Li, the co-founder, chairman and chief executive of Chinese internet search giant Baidu, told his audience at the annual X-Lake Forum in Shenzhen on Wednesday that this frenzy over LLMs – the technology used to train intelligent chatbots like ChatGPT – has resulted in 238 such AI models launched in the country as of October, up from 79 in June. By contrast, there are hardly any successful AI applications that are familiar to the public, he said.
“There are too many big models in China, but too few AI-native applications based on those models,” Li said. AI-native applications are developed on the “unprecedented” capabilities of AI, according to Li, who used the analogy of Tencent Holdings’ super app WeChat as a mobile-native app.
“Developing foundation models continuously and repeatedly is a huge waste of social resources,” he said. “We need 1 million AI-native applications, but we don’t need 100 big models.”
Baidu co-founder, chairman and chief executive Robin Li Yanhong speaks at the annual X-Lake Forum in Shenzhen on November 15, 2023. Photo: Handout
LLMs are deep-learning AI algorithms that can recognise, summarise, translate, predict and generate content using very large data sets.
“If our industry policies can be more encouraging on [creating more] AI-native applications based on the big models, we will surely create a prosperous AI ecosystem and drive a new round of economic growth,” Li said.
Li’s remarks extolled AI’s potential to help drive economic growth and become a useful daily tool, while urging China’s technology industry to be more circumspect in developing the technology to meet those goals.
In his presentation at the X-Lake Forum, Li criticised recent efforts by unnamed enterprises and cities to hoard advanced semiconductors and build intelligent computing centres as part of plans to build their own AI foundation models from scratch. He indicated that these new models would have no “emergent abilities” – referring to capabilities to perform a task via a few prompts – owing to a lack of parameter scale and training data sets.
Intel sees massive demand for AI chips designed for China: report
An LLM’s capability partly hinges on its number of parameters, a measure of the sophistication of a model. OpenAI’s ChatGPT, for example, was trained on 175 billion parameters, while most open-sourced Chinese LLMs in the market have between 6 and 13 billion parameters.
The government should support the demand side and encourage companies to deploy big models for building new AI applications, according to Li.
China’s major tech firms and AI start-ups have been engaged in an intense competition to provide their own ChatGPT-like services, as OpenAI’s popular chatbot and rival services like Google’s Bard are not officially available in mainland China or Hong Kong. OpenAI backer Microsoft Corp, however, has been pushing its GPT-4-powered Bing Chat in the region.
OpenAI’s GPT-4 Turbo likely to spark more competition among Chinese rivals
At the X-Lake Forum, Li claimed that Ernie Bot’s application programming interface call volume has shown “exponential growth” since its August launch, which “exceeded the other 200 models in combination”.
Baidu is currently incubating its own AI applications, such as the code-writing assistant Comate, Li said. Still, he asserted that the best AI-native applications have yet to be developed either in China or the US.
“Just like in the [previous] era that gave birth to ‘mobile-native’ apps like WeChat, Douyin and Uber, in the AI era there will definitely be excellent AI-native applications developed based on these big models,” Li said.
Denial of responsibility! Chronicles Live is an automatic aggregator of the all world’s media. In each content, the hyperlink to the primary source is specified. All trademarks belong to their rightful owners, all materials to their authors. If you are the owner of the content and do not want us to publish your materials, please contact us by email – chronicleslive.com. The content will be deleted within 24 hours.
