News

How many steps does Li want to take to make AI?

Publish Date: 2024.12.30

I want to choose a forest, I want to be the biggest in the forest. No matter how difficult it is, no matter what kind of difficulties it requires me to go through, I will definitely be more than just a tree

Not just to become a tree, to become a forest.

This is Li Xiang's expectation for himself and also for his ideals.

Li Xiang gradually regained his momentum on social media due to the failure of his ideal pure electric strategy. Now, what ignited Li Xiang's enthusiasm to "return to the world"?

One of the answers is, AI。

AI， Li Xiang's new 'stimulant'

AI， This is the 'stimulant' for all high-tech industry practitioners at present. Li Xiang is no exception.

Both Ideal and Li Xiang himself have been searching for a new "ignition point" to push Ideal Company to a new stage of development beyond the route of extended range technology.

At the beginning of 2023, Li Xiang solemnly issued a letter to all employees of the company.

The letter mentioned that Ideal aims to become a "global leading artificial intelligence (AI) enterprise" by 2030.

Li Xiang mentioned in the letter to all employees, "Intelligent electric vehicles with autonomous driving will also become the earliest artificial intelligence robots and create the mother ecosystem of physical world artificial intelligence: from AI algorithms to life changing AI products, from AI operating systems to AI inference chips, from AI training platforms to AI training chips, etc

He also stated that Ideal hopes to build a complete systematic capability in the field of artificial intelligence (software 2.0), continuously invest in research and development in the field of artificial intelligence, and continuously exceed user needs in product capabilities, using artificial intelligence robots to change the efficiency and experience of the physical world, and become a "global leader in the artificial intelligence industry" by 2030.

Moving forward in time, at the beginning of 2021, the ideal positioning was to become the world's number one intelligent electric vehicle enterprise, and by 2023, this company aims to transform into a leader in global AI enterprises.

This time, Li Xiang reiterated the brand positioning of Ideal at the 2024 Ideal AI Talk: "Ideal is an artificial intelligence enterprise. What we need to do is not the intelligence of cars, but the automation of artificial intelligence, and we will promote the popularization of artificial intelligence in every household

Our logo has never included the words' car ', and even today our operating company is still called' Beijing Chehejia Information Technology Co., Ltd. '. But hardware is very important to us. If we talk about our vision, a more complete way of calling it would be "connecting the physical world and the digital world," so that we can become a leading artificial intelligence enterprise

By reaffirming the ideal brand positioning at the end of 2024, it can be seen that Li Xiang has high expectations for AI.

He said, "When we released our strategy (vision) at the beginning of 2023, we made a fundamental change, transforming artificial intelligence from a hidden strategy to an open and strategic one

When Li wants to put his "mind" in the sunshine, it means that the ideal AI strategy cannot be just a paper talk, but must be tested from multiple dimensions.

Otherwise, if there are more criticisms, what AI brings to ideals may become a backlash.

So, is the ideal ready?

The steps are not big or deep, 'start a family first, then establish a career'

For a long time, in the eyes of many, there seemed to be a gap of "one hundred and eight thousand miles" between the terms "ideal" and "artificial intelligence".

But if we carefully examine the efforts and accumulation of ideals in artificial intelligence, the above ideas will not be so extreme.

In 2020, Ideal appointed Wang Kai, Global Chief Architect and Director of Advanced Driver Assistance Systems at Wistron, as the Chief Technology Officer (CTO) of Ideal Automotive, responsible for the research and mass production of intelligent vehicle related technologies, including electronic and electrical architecture, intelligent cockpit, autonomous driving, platform development, and Li OS real-time operating system.

It is reported that the ideal autonomous driving team had about 60 people at that time, and by mid-2021, it had surged to around 300 people, reaching a scale of 600 people by 2022.

By comparison, during the same period, the Xiaopeng autonomous driving team had a size of 1200 people, while NIO had 800 people. Obviously, there is still a certain gap between the ideal and the latter two.

In 2020, some Ideal models added simple assisted driving functions such as automatic parking, and began to explore and accumulate technology in the field of autonomous driving.

It is worth noting that Xiaopeng Motors released the G3 full scene automatic parking technology (XPILOT 2.0) in April 2018, which uses 20 sensors for visual functions to achieve automatic parking, mainly to achieve intelligent driving for the last kilometer.

At the time of researching automatic parking, Xiaopeng Motors had already launched its fully self-developed NGP high-speed autonomous navigation driving system in 2020. At that time, in the field of autonomous driving, Xiaopeng Motors became the second global and first domestically developed car company based on an open computing platform, following Tesla.

In 2022, changes in the senior leadership of the Ideal Autonomous Driving team have become particularly frequent.

It was also in this year, as mentioned earlier, that Ideal began to focus on the significance of autonomous driving for a car company.

The ideal new understanding has also caused a series of "turbulence" in the company's autonomous driving team.

In February 2022, Ma Donghui, the Chief Engineer of Ideal, took over the position from Wang Kai. Eight months later in the same year, it was reported that Wang Yilun, the head of the AI algorithm for Ideal Intelligent Driving and the chief scientist of AI, would resign. It is understood that Wang Yilun is mainly responsible for the research and development of AI algorithms such as Ideal Intelligent Driving Perception and Maps.

In December of the same year, Ideal announced a new personnel structure adjustment, with Chief Engineer and CTO Ma Donghui replacing Shen Yanan as President, responsible for the overall R&D and supply groups; Senior Vice President Xie Yan has been appointed as Ideal CTO, fully responsible for the Systems and Computing Group.

So in 2023, Ideal Autonomous Driving entered the stage of no image NOA.

In April 2023, Ideal announced the completion of map free NOA deployment for 100 cities by the end of the year. In July 2024, Ideal achieved full push of nationwide map free NOA, breaking away from dependence on high-precision maps. Through the vehicle's own sensors and algorithms, it can sense road structures in real time and generate driving paths, expanding the application scope of autonomous driving. Vehicles can achieve intelligent assisted driving in more scenarios such as urban roads.

At the same time, the ideal algorithm has evolved from traditional modular architecture to segmented end-to-end and towards OneModel.

Entering 2024, Ideal will introduce end-to-end+VLM technology (Vision Language Model). In July of this year, Ideal released a new autonomous driving technology architecture based on end-to-end models, VLM visual language models, and world models. On October 23rd, it completed the world's first full push end-to-end+VLM dual system intelligent driving, enabling autonomous driving to have logical thinking ability to deal with unknown scenarios.

But the actual application effect is still unknown.

Spend the most money and do the fastest thing

In the automotive industry, in terms of autonomous driving and artificial intelligence, ideals always seem to be "one step behind".

Of course, this is mainly because when Xiaopeng and NIO are "spending money", the ideal is "making money".

Every enterprise has its own development strategy and pace. As an ideal entrepreneur, Li Xiang believes, "One of the biggest benefits is knowing the pace of the entire enterprise's development, which is to solve what problems from 0 to 1 first, and what to do from 1 to 10 after obtaining income. This is actually a fundamental difference between me and new entrepreneurs entering the industry

He said, "Ideal Auto had the worst financing ability in the early stages. With very little money, the first step was to think about how to make the product good. We also gained very good market recognition. In 2020 and 2021, we went public in the United States and Hong Kong respectively, and had more money. So since the beginning of 2020, we have been working on platformization of the entire technology, such as our autonomous driving platforms AD Max and AD Pro, including our cockpit platform SS, as well as our vehicle's domain controller XCU. Later, we will also work on silicon carbide for models and motors, which is our evolutionary process

That is to say, the ideal strategy is to make money first, and then figure out how to spend it.

According to Li Xiang's interview, the current ideal is to have "over 10 billion yuan in research and development investment per year, with nearly half of it invested in artificial intelligence

Is this money considered a lot?

That's quite a lot.

According to He Xiaopeng, Xiaopeng Motors invests 3.5 billion yuan annually in AI research and development centered around autonomous driving.

By comparison, the answer to the above question is naturally clear.

Although relatively speaking, Ideal has spent more money and made smaller progress in research and development. But it cannot be denied that Ideal is indeed wealthy and powerful, which has become the confidence for Ideal to increase its efforts in artificial intelligence in the coming days.

Li Xiang stated in an interview that his requirement for the Ideal Team is to ensure that in the next few years, at least within the scope of China, the base model of the Ideal Big Language Model must be among the top three in the industry.

He said, "Based on such requirements, we are willing to invest in whatever training computing power is needed. We need to truly compete and compete with top companies. The core is that we need to truly build this ability, not just compare it in the automotive industry

So, what is the current ideal level of computing power within the automotive industry?

According to relevant information, Tesla, which is currently leading in the industry, has achieved a computing power of 100 EFLOPS, which is equivalent to the level of 300000 A100 cards. Its self-developed AI chip has reached 7000 TOPS and may be put into mass production in 2026.

Yu Chengdong stated at the Huawei Qiankun press conference at the Beijing Auto Show in April this year that Huawei's cloud training computing power has reached 3.5 EFLOPS. Huawei provides computing power support for autonomous driving through Ascend chips and other technologies, with a vehicle BU reserve computing power of 7.5E.

According to Global Times, the overall computing power of Ideal has reached 6.83 EFLOPS and is expected to exceed 10 EFLOPS by the end of the year. Xiaopeng Motors plans to achieve a computing power level of 7000 H100 cards by the end of the year, equivalent to approximately 4.2 EFLOPS; NIO is a 1.5E.

The significance of computing power level for a company aspiring to develop autonomous driving and artificial intelligence is self-evident.

The competition for computing power resources is also a new winning rule for the development of intelligent driving and artificial intelligence.

An insider once said in an interview, "Starting from the second half of the year, Ideal has almost bought all the cards in the hands of channel merchants

During an interview with the media, Lang Xianpeng, Vice President of Ideal Intelligent Driving, also said, "Li Xiang often asks me if my computing power resources are enough, and if they are not enough, I will buy them

The ideal development of computing power seems to be evolving according to Li Xiang's logic of "getting married first and then starting a career", where "having money makes things easier".

It is reported that in July this year, the ideal cloud computing power was 2.4 EFLOPS, and by the end of August, the ideal computing power had sharply increased to 5.39 EFLOPS. In almost a month, the ideal cloud computing power has increased by nearly 3 EFLOPS.

In addition, according to multiple experts in the intelligent driving industry, there are currently two main ways for car companies and intelligent driving companies to obtain data: one is to mine data from existing data; The second is to explore from mass-produced cars.

As for the latter, the ideal new car sales have already laid a good foundation for the company's computing resources.

In November of this year, Ideal delivered a total of 48740 new vehicles, a year-on-year increase of 18.8%, marking the sixth consecutive month that the delivery volume exceeded 45000 vehicles; So far this year, the cumulative delivery volume of Ideal Automobile has reached 441995 units in the first 11 months.

Since the start of delivery, the total historical delivery volume of Ideal Automobile has been 1075359 units.

Li Xiang said, "As an entrepreneur, it is important to look at different stages

According to his plan, the current stage is an important opportunity for the ideal development of artificial intelligence.

The requirement he put forward to the team is: "By next year, the goal is to become the first in China's space intelligence field, and enter the top three in language intelligence, big language models, and services provided

2025 is just around the corner, and the answer to whether the ideal can achieve the phased goals will soon be revealed.

Hot News

Exhibition

All categories

How many steps does Li want to take to make AI?

+86-15958763640（whatsapp/wechat)

kamen@pktautoparts.top

Mon - Fri / 8:00 - 18:00

PKT Auto Parts

Customer Services

PKT Auto Parts