BEIJING (Reuters) – The head of one of China’s leading AI firms, Baidu, warned on Wednesday that the fervor rushing to develop large language models in China could lead to a waste of resources, saying firms needed to focus their efforts on developing practical applications.
Robin Li made his comments at an industry forum in Shenzhen amid mounting market worries over a potential industry shakeout as companies developing large language models have yet to find viable business models.
Since the release of chatbot ChatGPT by OpenAI late last year, generative artificial intelligence (AI) has attracted significant interest in China, with both established companies and startups jumping into the field.
“I’ve observed a phenomenon (in China) where many industries, companies and even cities are purchasing hardware, stocking chips, (and) building computing centres to train proprietary large models from scratch.”
“A large language model itself is a basic foundation akin to an operating system, but ultimately developers need to rely on a limited number of large models to develop various native applications. Therefore, constantly redeveloping foundational large models represents an enormous waste of social resources,” Li said.
Li said there are many large models in China, but AI applications developed based on large models are still very few.
As of October, 238 large language models have been released, up from just 79 in June, according to Li, citing a third-party report.
Baidu’s own large language model, called Ernie, was opened to public use in August, joining other products that have received government approval for release. Last month, Baidu unveiled the newest version of its generative AI model Ernie 4.0, whose first version was first rolled out in March.
(Reporting by Yelin Mo in Beijing and Brenda Goh in Shanghai; Additional reporting by Eduardo Baptista; Editing by Kim Coghill)
