Some of the comments on the article about Tesla' Dojo (below) are FACINATING and would seem to lend quite a bit of weight to previous dot joining attempts IMO. Comments written about 18 hours ago and I hope they don't mind me posting them here?
First of all Dojo does not exist yet. They have a first functional engineering sample of a single integrated module that they just received last week. It is a long, long way off from being fully operational and there will be many supercomputers and other developments coming in the time between now and then.But the approach taken in Dojo is really impressive and has great potential, assuming Tesla keesp investing in its development and wider rollout. The only way to get things to a good cost/performance is scale and you are not going to get sufficient scale of use if the Dojo is there just for Tesla projects. It needs to be scaled up to multiple data centers and then rolled out like AWS service with both public and private nodes farmed out to other customers. Only then there will be enough use to be able to fund future hardware development and scale manufacturing.And they have very good pre-requsites for that. PyTorch integration and the LLVM based JIT are essential steps here, so the users do not have to learn a completely new language or computing paradigm to use Dojo. Support for FP32 is also very important for general use, even if the in-car networks are more limited in what they can execute. The low in-chip RAM might be limiting, however, but they acknowledged that and will be planning larger amounts for next generation (after Dojo 1).The only big risks Dojo has is the competition - other silicon manufacturers, like Intel, AMD, NVIdia, Samsung and Qualcomm have far more experience and larger R&D budgets and more direct access to manufacturing facilities. They are also working on both AI accelerators for various applications and AI training systems (both GPU-derrived and specific designs). If some of them come up with a far better design in 5 years, it would be sad if Tesla team would become too integrated into the Dojo and unable to switch to something else, even if it was better.
Ivan Ivanov's comment:
I’m more convinced than ever that autonomous driving requires a high degree of AI development, which Tesla doesn’t possess at this time. Musk probably realized it a few years ago. They can’t make FSD work as a fully autonomous driving system until they make a breakthrough in AI.My belief has been for a while now that fully autonomous driving systems will be developed by just a few companies (not necessarily automotive ones) and will be licensed out to automotive manufacturers. The two large companies that have a real chance of developing such a system are Alphabet and Apple. Tesla may or may not be a third one.We have seen a similar trend in car infotainment, as a microcosm. Many automotive companies tried to develop their own infotainment systems for decades, but eventually almost every one of them decided to go with the Android Auto / Apple CarPlay integration because none of those automotive companies could make their infotainment to be even close in functionality and usability to Android Auto or Apple CarPlay. The same exact thing will happen with fully autonomous driving systems. Automotive companies will develop their own Level 2 driving assist systems and will wait for the fully autonomous systems to be developed by the big AI guys.While Musk is focusing on AI to deliver on his FSD promise, the other automotive manufacturers will be focusing on delivering quality EVs and improving the range and the speed of charging. The EVs produced by the traditional automotive manufacturers will outclass Tesla within just a few years. We already see much more compelling EVs coming out of the competition now. By the time Musk is ready with the FSD (if ever), other automotive manufacturers will be able to buy fully autonomous systems from Apple, Alphabet or another specialized AI company.In my opinion, a fully autonomous Level 5 driving system will not come about until at least 2030, but that’s a very optimistic forecast. I wouldn’t be surprised if it were pushed to 2035 or even 2040. I’m still hoping to have it drive me around in my old age.
BRN Price at posting:
50.0¢ Sentiment: Hold Disclosure: Held
(20min delay)