Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Andrej Karpathy talks about how Tesla's NNs are structured and trained [video]

youtube.com

Post reply on HN
Print view View original

131–140 of 249 posts

Re: Andrej Karpathy talks about how Tesla's NNs are structured and trained [video]

#131

Post by xiphias2 »

navigatesol wrote: post #128
xiphias2 wrote: post #75

Earlier quoted context omitted.

Andrej Karpathy just started working on Tesla's software 2 years ago, before what Chris Lattner did was a mess (he wanted to just have 1 task that learns magically everything), Andrej had to start everything from scratch. Waymo had a 20 year advantage, but Google lost many key people there in the meantime as Larry Page didn't want to launch partial self driving. I think both approaches are great and I wouldn't want t…

Ah yes,the story of Tesla: everything yesterday was crap,but everything tomorrow will be awesome.

Did you really follow how drivers reacted to the changes before/after Karpathy came there? People were extremely dissatisfied with the first versions of Autopilot 2 (as it was worse than Autopilot 1 when it appeared), but it improved a lot, and now it's definitely better.

Re: Andrej Karpathy talks about how Tesla's NNs are structured and trained [video]

#133

Post by modeless »

antpls wrote: post #125
modeless wrote: post #123

Earlier quoted context omitted.

Google has Edge TPUs for use outside datacenters, and they don't support training. Neither do the chips Tesla made for their cars. It's a pretty different problem.

I wouldn't be so sure. Edge TPUs could be the exact same architecture than Google Cloud TPUs, but as you need less computation power for inferring than training, they have simply less transistors on the die and could be underclocked. In other words, Cloud TPUs could be the same architecture than Edge TPUs but scaled to an higher frequency and more packed. I guess we need sources to confirm.

Edge TPUs support 8 bit integer math only. Training is floating point. That's not a small change. https://coral.withgoogle.com/docs/edgetpu/models-intro/#mode...

Re: Andrej Karpathy talks about how Tesla's NNs are structured and trained [video]

#134

Post by tim333 »

new_realist wrote: post #21

Meanwhile Waymo is way ahead.

Waymo has a significantly different approach using lidar rather than just vision. The approaches seem to have different strengths and weaknesses. Waymo is actually able to do full autonomy but in very restricted environments - basically semi deserted suburbs. Tesla's autopilot works in real city rush hour traffic but not reliably enough to be let lose on it's own. It remains to be seen which will win or if it will be some other solution.

Re: Andrej Karpathy talks about how Tesla's NNs are structured and trained [video]

#135

Post by heavenlyblue »

jacquesm wrote: post #110
xiphias2 wrote: post #75

Earlier quoted context omitted.

Andrej Karpathy just started working on Tesla's software 2 years ago, before what Chris Lattner did was a mess (he wanted to just have 1 task that learns magically everything), Andrej had to start everything from scratch. Waymo had a 20 year advantage, but Google lost many key people there in the meantime as Larry Page didn't want to launch partial self driving. I think both approaches are great and I wouldn't want t…

> Larry Page didn't want to launch partial self driving That's the responsible thing to do.

It’s just that Google is a much larger target for negative PR response than any of the other companies.

Re: Andrej Karpathy talks about how Tesla's NNs are structured and trained [video]

#136

Post by codeulike »

jacquesm wrote: post #116
roystonvassey wrote: post #99

Earlier quoted context omitted.

Also, the software part of it (NNs and their algorithms) have been so widely researched and published that competitive advantages here are harder to come by than in hardware RD. Also, vendor lock-in is a huge challenge in the cloud space. I don’t think Tesla would be comfortable with the fact that all their training data sits on a potential competitor’s datacenter.

A car is a hardware device as well, and an electric car does not have the kind of power budget that allows you to throw oodles of standard pieces at it without paying a severe penalty in range.

Compared to the energy needed to move the car, everything else is pretty irrelevant. Power hungry features like Heating/AC only makes a few % difference to range.

Re: Andrej Karpathy talks about how Tesla's NNs are structured and trained [video]

#137

Post by tim333 »

semi-extrinsic wrote: post #102
panarky wrote: post #31

Earlier quoted context omitted.

> mind boggling to think people trust this crap It's also mind boggling to think we currently trust organic tissue to do this crap, some of which is bathed in psychoactive chemicals. And yet we do, and as a result, horrendous catastrophes occur every minute of every day. > It's all fancy neural networks until someone dies No, that can't be the standard, not when people are dying right now in the current regime. Unles…

To really be a measurable improvement over humans, and especially an improvement that is statistically distinguishable from just letting the safety tech of 2019 percolate into the average car, self-driving needs to achieve fatality rates around 0.1 per billion miles. Current averages for the US are around 10 per billion miles and decreasing; best countries in Europe are already below 5 per billion miles. There's some…

I'd quibble with your stats that self driving has to be 100x better than human to be a measurable improvement. You can estimate fairly well if a human driver is safe or not from a few hours as a passenger by seeing if they notice everything and if they have near misses. Also it's not to hard on published data to see the Tesla autopilot seems a fair bit worse than humans under the same conditions. I imagine they will improve.

Re: Andrej Karpathy talks about how Tesla's NNs are structured and trained [video]

#138

Post by xiphias2 »

jacquesm wrote: post #110
xiphias2 wrote: post #75

Earlier quoted context omitted.

Andrej Karpathy just started working on Tesla's software 2 years ago, before what Chris Lattner did was a mess (he wanted to just have 1 task that learns magically everything), Andrej had to start everything from scratch. Waymo had a 20 year advantage, but Google lost many key people there in the meantime as Larry Page didn't want to launch partial self driving. I think both approaches are great and I wouldn't want t…

> Larry Page didn't want to launch partial self driving That's the responsible thing to do.

If you look at self driving automation as a black box, sure.

But at the same time people understand that on a highway Tesla Autopilot is safe enough to be used on a long boring road, and dricers generally feel less tired (and can focus more on the harder parts of the road).

Re: Andrej Karpathy talks about how Tesla's NNs are structured and trained [video]

#139

Post by tim333 »

perl4ever wrote: post #49
Joky wrote: post #40

Earlier quoted context omitted.

You seem to use "regulated" as in "forbidden" right now, but maybe I am misreading so can you elaborate? Today thousands are dying every year on the road, are we just forbidding cars entirely? > And no, failover control is not acceptable given the past incidents and deaths. How many incidents/deaths per miles driven? How does it compare to all the other transportation systems?

Why are people so committed to the idea that self-driving cars are anywhere near human standards? It just seems like a groundless assertion of faith to me. Professional drivers can go for a million miles without an accident, and I don't believe anyone's autonomous driving software can get within an order of magnitude of that without a disengagement, even in favorable conditions. You may disagree that a disengagement…

I don't think most people think they are near human standards now. But there is the prospect for them being there before too long.

Re: Andrej Karpathy talks about how Tesla's NNs are structured and trained [video]

#140

Post by ben_w »

adamnemecek wrote: post #7

The trick for level 5 is learning the mapping between the lidar point cloud and the video stream. It’s the best of both worlds.

That might be a trick. It’s not the trick human brains use. It might be equivalent to the way that when we say we want to “look at” a thing, we often also want to touch it.

https://kitsunesoftware.wordpress.com/2017/07/27/why-do-peop...

Post reply on HN