Headline: The Rise of Chinese AI Models: An Analysis of Strengths, Controversies, and Future Implications
China’s artificial intelligence (AI) models have been grabbing headlines recently due to their impressive performance in numerous AI tasks such as coding and logical reasoning. However, these models aren’t without their share of controversies, with critics, including OpenAI staff, accusing them of censoring topics that are considered sensitive by the Chinese government, such as the Tiananmen Square incident.
In a recent French podcast, Clement Delangue, the CEO of Hugging Face, an AI model platform, expressed his concerns about the implications of Western companies utilizing these high-performing, open-source Chinese AI models. He cautioned that a chatbot developed using such models could exhibit biased responses to questions about topics like Tiananmen Square, unlike systems developed in Western countries like France or the U.S.
Delangue further highlighted the potential cultural influence that could ensue if China becomes the dominant force in AI technology. He has previously acknowledged China’s rapid progress in AI, attributing it to the country’s adoption of the open-source movement.
In the podcast, Delangue expressed his apprehensions about the increasing prevalence of top-performing open-source models originating from China. He emphasized the need for AI technology to be evenly distributed across all countries to avoid dominance by one or two nations.
Hugging Face is a globally renowned platform for AI models and a favorite among Chinese AI firms to display their latest Language Model developments. Recently, Hugging Face’s CTO announced the default model on HuggingChat as Qwen2.5-72B-Instruct, a model developed by Chinese e-commerce behemoth, Alibaba. Interestingly, this model doesn’t seem to censor queries on issues such as the Tiananmen Square incident, which are generally censored by the Chinese government.
However, another model from Alibaba’s Qwen series, QwQ-32B, available on HuggingChat, does appear to censor such topics. DeepSeek, yet another Chinese model lauded for its reasoning capabilities, also extensively censors topics deemed sensitive by the Chinese government.
Chinese AI firms are walking a tightrope, with the government insisting their models align with “core socialist values” and adhere to its extensive censorship mechanism. Hugging Face declined to comment further, however, Delangue has previously predicted that China is set to take the lead in the global AI race by 2025.
