Chinese open-source AI models have been in the news recently for their superior performance in various AI tasks such as coding and “reasoning.”
However, it has also drawn criticism, including from OpenAI employees, for censoring topics sensitive to the Chinese government, such as the Tiananmen Square massacre.
Clement DeLang, CEO of Hugging Face, said he has similar concerns. In a recent podcast (in French), he warned of the unintended consequences of Western companies building on top of high-performance open source Chinese AI.
“If you create a chatbot and ask people questions about the Tiananmen Square incident, it's not going to respond in the same way as a system developed in France or the United States,” Delang warned.
DeLang noted that if countries like China become “by far the most powerful in terms of AI, they will probably be able to spread certain cultural aspects that the West doesn't want to spread.” .
DeLang previously said that Chinese AI is rapidly catching up with Western AI thanks to its embrace of the open source movement.
DeLang warned on the podcast that the concentration of top open source models from China is “a fairly new development, and I'm a little concerned, to be honest.” “It's important that AI is distributed across all countries. It's important that there aren't one or two countries that are much more powerful than others.”
HuggingFace is the world's largest platform for AI models and a popular place for Chinese AI companies to showcase their latest LLMs. In fact, HuggingFace's CTO announced this week that HuggingChat's default model is Qwen2.5-72B-Instruct, developed by Chinese e-commerce giant Alibaba.
This particular model does not appear to censor questions about the Tiananmen Square massacre or other issues that the Chinese government typically censors.
However, a different model from Alibaba's Qwen family, the QwQ-32B, available on HuggingChat, clearly works, as TC asked:
TechCrunch previously reported that DeepSeek, another Chinese model that has become a hot topic in the AI community for its inference capabilities, also extensively censors topics deemed sensitive by the Chinese government.
Chinese AI companies are in a difficult position as the Chinese government forces their models to “embody core socialist values” and comply with an already extensive censorship regime.
A Hugging Face spokesperson declined to comment further, but noted that DeLang recently predicted that China would begin leading the global AI race by 2025.