(JW Insights) Aug 28 -- Chinese universities' research into large language models(LLMs) is being restrained by insufficient computer power and datasets needed to train such models, making it hard for them to compete with research bodies in the corporate sector, academics said at the World Artificial Intelligence Conference's Youth Scientists Forum on August 24.

It is difficult for universities to purchase the huge amount of computing power that LLMs need, said He Lianghua, deputy dean of Tongji University's Department of Computer Science and Technology. The academic circle is also unable to obtain the vast amounts of data that are needed to train parameters to be mature, he added, reported Yicai Global on August 25.

Only a handful of large companies have the resources needed to train LLMs, said Xie Enze, a researcher at Chinese telecom giant Huawei Technologies' Noah's Ark Lab. When building high-quality datasets it is important to know beforehand the kinds of specifications, how much data is available and how long the training will take.

Academic researchers may not have the same focus as the industrial sector, said Qian Xuejun, an assistant professor at Shanghai Technology University's School of Biomedical Engineering. The academic circle is more inclined to innovate in terms of new ideas and new applications, rather than leveraging large computing power and big data to make breakthroughs, reported Yicai Global.

It is an inevitable reality that the most advanced research in LLMs is being carried out in the business sector, said Qiu Xipeng, professor at Fudan University's School of Computer Science who, together with his team, developed MOSS, a chatbot like US AI firm Open AI's ChatGPT.

This is the case in the US where many prestigious universities in the US have already been outperformed by top corporate research institutes such as DeepMind and San Francisco-based OpenAI, Qiu said. But the academic circle has its own strengths as it has a more long-term view and its research is not profit driven, Qiu said.

For most business research institutes, their aim is to make money, Qiu said. There are many research centers set up by Chinese companies that are also under financial pressure in in just a few years' time.

With the success of ChatGPT, AI-generated content has become a hot topic among tech giants and the Chinese government. The country has launched 79 large language models with over 1 billion parameters each, the Ministry of Science and Technology reported in May, according to the Yicai Global report.

(Li PP)