OpenAI explodes one more time: What must domestic companies do to keep pace?
Persisting in vying for dominance in AI applications the way out for domestic companies
Published: May 16, 2024 03:37 PM
A concept photo of artificial intelligence (AI) Photo: VCG

A concept photo of artificial intelligence (AI) Photo: VCG

During a series of live demonstrations, GPT-4o's remarkable capacity for understanding, rapid responsiveness and human-like ability to empathize and convey emotions left audiences in awe. Within 24 hours, Google released its main updates including the launch of Gemini 1.5 Flash and Project Astra, among others.

According to OpenAI, the GPT-4o trained a single new model end-to-end across text, vision and audio, which allows it to present a much more natural human-computer interaction, with the response time to audio inputs being 232 milliseconds, which is similar to human response time in a conversation.

The standard that this model has set, combining all of text, vision and audio modalities, is something that China's domestic large-model companies are also attempting to achieve. Amid the rapid updates of OpenAI and Google products, what stages are domestic large-model companies at?

'Different punch'

"It is way too natural to tell that it's actually a machine that is talking. The variety of its tone, intonation and its ability to comprehend emotions is highly similar to that of a human being," Xiao Yanghua, a computer science professor at Fudan University, also director of Shanghai Key Laboratory of Data Science, told the Global Times when describing GPT-4o. 

Humans are unable to learn to comprehend and express emotions without adapting in a social environment. OpenAI achieved this significant technical challenge by training multimodal data, Xiao said. "The emergence of the GPT-4o basically achieved the goal of natural human-computer interaction that the computer science has been pursing over the past 80 years."

Many Chinese companies including ByteDance, Alibaba and MiniMax have been researching and developing products with comprehensive comprehension abilities. Yet, experts said more products in the market are focused on lightweight agents such as AI search engines and professional assistants. These are hardly able to compete with GPT, even in the audio real-time interaction level.

The recently-released C-end product "Small Conch" by Chinese startup MiniMax can provide real-time voice conversations. It showcases a clear sense of AI in terms of reaction speed, expression, timbre and emotion as the demo videos of the company showed.

Catching up with GPT-4o is something that can be done by domestic companies when the gap of domestic computing capability has gradually been made up, Li Baiyang, an assistant professor from the data management innovation research center of Nanjing University, told the Global Times. 

Developing basic large models should be carried out by top-tier AI companies with a strong financial background. From this perspective, China has some outstanding and competitive products, Li said.

For example, Alibaba Cloud, the digital technology and intelligence backbone of Alibaba Group, recently announced the launch of Tongyi Qianwen 2.5, becoming one of the most popular large models for Chinese enterprises. 

According to the latest data, Tongyi has more than 90,000 service enterprises through Alibaba Cloud and more than 2.2 million service enterprises through Dingding, and has now landed in PC, mobile phone, automobile, aviation, astronomy, mining, education, medical care, catering, games, cultural tourism and other fields.

But Li pointed out that not every company needs to develop its own large language models. More medium and small-scale AI companies in China should keep competing for space at the application level.

Li continued to say that the current gap with the US lies in the fact that the capabilities still lean toward a personal consumer-grade product, like the GPT-4o. However, Xiao added, China can leverage large models to achieve groundbreaking applications. 

"When the US throws a left punch, China could counter with a right punch. Regardless of the strength of its left punch, we can still deliver our right punch," Xiao vividly illustrated China's current pursuit of an advantage in AI technology development.

From technical tools to "assistants" capable of understanding individual needs, in the industrialization process of AI, by more evenly distributing the workload of large models across public clouds, private clouds, edges and terminals, it helps to drive the intelligent transformation of various industries, industry observers said. 

Prominently, AI-powered smartphones in China is expected to bring a leap in functionality to smartphones. At the beginning of this year, Honor smartphones released a new generation of full-scenario operating systems, with the monthly usage of its large model reaching as high as 15 million times, and the daily peak usage reaching 850,000 times, according to Beijing Youth Daily.

Another highlight on accelerated usage of AI is its deployment of applications in intelligent connected vehicles, ranging from interacting more naturally with people in smart cabins and accurately recognizing people and objects inside and outside the vehicle. Large models can also improve the efficiency and safety of autonomous driving systems. According to the Xinhua News Agency, there are more than 10 brands of cars carrying large models.

Tackle the core 

Since the first half of the year, major players in the large model race, such as iFlytek and SenseTime, have continuously iterated new features of their large models at the product level, exclusively targeting increasing efficiency of workflow segment. 

In terms of improving efficiency, the contribution from AI application is significant, especially in writing documents and even increasing the code writing of programmers by 30-50 percent, Xiao said. 

The value of AI application can be fully used when the technology is able to play a role in tackling "core links of industries," thus further expanding China's pursuit of an advantage in AI application with the US, Xiao explained.

He mentioned that AI helps to assist investors to make investment decisions in the financial sector such as spotting potential loopholes in contracts. In the medical field, when medical experts are unable to, AI is able to identify specific diseases that patients are suffering. 

AI can make up the limitations and flaws of human cognition working in various industries, Xiao summarized. In this regard, there have been no killer-level applications emerging from abroad, and the same goes for China, he added.

Li, who has recently been conducting extensive tests on large models, informed the Global Times that "Enhancing digital literacy nationwide will additionally unlock the immense potential of AI applications in the market."

China boasts a vast market, and the potential of this cutting-edge technology will only be fully realized as more users become acquainted with the latest advancements and gain complete access.