He Xiaopeng Talks About AI+"Fuyao": Xiaopeng G6 Equipped with GPT May Arrive in the Second Half of the Year!

"Fu Yao" architecture debuts, Xiaopeng G6 makes its first appearance, what other plans does Xiaopeng Motors have for the future?

In an interview with Cheyun.com, Xiaopeng Motors CEO He Xiaopeng first explained the significance of "Fu Yao" for Xiaopeng: globalization, modularization, low cost, and efficient rapid release of different price models.

Regarding Xiaopeng G6, autonomous driving, and the recent trend of artificial intelligence, He Xiaopeng said that GPT will have a significant impact on the automotive and robotics industries, "starting in the second half of this year, we will bring GPT's capabilities to G6 and integrate it into the 'Fu Yao' system. After next year, we will be able to combine GPT with autonomous driving."

In He Xiaopeng's view, GPT is a general artificial intelligence that is expected to accelerate the arrival of L4 and L5 autonomous driving by 5-10 years. Xiaopeng may launch a software and hardware system based on L4 in 2025, which is mainly due to the rapid development of GPT.

The following is the original interview with He Xiaopeng by Cheyun.com CEO Cheng Li:

Cheng Li: First of all, I would like to ask you about your feelings about this year's Shanghai Auto Show. Because it has been three years since we had such a big auto show.

He Xiaopeng: I think the auto show is a good way to integrate China and the world, and to connect technology and art. In this year's show, I think electric vehicles and new energy vehicles are the main focus.

Last time I attended the Shanghai Auto Show, there were still gasoline cars, but there were more new energy vehicles.

But I heard that in this year's show, no one really released gasoline cars, they are all facing the next generation of cars. There is a saying on the Internet that "no one can stop the trend, the trend has arrived."

Cheng Li: Today we saw Xiaopeng G6 on display, but what we are more concerned about is the release of our new global intelligent Fu Yao architecture the night before. The outside world is actually more concerned about this. What does this architecture release mean for Xiaopeng?

He Xiaopeng: I think in the past nine years, Xiaopeng has been exploring what is the intelligence of intelligent cars, and what is the electric and car of intelligent electric cars. So every car and every modification is changing, innovating, and exploring.

But I think the new Fu Yao technology architecture means for Xiaopeng: it can be globalized, modularized, low-cost, efficient, and able to quickly release models of different price ranges. **

I think that in the next three years, we cannot know who will win, but we can know who will not win in the future. So, in these three years, can we use a relatively low cost to make cars that are most suitable for customers at different price points, and can we go from China to the world, from non-intelligent to intelligent? I think this is the most important value of this architecture.

This architecture enables us to have the ability to release more than one car per year, which is different from before. It also allows us to have a great change in R&D efficiency and R&D costs, and it enables the operation of our intelligent cars to be consistent, efficient, and scalable, which was not possible before. I think this is a huge change.

Cheng Li: Does it include both hardware and software?

He Xiaopeng: Yes, it includes both local and cloud.

Cheng Li: Can it also adapt to globalization in the future?

He Xiaopeng: Yes, it can. All the cars we plan to globalize, including left-hand drive and right-hand drive, will be extended on the Fuyao architecture.

Cheng Li: I noticed that you used the term "architecture" instead of "platform" in the naming. Was there any thought behind this?

He Xiaopeng: In fact, in traditional cars, more platform-based modularization is used, but from our current perspective, adding "intelligence" to it is not even a platform in a certain sense. We can see that in today's Chinese Internet, global Internet, and mobile Internet companies, software and services are consistent. It is not a platform issue, but a continuous dynamic iterative consistency.

Therefore, under this architecture system, we focus on the car itself. You can still think of it as modularization and platformization, but for power, especially for the local and cloud of intelligence, we still hope to standardize integration and operation, and scalability.

Cheng Li: You think that with this architecture, because other competitors also claim to have architecture, or they are also promoting architecture and platform, what do you think is the difference between us and them, or what do you think our advantages will be?

He Xiaopeng: I haven't observed and learned too much from other competitors. I think they must have their own advantages and characteristics. I think a very big advantage of Xiaopeng's architecture is that it is based on the foundation of our rapid iteration for the past 9 years, which is very different.

Secondly, we will consider the needs of intelligent development in the future, which I think will be very different from them.

Thirdly, we believe that to make a global car, it must comply with global safety standards and global data standards.

Fourthly, can we achieve high efficiency and low cost in both software and hardware?

Cheng Li: In fact, when we saw the emergence of Chat GPT, not only the technology industry, but also everyone was paying attention to this product. Its impact on various industries, such as Xiaopeng, how do you see its birth, and will it have an impact on your future products or operations? Do you have any relevant thoughts in this area?

** He Xiaopeng: I think GPT is actually a groundbreaking product. Although its capabilities today have not yet reached a truly groundbreaking level, it is on its way. I think GPT will have a huge impact on the automotive and robotics industries.

In the past, we were the leader in China's autonomous driving, but I never believed that high-end L4 or L5 could truly arrive. Because if you tell a car a simple rule, it cannot act like a real driver when it encounters various special situations. But if GPT continues to advance for another three years, the integration with the car will be completely different.

We started with G6 and will bring GPT's capabilities to G6 in the second half of this year. It will be fully integrated into the future system, and GPT's capabilities will be coupled with autonomous driving capabilities in the future.

In my opinion, GPT is a general artificial intelligence. You can think of it as a lightweight artificial brain. What is autonomous driving? Autonomous driving is like a person. It is responsible for movement and balance, so when a person does not need a brain, he can still walk forward. It is stable and can judge and adapt by itself.

Therefore, if autonomous driving is the cerebellum and brainstem, and electronic appliances and many embedded OAs are the spine and spinal system, I think the brain, cerebellum, brainstem, spine, and limbs, the car will eventually become more and more like a robot, more and more like a smart transportation tool, and will completely change the capabilities of L4 and L5 in the next 5 to 10 years.

Cheng Li: Do you think it will accelerate the arrival of L4 and L5?

He Xiaopeng: I think there is a chance to advance it by 5 to 10 years.

Cheng Li: Did you anticipate the birth of this product? For example, when Chat GPT 3.5 came out, we didn't see anything particularly amazing about it, but after 4, we really felt amazed. Did you predict this before?

He Xiaopeng: I think the development speed of GPT may be faster than the intelligence development speed of cars.

Cheng Li: Yes.

He Xiaopeng: According to our plan, we believe that Xiaopeng will launch a hardware and software system based on L4 at some point in 2025. I think with GPT, maybe in the high-end L4 or quasi-L5, everyone will see it between 2027 and 2030. At this time, the car can not only help us drive automatically, but it can even do more for us.

Cheng Li: So, in terms of judgment on the entire intelligent driving technology, you are relatively optimistic or relatively radical. It is similar to Huawei because there are indeed two very different views in the market. Actually, you should have noticed that both Yu Kai and Wang Chuanfu have mentioned that we can only walk on the road of L2+++ now. The road to L3, L4, and L5 will be very long, or even there is no optimistic expectation for this matter.

Because Xiaopeng has been working towards this direction, where does your underlying logic and confidence come from? You just mentioned that it is possible in 2027?

He Xiaopeng: I think the market has different views on this matter.

As I just wanted to say, I think no one can beat the trend. First of all, the trend is coming. Then, another challenge in the trend is what kind of rhythm it is. I think different people have different views on the rhythm.

I think that Xiaopeng is different from other suppliers. An important point is that we are a complete vehicle manufacturer plus a self-developed software and hardware manufacturer. So why can we do standardization and scaling so well?

I think some manufacturers and partners are relatively pessimistic. They believe that the cooperation of hardware manufacturers will be slow. For example, the brother of a hardware manufacturer we just talked about would say that he is not very optimistic.

I think this is because different entrepreneurs have different perceptions of this matter, and there are significant differences.

From my perspective, I am extremely optimistic about the future. But for the process, I think that every process needs to grasp the rhythm. There may be many challenges in the process, but we cannot be pessimistic about the future because of the challenges in the process. This is the first point.

The second point is that to make this happen, I think it must be a complete vehicle manufacturer's self-developed intelligent software and hardware. Otherwise, it is impossible to achieve the three-in-one of experience, safety, and cost.

Cheng Li: In fact, there are some obstacles now, or breakthrough points are needed. How many breakthrough points do you think are there? Can it eventually be widely used?

He Xiaopeng: I think there are mainly three places:

The first place is whether the experience can be consistent throughout the entire process.

What have we solved in the past time? Only automatic driving on the highway and parking. To be blunt, these two scenarios only account for about 5% of the scenarios. What else do you need to achieve in the remaining 95% of scenarios? You need to be able to solve the community and parking lots without using maps. You need to be able to recognize various special situations beyond toll stations, speed bumps, pedestrians on the road, and animals. Therefore, you cannot use lidar, or you cannot use lidar completely. If these two things can be achieved, it is first of all a full-scenario. This will be solved in all of Xiaopeng's future 18 months.

The second is whether it can give people a relaxed and safe feeling like an experienced driver.

Our goal is to increase one year of driving experience every year. We launched the NGP in the city first in China last year. I saw that many competitors also want to launch it this year. But from our perspective, the core is the combination ability of the front and back ends. Can we increase the price by one year every year? So if we can achieve a driver with 4 to 5 years of driving experience in 2023, 2024, and 2025, I think this will solve the safety problem for people. The third issue is about cost.

I have always seen how mobile phones have succeeded, and their success has gone through three stages. First, Apple and Android launched a new generation of operating systems and phone models; second, Chinese phones made it possible for global phones to reach 1000 yuan; and third, there were many new content services, such as WeChat and Didi.

The logic of the second is how to make it cheap and scalable, so I also want to say that Xiaopeng will make autonomous driving standard and scalable next year, which is important.

So why do I say that the turning point of autonomous driving will be very clear in the next three years, and those who used to do Level 2 will find that their abilities are no longer needed. Because from another perspective, the functions that a Level 2 can achieve are like our previous NGP, which only solves 5% of the scenarios, and although its CPU is cheap, it cannot be used elsewhere.

There should be an integrated solution that ultimately allows a car to solve everything with only one series of computing power, which is the right way.

Cheng Li: What do you think about breakthroughs in laws and regulations in this area?

He Xiaopeng: This year I am a representative of the National People's Congress. In my proposal for the representative this year, I made suggestions on how to solve insurance issues, financial issues, and policy and regulatory issues.

Now I see that multiple ministries are constantly promoting it. I think that China is promoting what is called high-quality development. China must make fully autonomous driving the world leader. I think we are now standing at the best juncture and we must move forward.

Cheng Li: Another topic is actually about the operation of Xiaopeng as a whole. We have seen that the entire architecture of Xiaopeng's operation has undergone some adjustments in the past six months. I don't know if you think this adjustment is in place now?

He Xiaopeng: I think we are on the road.

I think the most core enterprise's final sales performance is good, which comes from product technology architecture and organizational management capabilities. How to find a management that can face customers in China or even globally, and can have rapid iteration and cost control in fierce competition, based on such a range.

To be honest, some companies that used to do cars well need to learn software, intelligence, and iteration; I think some new car companies that used to need to learn cost, quality, and module-related things in the automotive industry need to learn from each other. So who can learn fast and walk smoothly in these 3 to 5 years will eventually succeed.

Cheng Li: Actually, this adjustment is a very critical lesson for Xiaopeng. I think in this lesson, I believe that you actually have a very in-depth thinking about company operations.

After this adjustment, what are your conclusions? What do you think Xiaopeng lacked in the past and what it needs to grow in the future in terms of company operations?

He Xiaopeng: There are many points that I think are: innovation based on capabilities, and increasing innovation based on costs. In fact, I even think that the difficulty of innovation based on costs is even greater than that of capabilities. The key is how to truly get to the front line, instead of just treating customer first as a slogan. This is very important.

Because the competition among traditional car companies was not intense enough before, it is only now that it has become truly fierce. Previously, the gross profit margin was relatively high, but now it has become very low, even negative. So how to manage the entire process well in such a system is a challenge and a change for traditional companies.

I think these points are our views in the process of rethinking.

Cheng Li: Actually, Xiaopeng Motors previously mentioned 3 million vehicles, which is a goal that companies need to strive for. What is Xiaopeng's target for this year?

He Xiaopeng: I'm sorry, I can't tell you our target for this year.

Cheng Li: Another thing is your evaluation of Wang Fengying, who has just taken office, and the new head of autonomous driving, Xin Zhou. What do you think of them?

He Xiaopeng: Wang Fengying is a new colleague, and Xin Zhou is an old colleague. What is the logic of evaluating them together?

Cheng Li: Because I think they are both very important partners for you.

He Xiaopeng: I think there are many other important partners besides them, including Henry (Xia Hang) and Brend (He Tao), etc.

I think Wang Fengying has strong professional ability and a strong sense of responsibility. She is not like a traditional auto industry person, but more like an entrepreneur, which I really like.

I think Xin Zhou is rare in that he understands technology well, but also tries to grasp product engineering and management. Most technical directors do not have enough comprehensive ability in this area.

Cheng Li: Great, thank you very much! Thank you all!