Baidu teases Ernie 4 large language model after China approves ChatGPT-style chatbots for public use

The Beijing-based search giant will soon launch a new version of its large language model, founder and CEO Robin Li says
Baidu Cloud says nearly 10,000 clients are actively using Qianfan, its model-as-a-service platform, to train AI models and build applications

Artificial intelligence

Ben Jiangin Beijing

Published: 7:30pm, 5 Sep 2023

Why you can trust SCMP

Baidu, the Chinese internet search giant that opened its chatbot to the general public last week, said it would soon launch the latest version of its large language model (LLM), casting a fresh vote of confidence in the nation’s artificial intelligence (AI) sector after Beijing granted its first batch of approvals for ChatGPT-style services.

Robin Li Yanhong, founder and CEO of Baidu, announced the imminent arrival of the Ernie 4 model at a conference held by the company’s cloud computing unit in Beijing on Tuesday, calling it “one of the best LLMs”.

Ernie 4 can be used to build AI applications for use across a wide range of traditional industries and business scenarios to improve efficiency, Li said.

LLMs are the deep-learning algorithms that underpin a new generation of AI chatbots capable of understanding and generating human-like answers to specific prompts. Last week, for the first time, China gave several tech companies the green light to introduce such AI chatbots to the public.

Baidu Cloud on Tuesday also announced an upgrade to Qianfan, its model-as-a-service platform, and the launch of an ecosystem partners programme.

ChatGPT sparks investment frenzy and soul searching in China

Qianfan now hosts 42 mainstream LLMs from China and abroad, including Facebook owner Meta Platform’s Llama 2 and Google’s Flan, as well as models from the Beijing Academy of Artificial Intelligence, according to Shen Dou, president of Baidu Cloud.

Missing on the list are open-sourced models created by Baidu’s rivals, such as Alibaba Group Holding, owner of the South China Morning Post, and Tencent Holdings-backed start-up Baichuan.

Baidu Cloud pledged to provide financial and technological support, market access and business opportunities to more than 10,000 firms that build AI applications based on the Ernie LLM.

More than 150,000 clients have already signed up to use Qianfan, with nearly 10,000 actively using the platform to train their own AI models and build applications for use in some 400 business scenarios including finance, digital government, heavy industries, marketing, education and transport, the company said.

Since launching a beta version of Ernie Bot in March, Baidu has frequently touted the prowess of its AI technology.

When the company launched Ernie 3.5 in June, it claimed that the model had surpassed OpenAI’s GPT-3.5 in comprehension and outperformed GPT-4 in some Chinese-language capabilities, based on a test by state-backed newspaper China Science Daily.

08:54

Is China’s technology falling behind in the race for its own ChatGPT?

Baidu’s latest move is expected to heat up the country’s increasingly competitive cloud service market, according to a research note by Wang Kai, senior equity researcher at Morningstar Asia.

“Ernie will empower [Baidu’s] cloud business to use generative AI in its services to cater to new clients,” Wang wrote.

Other Chinese tech giants have also sought to upgrade their cloud computing products with AI. E-commerce juggernaut JD.com earlier this year introduced its ChatRhino LLM, designed for industrial applications. Alibaba in April started a programme to bring LLMs into the hotel, energy and telecommunication sectors through partnerships with industry players.

Despite the enthusiasm, analysts said it would take time for LLMs to create substantial returns for these companies.

“Baidu could generate incremental revenue to their cloud or advertising business from incremental clientele outside of its usual scope, who are willing to pay for generative AI that can enhance existing products or improve operating efficiency,” Morningstar’s Wang wrote.