Advertisement
Artificial intelligence
TechTech Trends

China’s AI players from Alibaba to state-run institute rush to open source their models in race for global influence

  • The Beijing Academy of Artificial Intelligence, Alibaba Cloud and various Chinese start-ups have shared their large language models with global users
  • Open-sourcing is a way for industry players to extend their reach and stand out from the competition, according to experts

Reading Time:2 minutes
Why you can trust SCMP
3
A sign seen at the World Artificial Intelligence Conferenc in Shanghai, China in July. Photo: Reuters
Ben Jiangin Beijing
China’s Big Tech firms, start-ups and institutions are rushing to share with the rest of the world their large language models (LLMs) – the software that underpins chatbots like ChatGPT – as they race to acquire global users amid a competition over artificial intelligence (AI) supremacy.

The Beijing Academy of Artificial Intelligence (BAAI), a government-backed facility focused on foundational AI research and its application in areas from medical use to life science, became the latest major Chinese player to open source a new AI model when it made its BGE model freely available for commercial use on Tuesday.

In June, the academy also opened up its Aquila model, which had been pre-trained on 7 billion parameters, 40 per cent of them Chinese and 60 per cent English, according to the model’s profile introduction on Github.

A model’s capability hinges partly on its number of parameters. OpenAI’s ChatGPT, for instance, was trained on 175 billion parameters. While most open-sourced Chinese LLMs right now have between 6 and 13 billion parameters, BAAI said it planned to launch a more advanced version of Aquila with 33 billion parameters.

08:54

Is China’s technology falling behind in the race for its own ChatGPT?

Is China’s technology falling behind in the race for its own ChatGPT?
BAAI’s moves come as the cloud unit of Alibaba Group Holding, owner of the South China Morning Post, last week open-sourced two 7-billion-parameter AI models, based on its flagship Tongyi Qianwen LLM launched in April.

Available on both Alibaba’s own repository ModelScope and US platform Hugging Face, the two models can be used by scholars, researchers and companies free of charge.

Advertisement