China’s AI Ambitions: A New Era of Open-Source Innovation
The Beijing Academy of Artificial Intelligence (BAAI) has taken a significant step forward in China’s artificial intelligence (AI) development, launching a suite of open-source AI models and tools to fuel the country’s adoption of the technology. This move is seen as a fresh example of China’s efforts to make advances in AI, despite US attempts to curb its access to advanced technologies.
China’s AI development gets a boost with open-source models
The BAAI, which received a visit from Chinese Premier Li Qiang in March, unveiled a suite of large language models (LLMs), including its latest iteration of the BGE model, Emu 3 multimodal, and a family of models used to develop the world’s first clinical robot for cardiac ultrasonic scanning. These open-source tools are designed to make it easier for institutions and companies to conduct AI training.
“[BAAI is] committed to frontier, strategic, and original research and breakthroughs in the AI field,” Wang Zhongyuan, the organisation’s head, said in a keynote speech delivered at an industry conference. “In the past few years, BAAI has made extraordinary contributions to technologies that propelled the whole industry’s development.”
The BAAI’s open-source LLMs have been downloaded over 47 million times globally, surpassing Facebook parent Meta Platforms’ Llama 3 series, which has been downloaded over 1.2 million times on developer platform Hugging Face.
BAAI’s contributions to AI research
Founded in November 2018 with the backing of the Beijing municipality and the Chinese science and technology ministry, BAAI has long been hailed as the cradle of China’s LLM research. It was established several years before Microsoft-backed OpenAI shook the global technology world with its groundbreaking ChatGPT debut in November 2022.
In 2021, BAAI launched Wudao 1, the country’s first LLM. A number of researchers who worked on the Wudao project are now spearheading some of China’s high-flying AI start-ups.
AI start-ups thrive in China
Yang Zhiling, who heads Moonshot AI, worked on the Wudao project at BAAI before becoming a tech entrepreneur. Yang Zhiling and Zhang Peng, who head Moonshot AI and Zhipu AI, respectively, both trace their roots back to BAAI and the Wudao project. The two start-ups are both valued at more than US$3 billion in recent funding talks, making them the most valuable Chinese AI firms.
During Li’s visit to BAAI, he was told that China was falling short of catching up to the United States in AI because the nation’s efforts were “littered with many essential challenges in theory and technologies.”
As Washington considers additional measures to contain China’s AI rise over national security concerns, BAAI remains hopeful of more global collaboration and dismissed the idea that the US could deter China’s AI development.
Global collaboration in AI development
In a recent interview with local media, Wang said that the US eventually will choose to work together with China, especially when the era of artificial general intelligence arrives, and the world needs to cooperate to contain AI-induced risks. Wang said the restrictions the US has placed on China’s access to AI computing power, talent, and the technology exchanges between the two countries will “certainly have an impact on China,” but it can only delay, not stop, China’s AI progress.