from Musk only seem to be with Tesla's FSD. Unless I missed something, Musk's SpaceX tweets are generally more reliable. He'll tweet a pic showing what SpaceX is actually doing and seems to avoid too many far off predictions that get missed. I mean, I don't see him tweeting that BFR will be done in "3-6 months" when they are still years away.

Falcon Heavy was, what, 5 years late? A lot of Elon's SpaceX predictions have been 1-5 years late. (Source: Bloomberg.)

I think some people interpet Elon's comments in an extremely uncharitable way. No one can reliably predict when these things will get done. It's not just Tesla and SpaceX — all kinds of companies whizz past the original timeline for a product without launching. For example, Audi and Nissan haven't launched the ADAS products they said they would have by now. I believe Volvo cancelled its plans for an SAE Level 3 product altogether. (See the thread “Four Upcoming Self Driving Level 3 Cars by 2019”.)

ULA was supposed to launch the Vulcan rocket in 2019, then it was pushed to 2020, now pushed to 2021 — and we'll see it really launches that year. So it’s 2 years late and counting. (Source: ULA now planning first launch of Vulcan in 2021 - SpaceNews.com)

In 2012, Sergey Brin said about Google’s self-driving cars: “You can count on one hand the number of years until ordinary people can experience this.” So we are 2 years late and counting on that one too. (Source: Self-driving cars a reality for 'ordinary people' within 5 years, says Google's Sergey Brin)

Elon is not exceptional in kind, he's just exceptional in degree. He seems more likely than most execs to lowball how much time something will take, and he is atypically open about his best guess about when something will be ready. Some execs will just decline to answer questions about "when", but Elon always answers and often gives a timeline unprompted. Personally I like that Elon is transparent about his thinking. I prefer people to be forthcoming about they think when they might turn out to be wrong, rather than to be withholding with their thoughts in order to avoid making a public mistake. Mistakes are good. They are fuel for intellectual progress.

Making mistakes publicly also fosters humility and discourages people from BSing by pretending that they’re always right.

It's unrealistic to say you have to be incompetent or lying to not be able to precisely predict the future of cutting-edge technologies in space and AI. It’s a fact of life that these things are hard to predict.

Across many domains, experts are often bad at predicting the future. The unrealistic assumption is that the future should be easy to predict, and if you get it wrong that means you lied or were negligently ignorant. No. That is not true. No one can predict the future of technology. The default assumption should be that predictions are probably wrong, and it should be surprising when they’re right.

What makes SpaceX and Tesla different than, say, Apple is that SpaceX is trying to design a new interplanetary spacecraft and Tesla is trying to push the boundary of science and engineering in AI. This is not like launching a new iPhone that is incrementally different from the last iPhone. SpaceX and Tesla have to create new knowledge in uncharted territory.

The flip side is that we — the public, enthusiasts, commentators — have to honour our part of the bargain. If we want people to be forthcoming about their thoughts, we can’t attack or accuse them when they do. If we want people to make mistakes, and do it publicly, we can’t act like making a mistake is unacceptable or humiliating. If we want people to take the risk of creating new knowledge, we have to be supportive when the journey into uncharted territory doesn’t go smoothly.

The “safer” option is always not to try. We want people to try, don’t we?
 
Last edited:
If Tesla had nothing when AP2 launched as @S4WRXTTCS says, then Tesla lied by saying the system awaits validation — there would have been no system to validate yet.

Moving onto the HW3/Karpathy saves angle, I am old enough to remember when it was the ”FSD codebase” saves. Or any number of new names that replaced the old ones, or new features or versions promised in Elon tweets...

Trust me: If Karpathy/HW3 fails, the narrative moves onto believing in the next big thing.

This is how Tesla got away with AP1 promises, P85D HP promises, all the P90D failures etc. Right now on Model X forum there is a thread on 90 kWh degradation (which is a real problem with that product) and people are talking about buying new Teslas as the solution.

Nice work for Tesla if you can get it...
 
Last edited:
@strangecosmos No, we don’t want people to break ethical codes in the names of progress.

We don’t want people to lie to make progress. The end does not justify the means.

As for Apple (or Jobs, if you include Musk’s other enterprises), you forget history if you think their contribution to the world is iterating on iPhone. But then again I never liked it or was silent when Jobs was unethical either.
 
Last edited:
I already have a suggestion:

In Goodfellow/HW4 we trust! :)

Also how about this ”the new-FSD codebase is different from the old-FSD, just wait for it, after they merge all the EAP bits to FSD to reflect the new Design Studio, it will be very different compared to today — just wait for it”.
 
Last edited:
  • Like
Reactions: rnortman
Trust me: If Karpathy/HW3 fails, the narrative moves onto believing in the next big thing.

This is how Tesla got away with AP1 promises, P85D HP promises, all the P90D failures etc. Right now on Model X forum there is a thread on 90 kWh degradation (which is a real problem with that product) and people are talking about buying new Teslas as the solution.

Great job for Tesla if you can have it.

I fail to see how AP1 promises or any of those other things measure up to the sheer scale of the singular promise of FSD.

Any company that produces products will have its fair share of mistakes. I type this on a MacBook Pro that works just fine, but I know lots of people with the same model have display issues due to flexgate. For Apple to get anywhere close to FSD they'd have to promise a Quantum Computer.

I can't think of the last time any established company made such a massive promise that is FSD.

Karpathy can't save HW3 in terms of FSD as it can't be saved solely by having better vision processing,.

Karpathy's entire purpose is to make sure Tesla is in position when everything comes together in 2 years for L3 driving. At which time hardware specifications for L3+ driving will be a lot more standardized.

2021 will be the year of L3 driving and Tesla has to be ready.

It won't be with HW3 though.

If Tesla fails to achieve L3+ driving when other EV manufactures have L3+ driving there won't be a next big thing.

A fantasy is great when reality doesn't give us what we want. Especially when that fantasy gives us glimpses to what it might be like.

When reality gives us what we want there is no point in a fantasy.

Nissan didn't get away from the battery issues with their Leaf. Instead they lost customers to Tesla.
Tesla didn't lose any P85D owners because they didn't have anywhere else to go to.
 
@S4WRXTTCS Sure Tesla might fail as a company, but barring that the conversation just moves on to the next big thing when the last one fails. That’s how it has gone and that’s how it will likely go with Tesla. Due to the unique dynamic inside the company and outside it. (Barring a radical change inside and out of course.)

I thought Karpathy’s role was to deliver on Tesla’s past promises, but glad to see you are already taking a forward-looking stance, now we are discussing Level 3 that Tesla has not ever even promised themselves. ;)

Tesla has the best people following it, for them, really does.
 
I already have a suggestion:

In Goodfellow/HW4 we trust! :)

Also how about this ”the new-FSD codebase is different from the old-FSD, just wait for it, after they merge all the EAP bits to FSD to reflect the new Design Studio, it will be very different compared to today — just wait for it”.

What I ponder if whether Goodfellow can save autonomous driving in general.

Autonomous driving doesn't currently work.

It has a massive problem that no one has solved.

That problem is humans.

We don't follow written rules
We bully
We don't particularly care about our own safety.
We have a different set of unwritten rules depending on where we're driving.
We liter the road with debris because we're idiots.
Occasionally we just go berserk for no particular good reason
 
@S4WRXTTCS Sure Tesla might fail as a company, but barring that the conversation just moves on to the next big thing when the last one fails. That’s how it has gone and that’s how it will likely go with Tesla. Due to the unique dynamic inside the company and outside it. (Barring a radical change inside and out of course.)

I thought Karpathy’s role was to deliver on Tesla’s past promises, but glad to see you are already taking a forward-looking stance, now we are discussing Level 3 that Tesla has not ever even promised themselves. ;)

Tesla has the best people following it, for them, really does.

As a company I don't think Tesla will fail.
As a leader in autonomous driving I think failure is more likely than success. They just don't have the resources of other companies or the level of focus on it.

With Tesla past promises don't mean a whole lot when it comes to exactness. You got the AP1 promises from the Tesla blog, but it's full of things that we never heard about again. I fully expect Tesla to deliver on promises of EAP with ULC, and Enhanced summons. But, just barely. Maybe an iteration or two of improvements to fix big ticket items before they switch focus to FSD/HW3.

You don't sell cars on past promises. You sell cars based on what you can promise now or in the case of Tesla what they can promise in the next year. :p

But, more seriously I think we have to look at the nature of the buyer of the vehicle. Tesla used to sell to rich people who could afford to simply buy the next greatest thing when the current thing didn't deliver on promises. They might huff and puff, but their bank balances were big enough to limit the impact to them. Plus a lot of them made bank off Tesla stock. At times it felt like it was part of some big machinery that ran on the vapers of unmet promises.

That's not really the case anymore. The newer buyers are a lot more careful with their money. You can see that in the polls that show more people getting AP for $2K because it was a deliverable that was ready to go. They were still hesitant on the whole FSD even when told they'd get HW3 for free.

Tesla very much caters to the nature of the kind of person that buys the car. I imagine I'm like a lot of Tesla owners who are attracted to the SW upgradability of the vehicle. It's most definitely a double edged sword that is both the best attribute, and the worst attribute. One of these days this week or next week I'll wake up to a car that is 5% more powerful than I had the day before. That's pretty neat I say. Everyone makes fun of things like fart Easter egg that wasn't an Easter egg. It was silly yet it's part of what makes owning the car fun.

If one isn't completely locked into the whole FSD fiasco there is a lot of fun elements about the car itself. When focused too much on unmeetable FSD promises it's a bit depressing. They used the promise of Software upgrades as a way to sell a farce.

I focus on L3 because it's a tangible thing that I know the competition will offer. For myself it's what matters the most. If EV competitors have L3 self driving and some resemblance of a Supercharging network then I doubt I'll stick around in whatever goofiness Tesla is promising. There is radical change coming. That change is competition. Competition allows people to move to things that more exactly fit their needs.

As to Tesla promise of city driving that I call goofy? That would be neat if it works. I'll probably buy my mom a Tesla if they somehow pull that off, and it meets my expectations. I love the idea of putting my mom in one simply because it could simultaneously torture her, and give me peace of mind.

I put a lot of faith in Karpathy because vision is Tesla's biggest problem. I might gripe about NoA, and it's driving policy/mapping issues, but nothing is worse than the stupid car not seeing an obvious thing in the road.

A lot of what I want is largely under the hood.

The number of objects it can detect
The accuracy of the detections
The working resolution of the neural network
The speed of the neural network inference

I don't expect other people to look at those things as they aren't the big ticket promises.

I'll be disappointed in Karpathy if he doesn't significantly improve the Vision capabilities of the system with HW3.
 
Last edited:
One of these days this week or next week I'll wake up to a car that is 5% more powerful than I had the day before.

Sometimes you also wake up with a car that is 10% less powerful than it was the day before. That’s less neat.

Pack Performance and Launch Mode Limits

And let’s not even talk about V9 on Model S/X...

You don't sell cars on past promises.

You can lose sales based on missed past promises though, but I do agree Tesla probably is just as cynical about this as you are, because they probably feel they can afford to leave some carnage behind.

All that is neither here nor there for me though. What matters is a realistic portayal of Tesla for great conversation and best future decision-making. That’s my goal for participating and following Tesla the company.

As I said, the car itself in many ways is good and there are reasons to like it. Less so with the company.
 
Last edited:
  • Like
Reactions: kavyboy
No one can reliably predict when these things will get done. It's not just Tesla and SpaceX — all kinds of companies whizz past the original timeline for a product without launching.

Only Tesla puts up a product for sale, and accepts payment for it, before the product has been developed, and before it is clear even that the product being promised is possible, much less reliably predict when it will get done.

That is the problem here. Shooting for the moon is not a problem. Promising the moon, putting out a demo video suggesting you've actually already reached the moon, and then taking orders and cashing checks for it, when in fact you have only gotten as far as identifying that there is a large object, primarily mineral in nature, orbiting the planet -- this is the problem.

The “safer” option is always not to try. We want people to try, don’t we?

Yup, we want people to try. But we don't want people to lie or misrepresent the current development status in order to convince people to buy a non-existent product. As it happens, we have laws against that in fact.
 
Only Tesla puts up a product for sale, and accepts payment for it, before the product has been developed, and before it is clear even that the product being promised is possible, much less reliably predict when it will get done.

That is the problem here. Shooting for the moon is not a problem. Promising the moon, putting out a demo video suggesting you've actually already reached the moon, and then taking orders and cashing checks for it, when in fact you have only gotten as far as identifying that there is a large object, primarily mineral in nature, orbiting the planet -- this is the problem.



Yup, we want people to try. But we don't want people to lie or misrepresent the current development status in order to convince people to buy a non-existent product. As it happens, we have laws against that in fact.

Agree that it could have been clearer.
However, Tesla did give a timeline and did deliver on EAP. Tesla did warn that FSD was an unknown timeline and feature set, as opposed to 'available soon'. They have improved in that the new wording provides the timeline vs feature set (along with being 3 years further in development).


Edit: Apparently I fail extremely at reading retention: EAP was called out to have features that still aren't working so great.
 
Last edited: