(Yicai Global) May 30 -- Large language models developed in China are far behind OpenAI’s ChatGPT and must overcome challenges in raising money and securing data to catch up, according to researchers.

China-made LLMs are similar to ChatGPT when it comes to chatbots but fail to keep up as productivity tools and will not reach the same level for months, Qiu Xipeng, head of Moss, the LLM of Fudan University Natural Language Processing Group, told Yicai Global.

Domestic LLMs are one to two generations behind OpenAI, including in knowledge and learning ability, said Yang Zhiming, founder of Huawei Technologies-backed iDeepWise Artificial Intelligence. It is difficult to use time to measure the gap as OpenAI will develope at an increasingly faster pace, Yang noted.

Computing power is not an obstacle when trying to catch up with OpenAI because this can be resolved by securing “garbage time” from global cloud service providers at night, Zhang Zhen, founder of Beijing Whaty Technology Development, pointed out. The core issue is whether algorithms and data can generate intelligence, Zhang added.

China’s LLMs cannot equal OpenAI in the short term because it requires huge investment, and no local developer has invested as much as OpenAI, said Huang Yan, president of Shanghai Guochuang Technology Industry Innovation and Development Center. Domestic investors have a wait-and-see approach as the field entails big investment, high risk, and potential high returns, noted Huang, who is also a managing partner of Lantern Capital.

In addition to computing power, data costs are also high because the training of LLMs needs huge cross-language, cross-industry, and cross-field data, and acquiring it is not easy, according to Tian Feng, head of SenseTime’s intelligent industry research institute. 

Leading companies, tech firms, and national laboratories should jointly build open ecology and form the world’s largest public open-source Chinese dataset to increase the training speed and quality of China-made LLMs, Tian added.

Editor: Martin Kadiev