3 ms·
Find it hard to believe there's anything valuable to steal from Tesla. Most the hardware is proprietary so stealing source for main computer would be kinda poin
by dzhiurgis 7y ago
Find it hard to believe there's anything valuable to steal from Tesla. Most the hardware is proprietary so stealing source for main computer would be kinda pointless.
Stealing neural net would be even more pointless - it's mainly built for US roads, to run on specialised AI computer and control specialised hardware and can probably be extracted via firmware update anyway.
Plus Musk is always claiming to open source their patents and innovate so fast that their old software is irrelevant.
Disclaimer: I know nothing about self driving car industry.
- sfblah 7y agoBest comment downvoted as usual. Well done, HN.
- ganzuul 7y agoThe poster gives the perfect reason to downvote right at the end of the post. Open and shut case.
- surfpel 7y agoHardware being proprietary doesn’t diminish the value of stealing code. In fact, there is much to be gained regardless of specific hardware, even if the code was super hardware dependent. And besides, what’s to stop them from developing similar hardware in the end? The point isn’t to just blindly run the code on their hardware, it’s to reverse engineer it. There’s also much more to a neural net than just the training of it. Its methods and architecture can be re-applied to different situations, regardless of what a stop sign or roads look like. In the same vein, iPhones use a different chipset than android phones but stealing Apple code and taking it to Google would still be of benefit.
- ironman1478 7y agoThis stuff would be useful. Self driving is getting to the point where it's a game of inches and there are only extremely hard problems to solve. Each major player definitely has non-overlapping sets of the solution. Understanding how their main compute works could hint at how their system works as a whole. Like how are they getting data from sensors? Are they doing HW acceleration? On which parts of the pipeline are they doing that and should we? How are they synchronizing all of their sensors to do sensor fusion? It's clear that a self driving car will need to make use of a large suite of sensors which require tons of data processing. If Tesla is primarily camera based, it could be possible that they have invested in better cameras and better image processing algorithms + hw acceleration that competitors could learn from. Same goes for any serious player in this space.
- heavenlyblue 7y agoYeah, I’ve got a feeling that the secrets that are actually worth stealing are not the secrets about the high-level workings of the autonomous cars. It’s the necessary piping that gets all of these together that is probably the (major) long-tail of the problem and where most of the costs lay.
- Erlich_Bachman 7y ago> Disclaimer: I know nothing about self driving car industry. I just wonder, why do you comment then? If you by your self-assessment know nothing, you would just be adding to confusion around the topic, or in the best case scenario would randomly get some correct fact or intuition and thus increase confusion even more because some people would assume the rest is correct. Why not just comment on the things you know something about?
- gt2 7y agoIf any of the other things he said are true, he brought something to conversation.
- MauranKilom 7y agoI would read the post with implied ", right?" after every statement given the disclaimer. Sure, it would've been possible (and likely better) to phrase everything as questions directly, but raising interesting talking points to start a discussion is... kind of what we are here for.