Fun fact for all of you:<p>Some time ago (around ~10 years) this guy (the presenter) was internet famous for being a Rubik cube speed solver and making tutorials and videos about that: <a href="https://www.youtube.com/watch?v=609nhVzg-5Q" rel="nofollow">https://www.youtube.com/watch?v=609nhVzg-5Q</a>
The competition in this space is great but I can't help but wonder what would happen if instead all these companies pooled their resources and went after the goal collectively. There is so much duplication going on and the paths do not seem to me - as an outsider - to be all that divergent, which is usually a pre-condition for having a lot of independent efforts one of which will succeed.<p>It's as if everybody wants to be the one to exclusively own the tech. Imagine every car manufacturer having a completely different take on what a car should be like from a safety perspective. We have standards bodies for a reason and given the fact that there are plenty of lives at stake here maybe for once the monetary angle should get a back-seat (pun intended) to safety and a joint effort is called for. That would also stop people dying because operators of unsafe software are trying to make up for their late entry by 'moving fast and breaking things' where in this case the things are pedestrians, cyclists and other traffic participants who have no share in the monetary gain.
His team is hiring;<p><a href="https://www.tesla.com/careers/job/software-engineerdeeplearning-49779" rel="nofollow">https://www.tesla.com/careers/job/software-engineerdeeplearn...</a><p><a href="https://www.tesla.com/careers/job/machine-learninginfrastructureengineerautopilot-48125" rel="nofollow">https://www.tesla.com/careers/job/machine-learninginfrastruc...</a><p><a href="https://www.tesla.com/careers/job/machine-learningscientistautopilot-48414" rel="nofollow">https://www.tesla.com/careers/job/machine-learningscientista...</a>
Awesome presentation. Crazy that they're developing their own training hardware too. It's going to be a very crowded space very soon. Can they really stay ahead of everyone else in the industry? Can it really be cheaper to staff up whole teams to design chips for cutting edge nodes, fabricate them, build supporting hardware and datacenters and compilers, than to just rent some TPUs on Google Cloud?<p>I can see the case for doing their own edge hardware for the cars (barely), but I really don't think doing training hardware will pay off for them. If they're serious about it, they should spin it out as a separate business to spread the development cost over a larger customer base.<p>Also, I'm really curious whether the custom hardware in the cars is benefiting them at all yet. Every feature they've released so far works fine on the previous generation hardware with 1/10 the compute power. At some point won't they need to start training radically larger networks to take advantage of all that untapped compute power?
Really liked this talk.<p>Looks like they are really nicely orchestrating workloads and training on numerous nets asynchronously.<p>As a person in the AV industry I think Tesla's ability to control the entire stack is great for Tesla... maybe not for everyone who can't afford/doesn't have a Tesla.
I'm still <i>amazed</i> that Teslas team isn't using a map... I know maps get outdated and are sometimes wrong, but having inaccurate knowledge of what's around the corner is far far more helpful than not having any clue whats around the corner.<p>The smart solution would be to consider a map a probabilistic thing, which neural networks are really good at handling.
Interesting that they don't have a full 3D world model. I'm certainly not a machine learning expert. I'm still amazed the route from image recognition to a 2D map of "what's drivable" to autonomous driving is so direct. One would expect to hit a ceiling really soon with that approach.<p>To me it seems we're still in really early days.
One thing I didn't quite understand is how training sub-graphs in parallel works. If you are editing a sub-graph of a monolith type model, aren't you affecting other graphs that have dependencies on the one you're editing? If these are independent graphs, then what's a "sub-graph" even mean?
For those who want to learn more, I would start with Mask-RCNN where you have a very similar architecture: one shared backbone with multiple heads that can be retrained for various tasks (bounding boxes, masks, keypoints, etc): <a href="https://youtu.be/g7z4mkfRjI4?t=628" rel="nofollow">https://youtu.be/g7z4mkfRjI4?t=628</a>
The good news for me is that the upper bound for fully autonomous self-driving cars is no more than 50 years away. What a time to be alive. If it happens before then, that will be an absolute bonus.
Andrej Karpathy is such a treasure.<p>He is an excellent presenter who really has a passion for teaching.<p>Im not really involved with the industry, so I cant really speak to how he holds up to other experts. However he is by far the most digestable resource I have found for learning about NN and science behind them.<p>If you are just discovering him now, google his name and just start reading. His work is truly binge worthy in the most meaningful way.
The description of SmartSummon about halfway through the talk is interesting. One of the views looks like SLAM using a particle filter, but Andrej seems to say that it's done entirely within a neural net.
Just listening to this talk scares me. The amount of errors - even in a seemingly normal, sunny day - is mind boggling to think people trust this crap.<p>How can we rely on the output of eight cameras? This is not a kid's science project.<p>It's all fancy neural networks until someone dies. Pretty callous and Silicon valley-mindset for such an important and critical function of the car.<p>Will never buy a Tesla after having seen this.