BEIJING, Oct 17 (Reuters) – Chinese technology giant Baidu (9888.HK) on Tuesday unveiled the newest version of its generative artificial intelligence (AI) model, Ernie 4.0, saying its capabilities were on par with those of ChatGPT maker OpenAI’s pioneering GPT-4 model.
CEO Robin Li introduced Ernie 4.0 at an event in Beijing, focusing on what he described as the model’s memory capabilities and showing it writing a martial arts novel in real-time. He also showed Ernie 4.0 creating advertising posters and videos.
Analysts were unimpressed. Ernie 4.0’s launch lacked major highlights versus the previous version, said Lu Yanxia, an analyst at industry consultancy IDC.
Baidu’s Hong Kong shares fell 1.32% in morning trading, underperforming a 0.7% rise in the broader Hang Seng Index (.HIS).
“We should see significant improvements once Ernie 4.0 is used hands-on, but concrete upgrades aren’t immediately clear,” Lu said.
Lu said other key announcements from the event included Baidu’s integration of generative AI across all its products, including Baidu Drive and Baidu Maps.
Li demonstrated how Baidu Map now allows users to access functions with natural language queries powered by Ernie, whereas previously users had to search through thousands of options.
Baidu, owner of China’s largest internet search engine, is at the forefront of AI models in China amid a global craze over the technology sparked by the introduction of ChatGPT last year.
The firm launched a chatbot powered by Ernie in March, dubbed ErnieBot, though investors were disappointed to be shown only pre-recorded demonstrations.
In August, Baidu was among a number of firms to receive government approval to release AI products to the public. Ernie has amassed 45 million users since being opened for public use, Baidu’s chief technology officer Wang Haifeng said during the event.
China now has at least 130 large language models (LLMs), representing 40% of the global total and behind only the United States’ 50%, data from brokerage CLSA showed.
Last week, Beijing published proposed security requirements for firms offering services powered by the technology, including a blacklist of sources that cannot be used to train AI models.
Reporting by Yelin Mo and Eduardo Baptista; Writing by Brenda Goh; Editing by Christopher Cushing and Sonali Paul
: .