Waymo founder John Krafcik: Tesla’s Full Self-Driving has ‘bad case of myopia’

Meanwhile Elmo just bribes local officials to change the rules so FSD can be considered autonomous. Texas, truly a one star state.

I recommend clicking the link and reading the article. It’s not a long read and it’s basic, but it’s actually a gem.

That turn of phrase is an insta-classic! :)

Does it matter to the main goal? Making people “rich”. There’s an endless supply of naive people who are smitten by that fraudulent system. It’s fascinating.

You can also listen to the section of the interview with Krafcik the article seems to have been based on [here](https://youtu.be/Dp4MVp2jA7w?si=TLNIEAF4Ot4oCIPD&t=539): it should jump to just before 9 mins in.

A select few, yes. But it is clear to me that, as time passes, and more and more autonomous driving solutions overtake Tesla's FSD and robotaxi offerings, confidence will erode. There will never be dividends, and at some point the share price is bound to collapse, eclipsing Enron by a factor of about twenty, doubtless making a few hapless "all in" Tesla fanbois very poor indeed.

Elon Musk himself is a bad case of myopia. Any and everything he does is to either pump his stocks for the quarter or chase the current trend to fill his deep desire to be liked by everyone.

I doubt it will last much longer. His AV projections used to be in terms of years, switching to months in the middle of 2025 and finally weeks towards the end. How many quarters of awful results, combined with 'safety monitors' still being required in his Austin robotaxis, before this house of cards collapses? One? Two? Does anyone seriously think it can carry on longer than that?

Both Tesla and Elon's valuation is literally at the highest it's ever been. Elon is about to become the first trillionaire in human history. His fraud and constant lies, instead of catching up to him, appear to be strengthening his wealth and power status. I have no hope for people.

Don't you think that, much like Trump, he's jumped the shark, though?

Has even Trump jumped the shark? No matter what he does, his cult will always stay with him. He still has around 40% approval in the US. He has never been more powerful, richer, and stupider at the same time. I'm sorry, I've just lost faith in humanity. I hope you're right and these men at some point will fall into a black hole of karma, but I doubt at this point.

I love that one of the fundamental flaws of elons argument is that it fails to take into account for our proprioception. Like even though we don't talk about it there is a sensation that is very noticeable if you're a driver. I drive an Uber and there's a sixth sense when someone's in my blind spot. Not using lidar was the nail in Teslas coffin

I am listening to Trump addressing the world as I type. Just to pick one example, he stated that he couldn't find any wind farms in China, implying they are useless, China knows that, doesn't use them, just sells them to other countries, despite it being universally acknowledged that China generates more power from wind energy than any other country on the planet. He just stated, as fact, in front of an international audience, that it is universally acknowledged that the 2020 election was rigged. Apparently everyone agrees. Well I don't, I doubt you do, nor does anyone else in the room with him in Davos. And I could go on and on; I'm half expecting him to talk about reducing prices by hundreds of percent again! On a personal note, living as I do in a town in the UK that has a great number of veterans, many who lost limbs, when they responded to the invocation of Article 5, to aid the USA, each time he says NATO never did anything for the US, my blood boils a little more. I've had enough. The EU has had enough. I expect the Supreme Court will decide it is time to draw a line on the tariffs, and, God-willing, Congress will realise they have to rein him in. After what he is saying now, I expect the US markets will open lower again today. Money talks. EDIT: He is now rambling about Biden, saying "what he did to this country"; does he know he is in Switzerland? Then he read his script, which said he had reduced drug prices by up to 90%, but couldn't help ad libbing "or up to 500%, depending on how you calculate it". He started late, he's running over, talking gibberish, insulting the world as he does so. He's jumped the shark. It is all downhill for him from here.

“‘Human vision is so much more capable than the vision of a car equipped with seven 5-megapixel cameras, only one of which is narrow-view, while all the others are wide-view. So you’re basically dispersing those 5 megapixels in a way that makes the actual effective vision more like 20/60 or 20/70. The rest of the cameras in a car like that wouldn’t even pass a DMV vision test.’ He argued that by removing radar and ultrasonic sensors (USS) years ago, and refusing to adopt LiDAR, Tesla has “handcuffed its AI” to a data stream that is inherently noisier and less reliable than what competitors like Waymo or Zoox are using. Krafcic added: ‘And then you have LiDAR and radar providing completely different modalities of active sensing to complement the passive sensing from the cameras. That’s truly superhuman. With that level of data, you can do amazing things, as opposed to a car that has a really bad case of myopia, should be wearing glasses, and operates on a very limited data cycle.’”

Won’t believe it until I see him hit rock bottom of that hill

Now all we need to do is feed ketamine to tesla cars and they can see clearly

The John Johnston Lounge take on the same topic, covering a few other Tesla specifically, and Musk in general, tidbits also: [Waymo Founder: Tesla’s FSD Wouldn’t Pass A DMV Vision Test](https://www.youtube.com/watch?v=NyyofUG7B6M)

While I generally agree with the Waymo founder the fact that he is the Waymo founder makes him incentivized to back Waymos approach over Teslas approach. We as consumers need at least three successful companies otherwise this will be priced at the replacement cost of a Gina driver and not at the cost of the service. Having only one successful company would be a disaster.

It's a wonder a tech guy like Elon would believe mimicking human vision with cameras is so effective. Using that logic, boats would just have a camera at the top of the mast like a dude in a crow's nest instead of using radar.

The phrase jump the shark has jumped the shark. We live in weird times no one seems to go away anymore for being a bad person or Businessman or but they just have ups and downs but they get entrenched in society. Elon right now has so much money he’ll never go away so you’ll just have to live with him.

Yep, whole state is an L. One Star.

We all watch in horror and wonder how it’s possible. But everyone is waiting for some adult in the room to do something, and no one is doing anything. Instead the grift continues, and the crazy talk gets crazier. We’re still wondering how much is too much and it’s TBD if a critical mass will finally say “enough” or if the proverbial lemmings all jump into the ocean. Meanwhile he’s busy removing more and more guardrails to make stopping him more difficult.

Exactly. And to be fair, for what it is, it is pretty good. But when you only have one stream of data you need to develop your algorithms for, you can advance pretty fast... till you hit that brick wall. As elegantly said, they are handcuffed. Everyone else is doing a steady march to a working product and Tesla is still in an accident every 40,000 miles.

>It's a wonder a tech guy like Elon would believe...  He doesn't believe it will ever work any more than you or I do. He's in the business of selling dreams - not delivering them.

And elektrek is reporting Musk's bullshit. That's how you know they're in the losing side, the site and its owner have been musk simps for years.

A lot of people do not realize that a company has zero value (from an investment perspective) if it does not return all the money investors gave it plus some level of interest for the risk they take. That money has to come back in the form of dividends or buybacks of some sort. Right now it is just a company that is changing money between investors less the billions that they keep paying out in compensation packages. There has been zero return to investors to date and at the valuation it is at, there is near zero chance of that ever happening. ('happening' to be of value to those holding at this price that is)

It actually is not as good as you think. Tesla has been a dog since 2022. If you bought after that date it just just been ups and downs with no real gains. Valuation way out to lunch that whole time but no money to be made unless you time the market which is just saying someone else is taking a loss.

5 megapixel? Fucking hell, is he still living in 2005 or something? Any smartphone camera has many times that, how much depending on the model.

TESTIFY, MY BROTHER IN COLUMB!

I recall hearing that Tesla was using structured light with the robotaxi. That could potentially negate the downsides described in the article. But upon a little more [research](https://www.notateslaapp.com/news/2653/is-teslas-cybercab-using-structured-light-what-it-is-and-potential-uses) it seems that this is not true.

One of the biggest lies Musk perpetuates is the idea that he's actually an engineer and that he has some kind of strong technical background. He might be a bit better about it with SpaceX and rocketry since he actually spent some amount of time reading into it but pretty much every interview he's given that goes into any kind of remotely technical detail in other topics usually results in him just saying nonsense or making statements that mean of nothing. Odds are this has been pointed out to him multiple times by multiple engineers and Musk has simply stated they need to find some way to make it work and if they can't he'll hire someone smarter who can. Etc.

Automotive cameras. Also limited by computer vision hardware. HW3 was 1.2 megapixel, HW4 is 5.

Yep, been saying this for years. It's not just the quantity of the data that matters, it's the quality too. Having a shitload of lower quality data of people doing their commute down a highway doesn't help solve difficult coordination and partial information problems in a city environment for example. It's also why the whole 'real world data is all that matters' argument they were making years ago (I think 2019ish?) was complete bullshit. Tesla was saying simulation was a lost cause back then and anyone in the field with half a brain knew you would need to do it because there's just too many high impact low probability events out there that need to be accounted for but couldn't reliably be farmed from data. Everything about Tesla's system is an estimator piled on top of an estimator with some data smoothing to make it look prettier in the visualizations when they want to show it off in some tech demo (the stuff in vehicle is literally prerendered).

The story on this, which is understood by many, but rarely told, is this: Years ago, Musk decided that he would manufacture reasonably affordable cars that would become self-driving very soon - his tech people just needed to figure out how. But he was faced with a choice: he could have lidar and radar OR he could have affordable, but not both. He chose "affordable" cars with camera only, betting that camera only would be enough, and some day Tesla would get to level 4 self driving. Now it seems like Tesla has two options (1) Stick with camera only and hope your engineers figure it out but perhaps never get to level 4, and never launch its "robotaxis" and hope that the investor market keeps the faith/hope, regardless of how many years it takes or (2) Now that it is close to affordable, add lidar and radar to all future Teslas, and re-do your FSD neural net program, and admit that all previously sold Tesla will never reach level 4 and face the legal consequences. Option 2 is basically a non-starter as long as Elon remains the CEO, right? I would note that if Tesla does stick to camera only, and tries to launch robotaxis, there will come a day when someone dies or is seriously injured. And when that happens, their lawyers will need to explain why they chose not to add lidar/radar, and the explanation cannot be "to increase corporate profits".

That reminds me of one of my favorite things to hate that engineering managers say when something goes wrong. "How do we make sure this doesn't happen again?" I'm always thinking, "Who's 'we', jackass?"

I just dislike spreading the idea that data is locked to a sensor or sensor type. It is a false narrative. Both Tesla and Waymo are guilty of playing into it.

"*Tesla's self-driving cameras have significantly upgraded with* [*Hardware 4 (HW4)*](https://www.google.com/search?q=Hardware+4+%28HW4%29&rlz=1C1GEWG_enUS1001US1014&oq=how+many+megapixels+is+tesla+self+driving+cameras%3F&gs_lcrp=EgZjaHJvbWUyBggAEEUYOTIHCAEQIRigATIHCAIQIRigATIHCAMQIRigATIHCAQQIRirAjIHCAUQIRiPAjIHCAYQIRiPAtIBCTExNDQ2ajBqN6gCALACAA&sourceid=chrome&ie=UTF-8&mstk=AUtExfCkm0hcM5PE3f0zoRZ-7HX44dEC3R1jpSO0OTAcJnm67pNIGswwekF7l6a9csLr0fFz_LnKjMd6sW0NqJAWGKaRbvxdxAAMUWX3V-Kv6XVUDgnCV--QnrCnGiVzqk9mok1perYMypMvoRNGPQxD9jGpGx8WmEawOpxS6FMp3uYolgo&csui=3&ved=2ahUKEwiwnr-8o52SAxUhODQIHWGwJ54QgK4QegQIARAD)*, moving from 1.2 megapixels in HW3 to* ***5-megapixel*** ***resolution****, using Sony's IMX963 sensor for better clarity, dynamic range, and low-light performance, which greatly aids in object detection for Full Self-Driving (FSD).*  ***Key Details:*** * ***Hardware 3 (HW3):*** *Used 1.2-megapixel cameras.* * ***Hardware 4 (HW4):*** *Upgraded to 5-megapixel cameras (Sony IMX963 sensor).* * ***Recent Changes:*** *In early 2023, Tesla began transitioning to these 5MP cameras and added an extra forward-facing bumper camera for FSD.* " Im not sure why they use just 5 megapixel cameras, but it may have made sense in 2022 when plans were made. I am also not sure if jumping to 40mp solves the underlying issue. It is better vision and a lot more data, but all from the same modality. It will be an improvement but maybe not as much as you would expect.

Except in downtown Austin there are more Waymos than there are teslas. Not just tesla robo taxis, all teslas. If tesla robo taxis are marked then I didn’t see any the weekend I was there recently.

To be fair with less and bigger pixels it would see MUCH better at night. But they just picked the shittiest CMOS.

I cannot speak for what is happening in the USA but in Europe there has been a shift. He tried the tariff BS, we humoured him, countries including mine (UK) made deals. Then, within a few months, because a single officer from here, one man, for three days, went to Greenland for a pre-arranged NATO exercise, that the US were invited to attend, it is all out of the window. As a previous POTUS famously misquoted: "Fool me once, shame on you; fool me twice, shame on me". The UK, and especially the EU, will react differently this time. I expect either the Supreme Court will clip his wings, or Trump will develop amnesia with respect to these latest tariffs. The British, the French, and many other European nations have been dealing with people like Trump for centuries. EDIT: And no surprise, within minutes of posting that, [the Dear Leader has announced the tariffs are now off the table](https://truthsocial.com/@realDonaldTrump/posts/115934734335579278). The Europeans stood up to him and he crumbled!

Can’t get fooled again!

I don't know--I think he's worked so hard at fooling others, and do some well at surrounding himself with sycophants, that he's talked himself into believing his own BS.

When he does stuff like direct his staff to fake the "Paint it Black" video or tele-operates robot demos, its pretty clear that he knows exactly what he's doing. He knows this junk doesn't work and never will work.

Oh it gets better. Now an insolvent insurance company wants to insure these drivers. Lemonade makes Tesla insurance look like Berkshire Hathaway.

Tesla does use LIDAR as part of the FSD training and validation; they just don’t use it for inference. Have no idea why the safety monitors are still there in Austin

Good point. He might actually be thinking that it *will* work and that he's just painting a picture of what it will be once some details are ironed out. Or maybe he is abandoning even that.

The safety monitors are still there because there would be far too many accidents if they weren't.

No doubt FSD v14 is good for what it is, but I’d love to see what Tesla could develop with the standard sensor set that most of the Chinese FSD competitors are currently shipping with in China, they basically all use a few more cameras, 5x radars and 1 to 3 lidars. It’s a shame that FSD is hamstrung by HW4, it’s as if it can’t live up to its full potential, especially as sensor fusion isn’t anywhere near as ‘hard’ as people claim it is with current AI models…

That would require Elon admitting that he was wrong and listening to real engineers

Wait, the car cameras are only 5 megapixels each? That’s fewer megapixels than digital cameras 20 years ago

I think aside from the Cybertruck, he hasn’t introduced anything into the assembly and production line. So the sensors might have been specced out by the execs and engineers he fired

Quick lesson here for any younger folk - engineers or whatever - if you make a mistake admit it and say "that one is my fault, I shoulda done X not Y". If someone else makes a mistake always say "how do **we** fix it". Never make it look like you're blaming someone, even if you know they're at fault. Keep the focus on solving the problem.

On the point. Camera sensors simply overflow and deliver incorrect data. It is transfer problem from the real world to the digital world. Every photographer knows that. Elon Musk is idiot and belives kameras wirk Like eyes….

Nice recommendation.

Yeah, it's obvious to anyone who's even a little bit interested in tech. It's like living in a cyberpunk world and given a choice of a bionic eye implant with infrared, night vision and zoom function, you say 'no, i want a bionic eye with worse vision than my original eyeball. And remove my other eye, please. I don't need depth perception. '.

You are also fighting dynamic range compared to an eye. That would make it especially difficult on snowy roads.

There have been 1,429 incidents involving Waymo. Tesla FSD has an average of 700% more miles between collisions than the average driver on the road. I'm confused as to why people can't research data and use basic math.

Apples and Oranges. If that "average driver" had someone sitting next to them, ready to take over if things got tricky, that comparison *might* be meaningful.