I've had the impression it's pure chaos over at Tesla's autopilot division for quite a long time. They have so much to gain if they get the pieces right, but I'm not sure it's guaranteed to happen.
> I've had the impression it's pure chaos over at Tesla's autopilot division for quite a long time.
That's interesting. I've heard this too from people at three different self-driving car companies ("Tesla's autonomy group is melting down; we get a ton of resumes...blah blah").
But...it's from their competitors, and you're probably always getting resumes from folks at your competitors all the time (and your colleagues' resumes are heading in the other direction, perhaps yours too). So, though I'm no fan of Tesla, I have discounted such comments. Though perhaps I shouldn't have been?
I just saw their latest marketing literature on their website. They promise:
> All Tesla vehicles produced in our factory, including Model 3, have the hardware needed for full self-driving capability at a safety level substantially greater than that of a human driver.
It seems somewhat irresponsible to promise that you have enough hardware to achieve something that nobody has ever achieved before, with any hardware.
Tesla has plenty of impressive-looking videos on their website of autopilot working well, but if you Google for "Tesla autopilot fail" you can also find plenty of stuff like this, which is just surprisingly bad: https://www.youtube.com/watch?v=UZ1XLqc5IUg
Tesla strikes me as over-promising and under-delivering here. I wouldn't want to work in a division like that.
They say "at a safety level substantially greater than that of a human driver". This is a low bar to meet - some human drivers are really bad. I know I would much rather trust my life to one of the Google self-driving cars than some of the people I got in a car with in college, for example.
It's also the wrong bar to meet, because lawmakers are going to be much tougher on AIs than on human constituents. It's not enough to be better than a human, you have to be so much better than a human that it'd be irresponsible not to mandate self-driving cars.
The bar to meet will be precisely "as good as Waymo", because if Tesla's technology kills substantially more people, neither regulators nor consumers will allow it to exist.
Human drivers die at a rate of about 1 per 100 million miles driven. Waymo cars have only driven about 2 million miles, as I recall. Therefore, as it stands today, there is nowhere near enough public evidence to confidently believe that Waymo cars are less deadly than human drivers. Especially considering that much of that driving has taken place in meticulously mapped areas with good roads in daylight hours.
Personally, I would trust a human driver over a Waymo car until strong statistical evidence shows that Waymo cars are safe in a wide range of conditions.
I also don't think the bar should be "safer than a human driver", but "safer than than an experiended, sober, well-rested, defensive driver who doesn't use this phone while driving".
> Why should the bar be any higher than the median actual human driver?
Marketing. The western world is mostly afraid of robots so they'll need to be at least 10x safer in a hybrid (human/autopilot) environment in order to gain adoption.
This would be FAR, FAR more believable if the guy placed both his hands in view. You can't see what the right hand is doing. I remain skeptical. Does he show his hands in the rest of the video? I stopped watching after a few minutes of not seeing both of them up in the image.
What do you think his right hand might be doing? I'm not familiar with the Tesla interior setup so I don't know what is possible, but I'm having a hard time imagining what he could be possibly be doing off-camera that wouldn't be reflected in any of the visible displays.
How about steering the car with his right hand on the bottom of the wheel to make it look like there's a problem with the auto pilot?
If someone wants to claim there's a problem with any auto pilot they better have a camera view showing they are not interfering with it in any way.
The idea that one can do things that are not seen on the various reflective surfaces isn't a new concept, television shows rely on camera and object placement all the time to ensure the camera and crew are not visible.
If I were to make a claim about the auto pilot I would have both my hands in plain view in a "look ma, no hands" stance right above the steering wheel for all to see. I would also have a second camera shooting from the side or farther back in order to make it visible that nothing is interfering with the steering, like my leg for example or someone pulling on strings from the back seat.
I don't have a problem with the idea of issues with any auto pilot software. I simply want to make sure claims are not falsified through the use of trickery.
To add on to this, how can they reconcile their /autopilot link with their car configurator? the /autopilot link implies all cars have the hardware to do full autopilot - but the fine print seems to be "if you add on $10k of options".
I wouldn't want people checking in code that doesn't conform to the coding practices of the rest of the project, even if the code came from Ken Thompson himself. If the coding practices are bad, his experience will hopefully allow him to convincingly argue this point. In any case, the company employs many other smart people too, and since the coding practices are there to facilitate collaboration, nobody should be above them.
I doubt this anecdote. There was no such test in 2011, and I haven't heard any tales from older timers about it either. It sounds like it's conflating some totally optional style-guide-conformance procedures you can do, with permission to check in code at all. And no-one gets to check in code without review regardless of how much expertise you've demonstrated.
I'm not sure why you doubt the anecdote. It's a published quote in a published book by a respected author, and to my knowledge Thompson hasn't refuted the statement. The original quote doesn't have the trashy editorialization. Thompson simply had no need to go through the approval process so he didn't.
When this was published, it was widely circulated and no one at the time stepped forward to claim it was untrue or that the requirement didn't exist. Given the fact that you make reference to "old timers" in this context, I doubt you were at Google in 2011 and are merely asserting your assumption as a fact.
IBM has people working on server-side Swift, but (as I understand it) those people are mainly focused on ways to make the IBM cloud service more enticing.
Getting a job is like getting into a nightclub. If you're a celebrity, the bouncers open the velvet rope and usher you straight into the club, leaving the long queue of punters waiting to be told that their shoes aren't right, or that its a private party.
I get what you're saying and agree it would be nice, but so long as we're wishing for stuff, I wish we could all, even those of us who are less successful, be able to declare our failures without it affecting our future job prospects.
I'm entirely unremarkable, but looking back I would say most of my biggest personal and professional developments have been built upon some sort of failure or mistake.
The most important men in town will come to fawn on me
They will ask me to advise them,
Like a Solomon the Wise
"If you please, Reb Tevye?"
"Pardon me, Reb Tevye?"
Posing problems that would cross a rabbi's eyes
Ya va voy, ya va voy voy vum
And it won't make one bit of difference
If I answer right or wrong
When you're succesfull they think you really know.
This has been my experience. I'll stick it out for 6mos, but if there is too much hazing, bro-culture, weird expectations, the power dynamics within teams are upside down (usually a hierarchy with project managers at the top and engineers at the bottom), no institutional memory, no 360's after projects, some weird hybrid of agile/kanban where you literally standup while the client shuffles priorities and deadlines daily and no sprint planning can effectively happen, everything is always an emergency, and there is little to no transparency then the bar is very very low to get the hell out and take my raise at a new company.
If nobody recognized his name then it wouldn't matter because nobody would know what he had done. Just ask the guy who wrote homebrew (Max Howell. I had to look up his name).
All he needs to do is put "I wrote homebrew" in his resume; and it'll matter.
(It is a wee more difficult to write LLVM than homebrew, though - so it might not have exactly the same impact. But I'm guessing that e.g. Apple would still hire him, if he wants to)
> All he needs to do is put "I wrote homebrew" in his resume; and it'll matter.
He did put it on his resume and it didn't matter one bit. Lattner got offers from Google without having to do the interviews. Max was subjected to the interviews and rejected. The lack of marketing and the fact nobody knows his name is a huge factor.
I wonder what's the connection between Lattner and autopilot. When Tesla hired Lattner, my impression was that he would oversee the entire systems software architecture. Why they couldn't have retained Lattner and also hired Karpathy is beyond me.
There does happen to be a research engineer position open at Mozilla right now, but it's for Servo, not Rust. To say nothing of the irony of having Chris Lattner working on Rust while Graydon Hoare works on Swift. :P But hey, Rust could undoubtedly benefit from a full-time LLVM engineer...
From one the presentations of this year's WWDC, I had the impression that there may be some of Rust's ownership concepts incorporated in future versions of Swift. It would definitely be very interesting.
>I don't get why he is leaving the company that soon
Because he doesn't like it there.
>nor what it announces it on Twitter before getting a new job.
Because it's his Twitter and he has nothing much to hide. It's not like he wont find employment (actually it might be not like he even needs employment that badly -- he could easily have a couple of million in the bank).
You really think they'd fire a showy public hire this fast? Unless he was committing felonies it's almost impossible. They'd be working with him to try to salvage the situation.
Impetuous quitting over some specific issue he holds dear is only a thousand times more likely.
acting or done quickly and without thought or care.
"her friend was headstrong and impetuous"
synonyms: impulsive, rash, hasty, overhasty, reckless, heedless, careless, foolhardy, bullheaded, headstrong, incautious, imprudent, injudicious, ill-considered, unthought-out
So he's got a good reason to leave but the decision to leave is rash, foolhardy, reckless? These don't seem to fit together. If you leave for a legitimate reason, I'd hardly call that an impetuous decision.
If you work in a good place with one huge flaw, and management assures you they will address the flaw, then you blow up one day because they haven't addressed it yet and quit, that's an impetuous decision for a good reason.
It's not necessarily rash, foolhardy or reckless, since those are synonyms. In this case they are similar, but not exactly the same.
Yea i'm wrong. Didn't realize it was 5 months since he started, was thinking it was more like two.
Still strange that a guy so super successful inside and out of Apple couldn't fit there. My guess is he was a bad fit for the role, Tesla said how bout we move you to another role that fits you better, and Chris said "no".
What's strange to me is why he would think it was a good fit in the first place.
His main work has been in languages, compilers and developer toolsets. Sure, Tesla makes use of those, and might even make a few of their own, but it's nowhere near their core interests.
VP Autopilot Software
January 30 - June 20, 2017
When I joined Tesla, it was in the midst of a hardware transition from "Hardware 1" Autopilot (based primarily on MobileEye for vision processing) to "Hardware 2", which uses an in-house designed TeslaVision stack. The team was facing many tough challenges given the nature of the transition. My primary contributions over these fast five months were:
We evolved Autopilot for HW2 from its first early release (which had few features and was limited to 45mph on highways) to effectively parity with HW1, and surpassing it in some ways (e.g. silky smooth control).
This required building and shipping numerous features for HW2, including: support for local roads, Parallel Autopark, High Speed Autosteer, Summon, Lane Departure Warning, Automatic Lane Change, Low Speed AEB, Full Speed Autosteer, Pedal Misapplication Mitigation, Auto High Beams, Side Collision Avoidance, Full Speed AEB, Perpendicular Parking, and 'silky smooth' performance.
This was done by shipping a total of 7 major feature releases, as well as numerous minor releases to support factory, service, and other narrow markets.
One of Tesla's huge advantages in the autonomous driving space is that it has tens of thousands of cars already on the road. We built infrastructure to take advantage of this, allowing the collection of image and video data from this fleet, as well as building big data infrastructure in the cloud to process and use it.
I defined and drove the feature roadmap, drove the technical architecture for future features, and managed the implementation for the next exciting features to come.
I advocated for and drove a major rewrite of the deep net architecture in the vision stack, leading to significantly better precision, recall, and inference performance.
I ended up growing the Autopilot Software team by over 50%. I personally interviewed most of the accepted candidates.
I made massive improvements to internal infrastructure and processes that I cannot go into detail about.
I was closely involved with others in the broader Autopilot program, including future hardware support, legal, homologation, regulatory, marketing, etc."
The original version apparently ended with the text "In the end, Elon and I agreed that he and I did not work well together and that I should leave, so I did."
83 comments
[ 2.9 ms ] story [ 164 ms ] threadThat's interesting. I've heard this too from people at three different self-driving car companies ("Tesla's autonomy group is melting down; we get a ton of resumes...blah blah").
But...it's from their competitors, and you're probably always getting resumes from folks at your competitors all the time (and your colleagues' resumes are heading in the other direction, perhaps yours too). So, though I'm no fan of Tesla, I have discounted such comments. Though perhaps I shouldn't have been?
> All Tesla vehicles produced in our factory, including Model 3, have the hardware needed for full self-driving capability at a safety level substantially greater than that of a human driver.
https://www.tesla.com/autopilot
It seems somewhat irresponsible to promise that you have enough hardware to achieve something that nobody has ever achieved before, with any hardware.
Tesla has plenty of impressive-looking videos on their website of autopilot working well, but if you Google for "Tesla autopilot fail" you can also find plenty of stuff like this, which is just surprisingly bad: https://www.youtube.com/watch?v=UZ1XLqc5IUg
Tesla strikes me as over-promising and under-delivering here. I wouldn't want to work in a division like that.
It's also the wrong bar to meet, because lawmakers are going to be much tougher on AIs than on human constituents. It's not enough to be better than a human, you have to be so much better than a human that it'd be irresponsible not to mandate self-driving cars.
Personally, I would trust a human driver over a Waymo car until strong statistical evidence shows that Waymo cars are safe in a wide range of conditions.
Adding vehicles operating at a higher skill level than the median to the pool of vehicles driving around should reduce incidents per mile driven.
And then if autonomous vehicles become prevalent, it becomes less punitive to raise the standards for licensing humans.
Marketing. The western world is mostly afraid of robots so they'll need to be at least 10x safer in a hybrid (human/autopilot) environment in order to gain adoption.
If someone wants to claim there's a problem with any auto pilot they better have a camera view showing they are not interfering with it in any way.
The idea that one can do things that are not seen on the various reflective surfaces isn't a new concept, television shows rely on camera and object placement all the time to ensure the camera and crew are not visible.
If I were to make a claim about the auto pilot I would have both my hands in plain view in a "look ma, no hands" stance right above the steering wheel for all to see. I would also have a second camera shooting from the side or farther back in order to make it visible that nothing is interfering with the steering, like my leg for example or someone pulling on strings from the back seat.
I don't have a problem with the idea of issues with any auto pilot software. I simply want to make sure claims are not falsified through the use of trickery.
https://www.tesla.com/modelx/design
"Enhanced Autopilot" as a $6,000 option. "Full Self-Driving Capability" is another $4,000 option.
/autopilot page seems very disingenuous.
https://techcrunch.com/2017/06/20/tesla-hires-deep-learning-...
https://go.googlesource.com/gollvm/
Unclear to me in what capacity it is "recognized" or "official" though.
https://books.google.com/books?id=2kMIqdfyT8kC&pg=PA474&lpg=...
When this was published, it was widely circulated and no one at the time stepped forward to claim it was untrue or that the requirement didn't exist. Given the fact that you make reference to "old timers" in this context, I doubt you were at Google in 2011 and are merely asserting your assumption as a fact.
I'm entirely unremarkable, but looking back I would say most of my biggest personal and professional developments have been built upon some sort of failure or mistake.
You clearly haven't because you, in error, think it'll nuke job prospects.
It won't. They won't care. Trust me.
And it won't make one bit of difference If I answer right or wrong When you're succesfull they think you really know.
Yeah, just not in the field Chris Lattner works in. This thread seems to have him confused for Kim Kardasian.
If it's not a good fit, why can't it be a failure on the employer side?
(It is a wee more difficult to write LLVM than homebrew, though - so it might not have exactly the same impact. But I'm guessing that e.g. Apple would still hire him, if he wants to)
He did put it on his resume and it didn't matter one bit. Lattner got offers from Google without having to do the interviews. Max was subjected to the interviews and rejected. The lack of marketing and the fact nobody knows his name is a huge factor.
Because he doesn't like it there.
>nor what it announces it on Twitter before getting a new job.
Because it's his Twitter and he has nothing much to hide. It's not like he wont find employment (actually it might be not like he even needs employment that badly -- he could easily have a couple of million in the bank).
More likely the other way round.
Impetuous quitting over some specific issue he holds dear is only a thousand times more likely.
acting or done quickly and without thought or care. "her friend was headstrong and impetuous" synonyms: impulsive, rash, hasty, overhasty, reckless, heedless, careless, foolhardy, bullheaded, headstrong, incautious, imprudent, injudicious, ill-considered, unthought-out
If you reread my comment, it didn't say he didn't have a good reason, just that he might have acted quickly or emotionally. Like many of us do.
It's not necessarily rash, foolhardy or reckless, since those are synonyms. In this case they are similar, but not exactly the same.
Still strange that a guy so super successful inside and out of Apple couldn't fit there. My guess is he was a bad fit for the role, Tesla said how bout we move you to another role that fits you better, and Chris said "no".
His main work has been in languages, compilers and developer toolsets. Sure, Tesla makes use of those, and might even make a few of their own, but it's nowhere near their core interests.
Google maybe?
https://news.realm.io/news/swift-on-android/
http://nondot.org/sabre/Resume.html#workhistory
"Tesla
VP Autopilot Software January 30 - June 20, 2017 When I joined Tesla, it was in the midst of a hardware transition from "Hardware 1" Autopilot (based primarily on MobileEye for vision processing) to "Hardware 2", which uses an in-house designed TeslaVision stack. The team was facing many tough challenges given the nature of the transition. My primary contributions over these fast five months were:
We evolved Autopilot for HW2 from its first early release (which had few features and was limited to 45mph on highways) to effectively parity with HW1, and surpassing it in some ways (e.g. silky smooth control). This required building and shipping numerous features for HW2, including: support for local roads, Parallel Autopark, High Speed Autosteer, Summon, Lane Departure Warning, Automatic Lane Change, Low Speed AEB, Full Speed Autosteer, Pedal Misapplication Mitigation, Auto High Beams, Side Collision Avoidance, Full Speed AEB, Perpendicular Parking, and 'silky smooth' performance. This was done by shipping a total of 7 major feature releases, as well as numerous minor releases to support factory, service, and other narrow markets. One of Tesla's huge advantages in the autonomous driving space is that it has tens of thousands of cars already on the road. We built infrastructure to take advantage of this, allowing the collection of image and video data from this fleet, as well as building big data infrastructure in the cloud to process and use it. I defined and drove the feature roadmap, drove the technical architecture for future features, and managed the implementation for the next exciting features to come. I advocated for and drove a major rewrite of the deep net architecture in the vision stack, leading to significantly better precision, recall, and inference performance. I ended up growing the Autopilot Software team by over 50%. I personally interviewed most of the accepted candidates. I made massive improvements to internal infrastructure and processes that I cannot go into detail about. I was closely involved with others in the broader Autopilot program, including future hardware support, legal, homologation, regulatory, marketing, etc."
The original version apparently ended with the text "In the end, Elon and I agreed that he and I did not work well together and that I should leave, so I did."