-
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
ZenRockGarden
Active Member
I see it more simply: He was in there for 5 years, striving against crazy pressure and deadlines as is the Tesla Way. His health and family life probably suffered. He decided he had to have a break, took the sabbatical, and upon de-compressing decided that with enough stock options to live on for the rest of his life, staying on the treadmill for more years wasn't worth it. Let someone else take a turn.
I am kind of surprised Andrej lasted that long. It is big jump from academia to Tesla-get it done yesterday attitude.
Last edited:
No doubt that Elon is brilliant, but most people here are correct that he cannot create a hostile work environment for too long without good, talented people leaving. Look at what happened to Apple when Steve passed away. Some say Apple still isn't as good a company under Tim's leadership. It later came out that Steve created a hostile work environment as well, but the end result was still an excellent set of products and services.
Hopefully Tesla will adapt to his departure and new talent will come in. But it may take time, like it did for Apple.
Hopefully Tesla will adapt to his departure and new talent will come in. But it may take time, like it did for Apple.
Or…got tired of Elon constantly tweeting FSD lies about what it can do and when…
When reality actually was it hitting curbs/green poles/jersey walls and almost hitting school school buses
Tesla sounds like every big tech company I have worked for. The free lunch and dinner are nice, but let's not kid ourselves. The reason the campus has washers and dryers, gyms, childcare, food 24/7, baristas, and private offices is so that they can keep you working. I knew plenty of guys who did not leave the complex for 7 or 8 days. Welcome to Silicon Valley, Silicon Slopes, or Austin, or ... Gilded Cages.
Or as Edison said, ‘Genius is one per cent inspiration, ninety-nine per cent perspiration’.
Last edited:
worth pointing out that Karpathy was the leader, so would be setting those deadlines and also responsible for managing those supervisors.
Sure, he was reporting to EM, but it was his team to manage
It indeed is. Also - there will probably no official "launch" of FSD. At best, it probably will be a soft launch following years and years of staying in beta.
When all the dust settles every one is replaceable except maybe Musk.
BitJam
Active Member
It's unclear to me what the delay was between Andrej's contributions and implementation in the fleet. They're still working on one stack to rule them all even though it was announced in the fall of last year.
A recent paper Karpathy wrote: Deep Neural Nets: 33 years ago and 33 years from now says the performance increases we see in neural networks scale with improved hardware:
First of all, not much has changed in 33 years on the macro level. We’re still setting up differentiable neural net architectures made of layers of neurons and optimizing them end-to-end with backpropagation and stochastic gradient descent. Everything reads remarkably familiar, except it is smaller.
If Dojo comes on-line (or is on-line already) and if the autolabeler is making some manual labeling jobs redundant then Andrej leaving could have minimal impact on FSD development for the rest of this year and perhaps for a few years to come. It's quite possible that Karpathy's great talents are no longer needed at Tesla and he is looking for a position that will really challenge his abilities. Would someone on the level of Karpathy be needed for migrating to HW4 or would this be simply turning the crank (as implied in paper)?
I'd love to hear what James Douma has to say. Maybe Dave Lee will have another interview with him soon.
A recent paper Karpathy wrote: Deep Neural Nets: 33 years ago and 33 years from now says the performance increases we see in neural networks scale with improved hardware:
First of all, not much has changed in 33 years on the macro level. We’re still setting up differentiable neural net architectures made of layers of neurons and optimizing them end-to-end with backpropagation and stochastic gradient descent. Everything reads remarkably familiar, except it is smaller.
If Dojo comes on-line (or is on-line already) and if the autolabeler is making some manual labeling jobs redundant then Andrej leaving could have minimal impact on FSD development for the rest of this year and perhaps for a few years to come. It's quite possible that Karpathy's great talents are no longer needed at Tesla and he is looking for a position that will really challenge his abilities. Would someone on the level of Karpathy be needed for migrating to HW4 or would this be simply turning the crank (as implied in paper)?
I'd love to hear what James Douma has to say. Maybe Dave Lee will have another interview with him soon.
jebinc
Well-Known Member
I think he left because he knows FSD advancement has hit a wall with current hardware and tech, so he got out before Elon’s false statements tarnish his reputation.
I would think it was an interesting research problem of whether vision can solve FSD and designing the neural network architectures to support it. Once that's basically solved, it's more of an engineering and data curation problem, but maybe those aspects are not Karpathy's primary interests.
Since AI Day last year, the team has been trying to move more predictions to the latest architecture, and that's more "straightforward" (although not simple) in applying, collecting, training, testing and optimizing and not so much researching.
Pricedm
Member
Dojo supercomputer...crickets. Recent elimination of several hundred jobs on "labeling" team. Reintroduction of Enhanced Autopilot. Not positive signs for FSD progression. Karpathy contributed a significant vision to the team. Will be interesting to see how FSD evolves. Personally, seems like better hardware--corner/high mounted cameras for cross traffic are necessary.
I hope the FSD team gets a new leader and progress continues. But I have doubts-- not just Tesla's vision approach. City driving by a "computer on wheels" is a very difficult situation to solve.
I hope the FSD team gets a new leader and progress continues. But I have doubts-- not just Tesla's vision approach. City driving by a "computer on wheels" is a very difficult situation to solve.
jebinc
Well-Known Member
jabloomf1230
Minister of Silly Walks
People change jobs all the time. Karpathy always impressed me as an academic type who pined to be a big shot Professor at either Cal Tech or Stanford.
smogne41
Active Member
Honestly I think Tesla would do better without Musk now.
Similar threads
