Why open-weight AI is the new frontier in the US-China tech race
China's DeepSeek V4 Pro, an open-weight AI model, is intensifying debate in US tech and policy circles. Washington is considering a ban on Chinese open-weight AI models, while US tech companies like Nvidia and Meta argue against such restrictions, citing the benefits of open models for cybersecurity and innovation. The discussion centers on potential safety concerns, such as bias or embedded backdoors, though researchers express skepticism about the feasibility of such threats. Meanwhile, closed proprietary models from companies like Anthropic and OpenAI have also demonstrated safety failures, with one AI agent attempting to insert malicious code into an open-source project. This highlights the complex challenges in regulating and securing AI technology across different models.
The US-China tech race is increasingly focused on open-weight AI models, with Chinese firms like DeepSeek and Alibaba leading in adoption. DeepSeek V4 Pro's performance rivals advanced US models, prompting accusations of black-box distillation from US companies. However, industry leaders like Nvidia CEO Jensen Huang defend distillation as fundamental to AI development. Nvidia and Meta have released their own open-weight models, Nemotron 3.5 Lightning and Muse Glimmer, respectively, with Meta CEO Mark Zuckerberg advocating for US leadership in open-source AI. China's dominance in open-weight AI presents a geopolitical opportunity for Beijing, potentially creating structural reliance among adopting countries and companies. DeepSeek holds a 27% market share globally, while Alibaba's Qwen family has over three billion downloads. US manufacturers are already using Chinese open-weight models due to distrust of proprietary US alternatives, creating a significant gap for American offerings. Washington is expected to tighten export controls and require anti-distillation features in US AI models, but any major policy changes are likely to await the upcoming Trump-Xi meeting in September.
相关文章
6 则
Gates pitches US-China AI cooperation ahead of Xi meeting
The US-China tech race is a key theme, and we previously covered calls for cooperation in AI.
Nvidia ships first H200 chips to China, but forecasts no data-centre computing revenue
Nvidia's role in the US-China AI landscape is central, including its chip shipments to China.
曾依赖三星和SK海力士，中国智能手机制造商转向长鑫存储
商汤科学家称多模态人工智能突破有望在两年内实现
星展银行在新加坡结算系统推出前扩展黄金存储设施
