[ad_1]
Tesla manufactures cars. Now, it is also the most recent company to find the limit artificial intelligence at his design silicon chips.
Pa an advertising event last month, Products revealed more about an AI device called the D1 machine learning machine behind Autopilot self-driving. The event takes a closer look at Tesla AI’s work and shows a dancer who looks like human robot the company wants to build.
Tesla is the latest chipmaker to make its own silicon. As AI becomes more important and cheaper to ship, other companies with more sophisticated technology — including Google, Amazon, and Microsoft — are also now making their own chips.
At the event, CEO of Tesla Elon Musk he said, squeezing a large number of computer systems that train the company neural networks it will be necessary for you to proceed with the self-driving car. “If it takes a few days for a trainer to exercise against a few hours, that’s a big deal,” he said.
Tesla has already started developing sensor-sensing chips in its vehicles, after using Nvidia tools in 2019. But making powerful and sophisticated tools needed to train AI algorithms is very expensive and challenging.
“If you believe that the only way to drive is to teach a great way to help, then what was followed was the way to connect around what you would like,” he says. Chris Gerdes, Supervisor Center for Traffic Research at Stanford, who attended the Tesla ceremony.
Many car companies use neural networks to detect traffic, but Tesla relies heavily on the technology, with one very large network called “transformer” receiving receipts from eight cameras at a time.
“We are better off building animals made from the ground,” said Tesla, AI, Andrej Karpathy, at the August ceremony. “A car can be thought of as an animal. It moves independently, recognizes the environment and protects itself. ”
Transformer models have provided great features progress in areas such as understanding the language in recent years; Benefits have been made from making the colors bigger and more informative. Teaching large AI applications it costs several million dollars important power cloud computing.
David Kanter, a chip analyst with Real World Technologies, says Musk bet that by sending the training faster, “then I can make both machines – a self-driving program – more advanced than Cruises and Waymos in the world,” said two Tesla fighters in self-driving.
Gerdes, of Stanford, says Tesla’s mind is built around his neural networks. Unlike most automotive companies, Tesla does not use lidar, a low-cost brand that can see the world in 3D. He only relies on interpreting the images using neural network networks to display inputs from his cameras and radar. This is very difficult because the algorithms need to re-create circular maps from the camera’s cameras instead of relying on sensors that can capture the image directly.
But Tesla also collects more education than other car companies. One of the more than 1 million Teslas on the street also sends the company video footage on its eight cameras. Tesla says it employs 1,000 people to print the images – and see cars, trucks, road signs, road signs, etc. – to help train a major transformer. At the event, which took place in August, Tesla added that it could select the images that it would prioritize in writing to make the project a success.
Gerdes says one of the risks of Tesla’s approach is that, at some point, adding more will not make the system any better. “Is it a matter of more detail?” he says. “Or are the neural networks just getting lower than you expected?”
Answering this question can be costly in any way.
The rise of large, expensive AI brands not only encouraged other big companies to make their own chips; has also made many well-paid startups working on a special metal.
[ad_2]
Source link
