21st Century Business Herald reporter Song Doudou reported from Guangzhou
"The automotive technology revolution with electric intelligence as the main feature has allowed the automobile industry for more than 100 years to return from a sunset industry to a sunrise industry, and also allowed China's automobile industry to quickly narrow the distance with the world's advanced level by changing tracks." If high-quality development is the only ticket in the second half, then intelligent networking is the new commanding height. At the 2023 Future Vehicle Pioneer Conference held recently, Wang Xia, president of the Automotive Industry Branch of the China Council for the Promotion of International Trade, said that compared with electrification, intelligent network will have a more far-reaching impact on the automotive industry.
In fact, in recent months, artificial intelligence represented by ChatGPT has set off a revolution in the entire industry. Especially in intelligent vehicles as the next generation of intelligent terminals, with the rapid popularization of intelligent driving and intelligent cockpits and the continuous improvement of functions, the industry has accelerated the application of large models on the car while also bringing greater imagination to the intelligent electric vehicle industry.
From a functional point of view, there are currently two main landing forms of artificial intelligence in automobiles, one is used in the field of artificial intelligence communication and dialogue, most of which are used in intelligent cockpits, such as Baidu's Wenyan Yixin, and nearly ten car companies such as Dongfeng Nissan, Hongqi, and Great Wall have announced access; During the Shanghai Auto Show, SenseTime unveiled the new Sensenova model, demonstrating the combination of its Chinese language models "SenseChat" and "SenseAvatar" with the cockpit. Previously, Alibaba also announced that the AliOS smart car operating system has been connected to Tongyi Qianwen model for testing. The other is to focus on intelligent driving, such as the release of DriveGPT, a generative model of autonomous driving, to help solve cognitive decision-making problems and ultimately achieve end-to-end autonomous driving.
However, it should be pointed out that there are still many challenges in the process of entering the automotive industry represented by ChatGPT - whether data security can be guaranteed, how to implement the envisaged scenario, what preparations should be made under the dual pattern of opportunities and challenges, and whether the large model will become a "Pandora's box", all need to be answered by the entire automotive industry in the practice process.
AI big models reshape smart cars
In the industry's view, large models open the AI era and will reshape thousands of industries, and new cars have many interactive subjects, many interaction methods, strong interaction stickiness, many computing parts, large data scale, spatial attributes and social attributes, etc., which determines that new cars are the largest interactive application scenarios of large models.
The application of large models on the car will also bring profound changes in interactive intelligence and service intelligence to the automotive industry, bringing new experiences including interaction with emotional digital humans, open and rich service ecology, and intelligently generated content and expressions.
"Automotive intelligence is an important focus for the combination of the new generation of ICT and automobiles, which will change and reconstruct the industry, expand the value chain, and now form a strong competitive ecology, and the development direction is irreversible." Li Keqiang, academician of the Chinese Academy of Engineering, professor of Tsinghua University and chief scientist of the National Intelligent Networked Vehicle Innovation Center, said that recently artificial intelligence technology has appeared a large model, focusing on the automotive field, the essential attribute of intelligent driving is to replace human operation through the device, from this point of view, artificial intelligence is a necessary professional technology for cars to achieve a higher level of intelligent development.
Li Keqiang believes that the advantages of AI large models in processing text, obtaining and processing data, and establishing training and iteration of scenarios will accelerate the intelligent human-computer interaction and intelligent driving.
Huatai Securities pointed out in the research report that the big model paradigm is expected to empower core links such as intelligent driving perception labeling and decision-making reasoning in vertical fields, accelerate the implementation of intelligent driving, and at the same time develop large models or promote the rapid growth of driving data and computing power demand.
Chen Zhuo, general manager of Baidu's autonomous driving business department, said that AI technology has accelerated the scale of autonomous driving, automatic driving is a typical application scenario of artificial intelligence, artificial intelligence gives intelligent cars super brain, taking planning decision-making as an example, the rules from rule-based to self-learning algorithms have been realized, self-learning algorithms can surpass the experience system, more intelligently handle complex scenarios, and greatly expand the design and operation scope of autonomous driving.
"The big model is the underlying technical support of Baidu's various businesses, and just like Wen Xin's words, autonomous driving and intelligent transportation will also benefit from the big model and achieve better development." According to its introduction, relying on the advantages of the Wen Xin Yiyan large model, the Wen Xin graphic model has thousands of object recognition capabilities. In addition, the integration of the large model into the smart cabin scene will reshape the car space, redefine the relationship between people and cars, and the cockpit will also evolve into the core carrier of the third living space.
Some industry insiders predict that in 2023, intelligent driving products will enter the full outbreak period, and large models will open the landing application on the car end.
"Only a set of universal intelligent driving solutions that can be used in the whole world and are cost-effective can usher in the 'ChatGPT moment' of intelligent driving." Zhou Guang, CEO of Yuanrong Qixing, said that the DeepRoute-Driver 3.0 solution launched by Yuanrong Qixing in March this year does not need to be equipped with high-precision maps, and the hardware cost is 80% cheaper than the previous generation solution launched in 2021. It is understood that at present, the high-end intelligent driving product D-PRO has obtained the designated cooperation of car companies and has been tested in Shenzhen.
It is worth mentioning that in addition to technology companies and autonomous driving companies accelerating the application of AI large models, many car companies are also taking action.
Ideal Auto said that it will apply the self-developed cognitive large model Mind GPT to the "ideal classmate", so that the upgraded "ideal classmate" will become the "housekeeper" of the car. It is reported that the technology is similar to the voice interactive version of ChatGPT, which has undergone the key training links of large language models and has the ability to generate safe, accurate and logical conversations.
Mercedes-Benz and Microsoft announced last week that the two companies are collaborating to test in-vehicle ChatGPT artificial intelligence that can be used by more than 900,000 cars equipped with MBUX infotainment systems in the United States.
At the Guangdong-Hong Kong-Macao Greater Bay Area Auto Show, Yu Chengdong, Executive Director of Huawei, CEO of Terminal BG, and CEO of Intelligent Vehicle Solution BU, revealed that the AITO M9 will be equipped with an AI model, and the detailed functional experience will be revealed at the AITO M9 conference this fall.
"Cars are moving towards the era of intelligence, the investment scale of car manufacturers is very large, and insufficient investment cannot support competition, and small and medium-sized players may not be able to keep up and become giants, because heavy asset investment is needed to support sustainable development in the future." Yu Chengdong said that Huawei invests more than 10 billion yuan in R&D in the automotive field every year, focusing on intelligent cockpits, intelligent driving, intelligent network connections and other fields, but the difficulty of intelligent driving is much higher than that of intelligent cockpits, and 70%-80% of Huawei's investment is in intelligent driving.
Data compliance and scenario implementation have become new challenges
Under the "infinite reverie space" brought by ChatGPT, the application of large models in the automotive field still faces many challenges.
Li Keqiang pointed out that at present, the application of large models in the automotive field is still in the early stage of experimentation, and the follow-up still needs to be opened up based on the cloud platform, and there is still a long way to go to promote the application of deep fields, and there is still a long way to go to completely replace human thinking.
"Now the real model data mobilization management needs to use intelligent networked vehicles and computing technology platforms, cloud control technology platforms and other platforms, only a large number of data convergence can do crossover, especially in vertical fields, which is different from entertainment systems and service systems, from the perspective of the industrial Internet, if the basic cloud platform can not be opened, it is difficult to achieve the advancement to the deep field." Li Keqiang said.
Yu Fei, academician of the Canadian Academy of Engineering and executive director of the Guangdong Laboratory of Artificial Intelligence and Digital Economy (Shenzhen), believes that "everyone's expectations for generative artificial intelligence represented by ChatGPT are a bit high". He pointed out that every time there is a breakthrough in the field of artificial intelligence, the car people want to apply it in the field of cars, but things are not so simple.
He cited Tesla's FSD (Full Self-Driving) system as an example, "When Tesla launched FSD in 2014, it was asked why Tesla couldn't achieve self-driving when artificial intelligence developed to such an extent? Musk said soon, explain the year, but it has been nearly a decade now, and Tesla still has not achieved full self-driving. ”
In his view, artificial intelligence used in the automotive field is mainly divided into three parts: perception, model or thinking, execution, at present, artificial intelligence has made a breakthrough in natural language understanding technology, which can help the car improve in the field of perception and model on a large time scale, but the era of playing a role in the small time scale, especially in the field of car control, has not yet arrived.
It is worth noting that although artificial intelligence is profoundly changing the automotive industry, the application of large models represented by ChatGPT is still far away. On the one hand, relevant policies and regulations have not yet been promulgated, and data security cannot be guaranteed; On the other hand, the landing scene is not yet clear.
Wang Zheng, vice president of business of Heduo Technology, previously said that automatic driving faces the contradiction between artificial intelligence and mass production, so that autonomous driving technology can be applied by the public, in addition to artificial intelligence, many dimensions of work are required. After truly liberating people, we can better experience and experience the so-called more human-machine and softer interaction.
Some insiders told the 21st Century Business Herald reporter that there are still many risks in the application of large models in the automotive field, including whether the entire ecology is ready, and whether the vehicle end has enough computing power and resources to run the model; How to land the scene, under the extremely high real-time requirements of the car, whether the feedback of artificial intelligence can solve the problem in milliseconds, and how to ensure data security; And ethically and legally, whether things generated by large models can be allowed are all challenges.
For more information, please download 21 Finance APP