Baidu’s CEO, Robin Li, has urged caution in China’s hurried development of large language models (LLMs). Li warned that the current trend could result in a wasteful misuse of resources and called for a stronger focus on practical applications of AI technology. Speaking at an industry forum in Shenzhen, Li highlighted the concerns surrounding the need for viable business proposals for companies working on LLMs.

The release of OpenAI’s ChatGPT last year provoked significant interest in generative AI in China, with established companies and startups joining the race. However, Li worried about the excessive investment in hardware, chips, and ‘computing centres to train proprietary models from scratch’. According to Li, developers should use a limited number of LLMs to create diverse AI applications instead of continually redeveloping foundational models. He stressed that the existing method signifies a waste of social resources, driven by the swift proliferation of LLM releases in China. Li cited a third-party report that showed 238 LLMs released by October, compared to just 79 in June. Baidu, a prominent player in the field, has its LLM called Ernie 4.0, which was made available for public use in August.

Why does it matter?

The surge in generative AI in China, sparked by the success of OpenAI’s ChatGPT, has led to a rapid influx of product announcements from various companies. This trend, dubbed the “war of a hundred models,” involves startups and tech giants, including Tencent, Baidu, Alibaba, and Huawei. China now boasts at least 130 large language models (LLMs), making up 40% of the global total. However, investors caution that many of these LLMs need more viable business models, are too similar, and face rising costs, indicating a potential shake-out in the industry.

cross-circle