Original title: Frontline | Tesla CEO Musk: Confident that L5 autonomous driving will be achieved this year, China's autonomous driving team is being established Source: 36Krypton

On July 9, the 3rd World Artificial Intelligence Conference (WAIC) opened. Tesla CEO Musk delivered a keynote address in the form of a self-question-and-answer session via video access.

Around Tesla's Chinese factory, Tesla's local manufacturing and R&D teams are all systematized. At the beginning of his speech, Musk also posted a job advertisement for China's autonomous driving team. He said, “We are building a relevant engineering team in China. If you want to work as an engineer at Tesla, you can join us.”

Musk also specifically explained that China's autonomous driving team is not just simply moving things from the US to China. “We will be original in engineering development, original design and engineering work in China.”

Autonomous driving is an important scenario for the implementation of artificial intelligence applications, and Tesla is clearly a leader in this field. Since Tesla began installing AutoPilot hardware in vehicles in 2015, it has driven more than 3 billion miles when enabled. Based on massive amounts of real road data, Tesla's intelligent driving system is also rapidly evolving towards autonomous driving.

In his speech, Musk said that he is very confident that Level 5 autonomous driving will be achieved in the future. “I think it will happen soon. At Tesla, we are already very close to L5 autonomous driving, and we are confident that the basic functions of the L5 level will be completed this year.”

The following is a transcript of Musk's speech:

I'm so happy to be here again, and I hope to visit in person in the future!

1. Autopilot autonomous assisted driving is a very popular feature of Tesla electric vehicles. How is it being used in the Chinese market?

Tesla's autonomous driving is doing a very good job in China. We are building a relevant engineering team in China. If you want to work as an engineer at Tesla, you can join us.

I want to emphasize that we will do a lot of original Chinese engineering development, not just simply moving things from the US to China; we will do original design and engineering work, so consider working at Tesla China.

2. How confident are we that we will finally achieve L5 level autonomous driving? When do you think this day will come?

I am very confident that L5 autonomous driving will be realized in the future. I think it will happen soon. At Tesla, we are already very close to L5 autonomous driving, and I am confident that we will complete the basic functions of the L5 level this year.

I don't think there are any fundamental challenges at the bottom to achieve L5 autonomous driving, but there are many detailed issues. The challenge we face is to solve all these minor problems, and then integrate the system to continuously solve the long-tail problem. It can establish the situation in most scenarios, but from time to time, some strange situations will occur. We must have a system to solve training and solve these strange scenarios. Therefore, it is also necessary to have realistic scenarios. Nothing is more complicated than reality; any simulation is a subset of the complexity of the real world.

We are currently very focused on dealing with the details of L5 autonomous driving. We really believe this can be done. We can use Tesla's current hardware. We only need to improve the software, and we can achieve L5 level autonomous driving.

3. Do you think artificial intelligence androbotsThe three pillars of technology: perception, cognition, and behavior. How are they currently progressing in their respective fields?

I'm not sure if artificial intelligence can be categorized this way. If we use this classification standard, technology has made tremendous progress at the perceptual level. In fact, it can be said that in the field of professional image recognition, it has done a better job than almost any human, even an expert. The question is how many computers, how much practice to train computers, and how efficient the image system is.

When it comes to image recognition or corresponding recognition, can any signal, any given byte stream, be accurately recognized by an artificial intelligence system?

Cognition is the weakest field. Is artificial intelligence able to understand concepts, make effective inferences, and create? There are many creative AIs now, but they have no way to control their creative activities. We may think that they are not doing a good job, but they will perform a little better in the future.

We can imagine it as a game. Obviously, in any game with clear rules, AI will be much better than humans. Mainly games with clear rules. If this game doesn't have a high degree of freedom, it's hard to imagine that in any game, artificial intelligence won't play better than humans, and it doesn't even consider the faster response time of artificial intelligence.

4. In what ways has Autopilot automated assisted driving promoted the development of AI algorithms and chips? How has it changed our understanding of AI technology?

When developing artificial intelligence chips for autonomous driving, we discovered that there were no affordable, low-power systems on the market.

If we use traditional TPU, CPUs, or other similar products that consume hundreds of watts of power, the trunk will be occupied by a huge cooling system. Since then, the cost is high, the size is high, and the energy consumption is high.

You need to know that energy consumption is very important for a car's mileage. For this reason, we have developed a Tesla smart artificial chip. This chip has an eight-bit accelerator for motor calculation. There is a lot of motor computation. If you know what motor computation is, you know that the amount of motor computation is huge. In fact, it has not fully exploited Tesla's ability to drive autonomously.

In fact, we launched the second chip system a few months ago, so it may take at least a year to fully utilize Tesla's fully autonomous driving system.

We're also seeing Tesla's training system to improve the training of artificial intelligence systems. Artificial intelligence systems, like FP16 training systems, are mainly limited by chip heat generation and communication rate, so we are also developing new wire and cooling systems to develop more efficient computers to process video data more effectively.

5. How do you view the development of artificial intelligence algorithms?

How do we view the development of artificial intelligence algorithms, I'm not sure if this is the best way to understand it.

Neural networks mainly obtain large amounts of information from reality, much of it without light sources, and create vector space to compress a large number of photons into vector space.

Whether humans can enter the vector space of the brain, we usually take reality for granted by analogy. I think you can enter the vector space in your own brain and understand how your brain processes external information. It obtains a large amount of information and keeps only the relevant parts. How do people create a vector space in the brain. His information only takes up a small part of the original data, but can make decisions based on vector space expressions. This is similar to the large-scale compression and decompression process.

It's a bit like physics, because a physical formula is essentially a realistic compression algorithm. This is the effect of physics. Obviously, a physical formula is a realistic compression algorithm.

Simply put, we humans are evidence of the effects of physics. If we simulate the universe in the true sense of physics, if there is enough time for a large amount of computation, and the end of absolute knowledge, humans will be the best proof. If you believe in physics and the history of the evolution of the universe, it was hydrogen for a long time, and helium and lithium appeared. Some of the heavy elements we learn to express, we humans essentially evolved from hydrogen. If we leave hydrogen for a while, it will slowly become us. I think everyone may not agree with this, so some people may not agree with this, so some people may not agree with this, so some people will ask me where our perceptual effects come from, the whole universe Special desperation is either whether there is a special nature, or when does hydrogen become perceptive during the transformation of humans.

The Tesla factory is progressing smoothly. I am extremely proud of the Tesla team, and I look forward to visiting the Shanghai Gigafactory soon. I don't know how to express this. Thank you very much to the Tesla team in China. We will be using more artificial intelligence and artificial intelligence software in our factories in the future. I think it will take some time for the factory to actually use artificial intelligence effectively. We can think of the factory as a complex combination; in fact, all companies are like this.

Thank you for the online interview. I hope to have the chance to participate in person next year. I really like going to China. China always surprises me. There are many smart and hardworking people in China, and China is full of positive energy. I am looking forward to the future. I am looking forward to making the future a reality, and I am looking forward to coming back.