China's AI Models Close Gap with US to Months, Talent Network and Compute Efficiency Key

According to the Wall Street Journal, China's AI models have rapidly narrowed the gap with top US models, driven by a talent network from universities like Tsinghua, open-source technology, returnees, and better compute efficiency. Key researchers include Tang Jie (Zhipu AI), Yang Zhilin (Moonshot AI), and Liang Wenfeng (DeepSeek).

Industry insiders say the gap is now just months. In June, Musk predicted China would catch up by Q1 2027, but Tang Jie said it won't take that long. The talent system dates back to 2005 with Yao Qizhi's 'Yao Class' at Tsinghua. Tang's lab trained many, and Yang studied under him before returning to found Moonshot.

Chinese firms face funding and chip shortages, spending less than a fifth of US peers. DeepSeek uses MLA and MoE to boost efficiency, and its low-cost open-source model shook markets. Talent return, like Tencent hiring ex-OpenAI staff, also helps. But chips remain key, as Liang Wenfeng says the chip gap is the biggest challenge.