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

Tesla is building its own AI chips for self-driving cars

techcrunch.com

Post reply on HN
Print view View original

61–70 of 238 posts

Re: Tesla is building its own AI chips for self-driving cars

#61

Post by titzer »

Nae3Au5x wrote: post #21
titzer wrote: post #3

This is pretty bad from a hardware/software verification perspective. People who care about correctness can just go home, I guess. It's amazing that we're gonna let these things on the streets. When you weigh almost 2 tons and can go >100mph, it doesn't matter if you have a gun mounted on you or not, you are a weapon .

In the same sense that any object that can potentially kill someone (hint: everything) is a weapon. Which is technically correct but a useless observation. If we use a more narrow definition of weapon, e.g. a tool optimized for the primary purpose of injuring or killing people then it certainly is not a weapon.

> any object that can potentially kill someone (hint: everything) is a weapon

Any object that can be remotely programmed to drive itself at high velocity into a target, yes.

Re: Tesla is building its own AI chips for self-driving cars

#63

Post by oliveshell »

Ultimatt wrote: post #27
tntn wrote: post #4

> By having its own silicone Will this mistake never go away?

Only when everyone learns a major property of fake breasts isn't that they solve computational problems via the field effect.

...and that semiconductor wafers don’t provide a flexible watertight seal around windows, even if you crush them up and squirt them from a tube.

Re: Tesla is building its own AI chips for self-driving cars

#64

Post by mattigames »

planteen wrote: post #44
slivym wrote: post #35

>By having its own silicone, Tesla can build for its own needs at its own pace. If they suddenly recognize something the hardware is lacking, they’re not waiting on someone else to build it. It’s by no means a trivial task — but if they can pull it off without breaking the bank (and Elon says it costs them “the same as the current hardware”), it could end up being a significant strength. I outright snorted at this. T…

They have also conflated silicon and silicone, which is humorous.

Blame the English language for making the name of 2 popular industrial materials with such similar names.

Re: Tesla is building its own AI chips for self-driving cars

#65

Post by titzer »

ben174 wrote: post #8
titzer wrote: post #3

This is pretty bad from a hardware/software verification perspective. People who care about correctness can just go home, I guess. It's amazing that we're gonna let these things on the streets. When you weigh almost 2 tons and can go >100mph, it doesn't matter if you have a gun mounted on you or not, you are a weapon .

It only has to perform better than a human.

Humans are not remotely programmable.

Re: Tesla is building its own AI chips for self-driving cars

#66

Post by TeMPOraL »

Nae3Au5x wrote: post #21
titzer wrote: post #3

This is pretty bad from a hardware/software verification perspective. People who care about correctness can just go home, I guess. It's amazing that we're gonna let these things on the streets. When you weigh almost 2 tons and can go >100mph, it doesn't matter if you have a gun mounted on you or not, you are a weapon .

In the same sense that any object that can potentially kill someone (hint: everything) is a weapon. Which is technically correct but a useless observation. If we use a more narrow definition of weapon, e.g. a tool optimized for the primary purpose of injuring or killing people then it certainly is not a weapon.

> In the same sense that any object that can potentially kill someone (hint: everything) is a weapon. Which is technically correct but a useless observation.

It's not useless; the damage that can be caused varies by degree between things. Then you have to look at two cases - suitability for object to be used as a weapon, and the damage it can cause on accident. Unlike knives or hammers, both those factors are very high for cars.

The fact how dangerous cars is is very much underappreciated by people in general, as evidenced by the number of morons on the road. We already lose hundreds of people daily in the US alone because of this; now we're trying to add another class of drivers into the mix - algorithms written by greedy optimizers caring primarily for short-term profit and being first to market. This should give us some pause.

I'm not saying this technology is not possible or not wonderful, but I think the current ecology of self-driving efforts is unhealthy. We have a for-profit race by companies, many of which can't be trusted with getting software right, and most (all?) of them pursuing self-driving capabilities by means of half-understood brute-force black boxes the neural networks are.

Re: Tesla is building its own AI chips for self-driving cars

#67

Post by bonestamp2 »

melling wrote: post #12

“Nvidia’s hardware was handling about 200 frames per second, its specialized chip is able to do crunch out 2000 frames per second “with full redundancy and failover”.” What are the real-world benefits going from 200 to 2000 frames a second? 200 seems quite fast.

I think that comparison is just to illustrate the computational improvements. In other words, the new system is 10x faster. As Elon said on the call, that means they can now process all the cameras at full resolution and full framerate, which can improve accuracy and therefore ability and safety... which sounds like the real world benefits you're looking for.

Re: Tesla is building its own AI chips for self-driving cars

#68

Post by visarga »

gallerdude wrote: post #6

The benefit they point out is 200fps vs. 2000fps, but is that actually that useful? In the post of OpenAO's robot, they said that improving the response times gave no significant gains. Only thing I can imagine would be allowances for wider and deeper neural nets.

I think latency makes a lot of difference when an autonomous system is interacting with the world. Take a look at this video from 2009 where they use a high speed vision system (1000fps!). The things it could do are probably unequaled even today.

https://www.youtube.com/watch?v=-KxjVlaLBmk

The higher the speed of the system, the less it has to predict in the future and better can adapt to unforeseen situations. It's basically using the world as a model, for free.

Re: Tesla is building its own AI chips for self-driving cars

#69

Post by TeMPOraL »

mhneu wrote: post #47
3pt14159 wrote: post #25

Earlier quoted context omitted.

I worry less about software / hardware crashes leading to death and more about cyberattack. I've written about it before[0]: We need regulations before a mass breach of autonomous systems happens. Not after. This whole bullshit where corporations claim all sorts of things about their security then get owned trivially has happened for every OS and device out there. The supposedly secure QNX was installed into countles…

YES. Security compromise is the huge achilles heel of self-driving cars - and anything that is algorithmically controlled. When you have 100,000 self-driving cars on the road, and someone figures out how to hack them, and drive them simultaneously into crowds, that will be a BIG BIG problem. And this problem isn't just about self-driving cars. It is a fundamental issue with all algorithms. It's a lot harder to hack h…

> It's a lot harder to hack humans to do bad things

I'd say, it's probably easier to hack individual humans, but - like everything in computing - hacking algorithms scales well, while human factors generally don't.

Re: Tesla is building its own AI chips for self-driving cars

#70

Post by Retric »

simion314 wrote: post #48
Retric wrote: post #38

Earlier quoted context omitted.

> what numbers are you basing this on I trust insurance companies to have good data, they think it improves safety when you compare rates.

But you are referring at current Teslas ? This cars require the driver to pay 100% attention, by law and by Tesla agreements so isn't obvious that driver+drive assist+expensive car has better average numbers then driver + average (maybe old) car ? I would like to see the numbers of how many times a driver had to intervene to save the situation

It's an option so insurance companies compare the exact same car with and without that option.

In terms of paying attention I really doubt most people are doing this very well. I am going to be stuck in the car either way so by default I am going to be watching the road, but as long as my relative stress level drops that's a huge win.

PS: You can make your own estimates. If people pay attention sufficient to avid accidents 99% of the time the car might be in 100x more accidents without drivers assuming those drivers where perfect. Drop that to IMO more likely 75% and it's closer to 4x at the absolute best case.

Post reply on HN