Let's be real, if there's anyone who can't embrace pragmatism and barks up one tree despite everyone else pushing for the contrary, it's Elon.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Karpathy leaves - what's next for AP/FSD?
- Thread starter sids911
- Start date
mtndrew1
Active Member
There are stretches on the 10, 110, and 14 freeways in Los Angeles where autopilot has been slamming on the brakes for incorrectly-targeted cars in adjacent lanes for YEARS.
At the very least, thousands of HW3 Teslas drive these roads every week. Maybe tens of thousands. And no change all this time.
I understand that Tesla is aiming for a generalized solution to L5 but as a customer I would be thrilled for an end to end solution giving me real L3 on geocached roads.
I’m starting to think GM Super Cruise might be the better path for my needs. I watched videos of people using it, truly hands free, for extended periods of time. Even when towing!
City FSD is a neat party trick, sort of, but after the better part of a year using it I’m realizing I just don’t care. It’s not useful. Real L3 highway driving is useful (for me).
Last edited:
Daniel in SD
(supervised)
Super Cruise isn't L3 though, it still requires driver attention and L3 doesn't.
Full speed divided highway L3 would be awesome but it's probably not happening very soon. Mercedes does plan on releasing traffic jam (<37 mph) L3 system in California soon though.
mtndrew1
Active Member
Yeah I know and I should have made that clear.
1. SuperCruise, today, looks more useful than FSD for my use case.
2. I wish Tesla would aim for real L3 instead of fussing with L5. For me L3 would be a quality of life improvement while FSD, even if Tesla can get it working properly in the city, is just of very little value and a novelty.
These are different than my opinions from a year ago; I’m just not seeing the rate of progress necessary to make FSD anything other than a party trick.
They said they would have FSD safer than a human driver at end of year. The FSD Beta agreement notes
“The currently enabled features require active driver supervision and do not make the vehicle autonomous. The activation and use of these features are dependent on achieving reliability far in excess of human drivers as demonstrated by billions of miles of experience, as well as regulatory approval, which may take longer in some jurisdictions.”
In other words IF it is safer than a human at end of year then it needs to become FAR safer than a human (I think they have already said 10x as a goal) and billions of miles to prove that… then there is the process of getting regulatory approval which will take god knows how long. In other words robotaxis that can drive you anywhere are years & years away at best.
Last edited:
To be fair Steve Jobs was known to be a very harsh leader. He was visionary and had a knack of inspiring people to do things that others considered impossible, which was why people were there, but he was not easy to work with and working under him would have been no 9-5.
I really don’t know much about Tim Cook but would think that the culture is more conventionally corporate and less frantic, but less visionary.
I would agree with all of this. Other thing thing to keep in mind is that pay for good software engineers is very good. Even considering crazy Bay Area cost of living it’s entirely feasible if you live sensibly to work 10 years out of school and be in a position, especially with stock options, that you don’t HAVE to work anymore.
Isn’t Supercruise basically enhanced autopilot but with eye monitoring instead of steering wheel torque? I really hope Tesla starts using the interior camera to remove the steering wheel jerk requirement. If so I would seriously consider the $6k enhanced autopilot. Really not interested in City Streets FSD beta and smart summon for $12k.
Tesla could remove all the attention checks anytime they wanted to - the downside is will regulators jump down their throat? As YouTube has shown us, if you put in a safety monitoring system, someone will attempt to circumvent it and wave at people from the back seat of their driverless car. I wonder how long before someone circumvents the Super Cruise eye monitoring.
Good point. I drove the HOV lane (no physical barriers) on the Long Island Expressway about 60 mi total last weekend and I must have had 20 cars cross the double line to pull into the lane in front of me. I don’t know if HOV lane entry and exit ever gets enforced there.
True. Consumer Reports will experiment and put together a “how to” guide for fooling the camera.
Hands-free towing really scares me. There is a lot that can go irreversibly bad very quickly when towing. I’m pretty surprised GM allows it to be activated.
Not only feasible, relatively common. Especially if you sell your home for millions and move somewhere else where houses are cheap.
If they plan to stay in the Bay Area, they will DEFINITELY have to continue to work after that ten years.
If they move to Montana? Prolly not
jebinc
Well-Known Member
DanCar
Active Member
Elon's FSD expectations have been unrealistic since 2016. I'd like to think that if he had more realistic expectations we wouldn't have so many departures at the top.
There was a known radar chip shortage when Tesla decided to remove it.
Probably will never know why for sure, but being the head of Vision and the phantom breaking that has been going on so long that the NHTSA opened an investigation can't be a positive.
AI day, coming up, might offer some answers to who is taking the reins and what the revised^11 roadmap will look like.
AI day, coming up, might offer some answers to who is taking the reins and what the revised^11 roadmap will look like.
Not necessarily. He mentioned stock options. The first set usually vests in 5 years. I know plenty of people who used that money to start slowing down in the Bay Area. At one company I consulted at it was called "Spring Break" because shared vested in the Spring.
Also, Montana is not so cheap after everyone bugged out to Montana and Idaho. Unfortunately, that drove prices up to unaffordable levels for locals.
Who is Andrej Karpathy? Tesla AI chief suddenly quit and that spells trouble for Elon Musk | Fortune
The oddly timed departure of the brains behind Tesla's self-driving program suggests either the work is done...or the EV company still has a long road ahead in the race for true autonomy.
fortune.com
So much disinformation in this thread and in the links here. He was not the brain behind Tesla Autopilot, he joined to replace Lattner who was a great lead for Software 1.0, but Elon saw that the projected needed a Software 2.0 approach to succeed. Karpathy‘s main contribution was improving how to balance data, how they clean the data, how to manage labelers, how to tune weight function when you run multiple tasks in a neural network and how to work as a big team training one large neural network. Basically he helped setting up the software 2.0 stack correctly.
Now that that is mostly complete, they will grow in complexity and have to deal with the problems of complexity. Two steps forward, one step back times many and in parallel and in serial. They need someone who has good intuition for this. Ashok and Milan seems like a great team for this and some other battle tested guy who has been responsible for validation and choosing what new features to push out will step up and become another leader.
Now that that is mostly complete, they will grow in complexity and have to deal with the problems of complexity. Two steps forward, one step back times many and in parallel and in serial. They need someone who has good intuition for this. Ashok and Milan seems like a great team for this and some other battle tested guy who has been responsible for validation and choosing what new features to push out will step up and become another leader.
Similar threads
