China’s open-source AI models rival US giants, making engagement urgent: Stanford
The report calls on US actors to engage with Chinese firms as the AI landscape continues to evolve
China’s open-source artificial intelligence models may have caught up or “even pulled ahead” of their US counterparts in capabilities and adoption, a Stanford University report has found.
The country’s status as a leading AI power meant that US firms should not avoid “selective engagement” with Chinese AI labs, academics and policymakers, given the wide range of AI governance and safety challenges Chinese players faced, according to the report published on Tuesday.
The report was prepared by Stanford University’s DigiChina Project, housed under the university’s Center for International Security and Cooperation, and its Institute for Human-Centered Artificial Intelligence, which compiles the influential annual AI Index report.
“Today, Chinese-made open-weight models are unavoidable in the global competitive AI landscape,” the report said. “There is space for academic collaboration with Chinese counterparts to increase our understanding of the risks of open-weight AI models and the efficacy of various guardrails.”
While Chinese AI developers are largely pursuing an open-source approach, releasing the weights of their models for end-users to freely deploy and modify, US tech giants such as OpenAI and Google DeepMind have kept their leading models as proprietary software.
Citing leading industry benchmarks, the report noted that Chinese open models now performed at “near state-of-the-art levels”, leading the world in the open-source landscape and barely trailing leading closed US models.