Transformative AI makes Chinese automobiles smarter
SHANGHAI -- Settling into the plush comfort of his vehicle cockpit, a driver surnamed Liu spoke to the sleek dashboard display: "Simo, how long before I reach home?"
"You have an hour's journey ahead," answered the onboard intelligence system supported by the large language model (LLM), an artificial intelligence (AI)-powered program that can recognize and generate multimedia text.
"Please activate the home air conditioning and close the curtains 30 minutes prior to my arrival." "Affirmative."
This scenario unfolded within Jiyue, an intelligent vehicle backed by Geely Auto, a prominent Chinese automaker, in collaboration with Yiyan, Baidu's LLM platform.
Despite the complexity of overlapping voices, simultaneous commands or ongoing conversations, the smart voice system can adeptly discern and address the unique requirements of the commander.
Its adoption rate has reached an impressive 98 percent among its car owners, who on average utilize the system more than 60 times daily, said Xia Yiping, CEO of Jiyue Auto.
"See it, say it, and get it done." Achieving this level of sophistication for smart cockpits is the goal of China's automotive industry.
China's competitive edge in the automotive market is not just a result of its burgeoning new energy vehicle (NEV) sector. The swift progression in the smart capabilities of these vehicles is equally pivotal in securing this advantage.
The country is committed to fortifying and elevating its vanguard status in the sector of intelligently-connected and new energy vehicles, according to this year's government work report.
China's auto landscape is now enriched with the integration of large-scale AI models, a collaborative effort between leading technology firms and automotive manufacturers.
Prominent tech firms contributed models like Huawei's PanGu, Baidu's Yiyan and iFlytek's Xinghuo. Carmakers also made strides with their own proprietary architectures, including BYD's Xuanji and XPeng's Lingxi.
Now, over ten Chinese vehicle brands have embraced this technological leap, incorporating advanced AI models into their lineup.
"Cars, serving as a terminal that harnesses a plethora of cutting-edge technologies, have evolved from electrification to intelligence, and are poised to seamlessly integrate with the broader transport infrastructure in the future," said Fang Yunzhou, the founder of Hozon Auto, a Chinese electric car startup.
Previously, drivers were required to issue direct commands to their voice assistants, like "activate the air conditioning." But now, with the incorporation of an advanced AI model, a simple statement such as "I'm feeling a bit chilly" suffices.
The sophisticated AI, demonstrating an understanding reminiscent of human cognition, can respond to the driver's needs by executing a sequence of actions. This includes closing the windows, tuning the air conditioning to the driver's accustomed temperature, and adjusting the fan to their usual setting.
"In the past, the process of voice recognition involved data analysis in the cloud followed by a download to the vehicle," said Xia. "Now, the task can be done offline, a transition that allows drivers to operate their vehicles seamlessly even in the absence of an internet connection."
"The AI was predominantly governed by explicit rules, while contemporary large-scale models are propelled by data," said Zhong Xinlong, an AI expert at CCID consulting, a think tank under China's Ministry of Industry and Information Technology.
"This is akin to the concept of learning from specific examples and then applying that knowledge broadly, a skill that mirrors human learning and application of knowledge," Zhong added.
Intelligence in automobiles is not confined to smart cockpits. Chinese automakers have taken a significant step forward by integrating the AI model across the entire spectrum of vehicle intelligence to achieve a comprehensive and vertical alignment of applications in multiple scenarios.
"Large-scale AI models are far more intelligent in perception, behavior and control than we might imagine, as they can, for example, interpret signs such as 'ETC ahead is about to be repaired, please change lanes'," said He Xiaopeng, XPeng chairman and CEO.
XPeng's intelligent driving system, which once received updates on a quarterly basis, has seen an average of 3.87 new updates being rolled out daily, according to He.
"The integration of large-scale AI models into vehicles has already brought about a transformation from zero to one," said Fang.
"Looking ahead, the evolution will be exponential. As this technology matures, we anticipate a future where the responsibility of driving is progressively handed over to the vehicle's autonomous systems," Fang added.