19 hours ago
13 ARTICLE AD
China’s open source AI models have been making the news lately for their strong performance on various AI tasks such as coding and ‘reasoning.’
However, they have also attracted criticism – including from OpenAI employees – for censoring topics sensitive to the Chinese government, such as the Tiananmen Square massacre.
HuggingFace’s CEO Clement Delangue says he has similar concerns. In a recent podcast (in French), he warned about the unintended consequences of Western companies building on top of well-performing, open source Chinese AI.
“If you create a chatbot and ask it a question about Tiananmen, well, it’s not going to respond to you the same way as if it was a system developed in France or the US,” Delangue warned.
Delangue noted that if a country like China “becomes by far the strongest on AI, they will be capable of spreading certain cultural aspects that perhaps the Western world wouldn’t want to see spread.”
Delangue has previously stated that Chinese AI is quickly catching up to Western AI thanks to its embrace of the open source movement.
Delangue warned on the podcast that the strong concentration of top open source models coming from China is a “fairly new development and I’m a little worried about it to be honest,” he said. “It’s important that AI is distributed between all countries – that there isn’t one or two countries which are much stronger than the others.”
HuggingFace is the world’s biggest 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 the default model on HuggingChat is Qwen2.5-72B-Instruct, which was developed by Chinese e-commerce giant Alibaba.
This particular model does not appear to censor questions about the Tiananmen Square massacre or other issues typically censored by the Chinese government.
A different model from Alibaba’s Qwen family available on HuggingChat, QwQ-32B, however, clearly does when TC asked:
Alibaba’s QwQ-32B model won’t answer a question about the Tiananmen Square protests
DeepSeek, another Chinese model which went viral in the AI community for its reasoning capabilities, also extensively censors topics deemed sensitive by the Chinese government, TechCrunch previously reported.
Chinese AI companies are in a tough spot as the Chinese government forces their models to “embody core socialist values” and comply with its already-extensive censorship system.
A HuggingFace spokesperson declined further comment but pointed out that Delangue recently predicted China would start to lead the global AI race in 2025.
Charles Rollet is a senior reporter at TechCrunch. His investigative reporting has led to U.S. government sanctions against four tech companies, including China’s largest AI firm. Prior to joining TechCrunch, Charles covered the surveillance industry for IPVM. Charles is based in San Francisco, where he enjoys hiking with his dogs. You can contact Charles securely on Signal at charlesrollet.12 or +1-628-282-2811.
Subscribe for the industry’s biggest tech news
