During last night's earnings call with investors, Elon Musk spewed out a bunch of late-night ideas from his dorm room. “What if he was AWS, what if he was Tesla?”
Musk, who loves making jokes on earnings calls, likened the unused computing power of millions of idle Tesla cars to Amazon's cloud services business. If they're just sitting there, he thought, why not put them to good use to run AI models? (Also, have you ever looked at your hands? No, I mean… I really saw it?)
“Distributed inference could actually be performed when the car is not moving,” Musk said. “Imagine a future where there are perhaps 100 million Teslas, each with an average of about 1 kilowatt of inference computing. This means that, distributed around the world, there are about 100 gigawatts of inference computing. equivalent to computing.”
In short, buy a Tesla. It's your property. But Musk wants to free up your car's unused computing power for…something. Maybe AI related? I hope it's not blockchain. (Incidentally, Tesla is now an AI company. Musk himself said so on a conference call.)
Will Tesla pay for this?unclear
Will Tesla pay for this? Unclear. After all, this is Musk's most hypothetical. Still, I'm not going to ignore him trying to take computing power away from customers' vehicles without consent or compensation. GM was providing your driving data to insurance companies without your consent. Baby, it's free.
But before you treat this as a serious idea, you need to determine if it's possible. I reached out to Sam Anthony, former chief technology officer of Perceptive Automata, a now-defunct company that developed a module that allows self-driving cars to perform “theory of mind” tasks. I took it.
Anthony said that conceptually, it is “totally possible” to split large computing tasks into many smaller nodes. We've seen it done with Bitcoin mining and his Folding@home, a decentralized computing project that develops new treatments. But just because something is possible doesn't necessarily mean it's a good idea.
Anthony said there are two main issues that make cars, especially electric cars, imperfect nodes for distributed computing projects. First, it must rely on the energy source of the car's battery, or charging station if connected to a power source, for power. And that power is usually not free, the owner pays a retail price for the power. Second, connectivity and speed are “big issues” in distributed computing, Anthony said.
“Especially reasoning, [machine learning] This is a workflow where speed is important. ”
“Especially reasoning, [machine learning] It’s a workflow where speed is key,” he added. “You're not going to do a ton of offline inference overnight; you're just going to answer questions. This is the big inference problem that AI companies are facing right now. This is what drives connectivity and availability issues in cars (see, , move around) is even more problematic.”
In Musk's mind, a decentralized network would only work when cars are parked or stationary. Still, Anthony argues that no one would be willing to create his architecture for a distributed computer out of millions of automotive ECUs (electronic control units) unless forced to do so in some way.
“It's a very strange-looking figure with a hammer, and I'm imagining the presence of a nail, which is very unbelievable,” he said.
Indeed, computer scientists have been trying to create fast computers from large numbers of small idle nodes for a very long time. One of his earliest examples was his SETI@home, in which Berkeley researchers thought they could discover extraterrestrial life by analyzing radio data using a distributed volunteer network of computers. Ta. So why not Tesla@home?
Indeed, computer scientists have been trying to create fast computers from large numbers of small idle nodes for a very long time.
First, the more geographically dispersed nodes are, the harder it is to make them work together, said Phil Koopman, a professor of electrical and computer engineering at Carnegie Mellon University and co-author of a book on supercomputers. Stated.
Like Anthony, Koopman acknowledged that the project could work as long as the vehicle is connected to a power source during computing to avoid draining the battery. Good his Wi-Fi was also a necessary component, so you may need to park your Tesla at home overnight for the decentralized network to work properly. However, you may still run into obstacles as you grow your project to benefit AI computing.
“Scalability to that scale is always difficult, and it's rarely successful enough to make it worth doing instead of building a data center,” Koopman said. “The devil is in the details, so I'd like to do some serious experimentation to make sure it's doable.”
Mr. Musk likes to boast about what will be possible in a future of self-driving, connected cars. Like a 24/7 robotaxi service, where your car is running and you can earn passive income while you sleep. sound Great in theory. But Musk's big ideas tend to deflate when reality sets in.
“So far, this is an interesting idea,” Koopman said. “But you have to keep in mind that most of these cool ideas are never ready for practical use.”