Meta changes course on open-source AI as China pushes ahead with advanced models

South China Morning Post

2 days ago

Facebook parent Meta Platforms, a major proponent of open-source artificial intelligence (AI) models with its Llama family, has indicated it would be more 'careful' going down the open-source road, a move that contrasts with China's embrace of open-source.
In fact, China has probably found the path to 'surpass the US in AI' thanks to the momentum in the country's vibrant open-source AI ecosystem, according to Andrew Ng, a renowned computer scientist known for his work in AI and the field of deep learning.
Wu, an adjunct professor at Stanford University's computer science department, praised China's open AI ecosystem, where companies compete against each other in a 'Darwinian life-or-death struggle' to advance foundational models. In a post published on DeepLearning.AI, the education platform he co-founded, Wu noted that the world's top proprietary models were still from frontier US labs, while the top open models were mostly from China.
Chinese companies have been launching open-source models in quick succession in recent weeks. Alibaba Group Holding and Zhipu AI rolled out their latest reasoning and video models this past week.
Alibaba claimed its Wan 2.2 video tool was the industry's 'first open-source video generation models incorporating the Mixture-of-Experts (MoE) architecture' to help users unleash film-level creativity. Alibaba owns the South China Morning Post.
Crowds seen in front of the Zhipu AI booth during the World Artificial Intelligence Conference in Shanghai. Photo: Handout
Zhipu boasted its GLM-4.5 as China's 'most advanced open-source MoE model', as it secured third place globally and first place among both domestic and open-source models based on the average score across '12 representative benchmarks'.