CEO of AI development platform says high-performance AI model development in China is 'highly likely to lead to political censorship'



As AI development accelerates worldwide, China's open source AI models have demonstrated excellent performance in various AI tasks such as coding and inference. However, Clement Delange, CEO of Hugging Face , a platform also known as the 'GitHub for AI development,' pointed out that developing based on China's high-performance open source AI may result in unexpected defects due to censorship.

Hugging Face CEO has concerns about Chinese open source AI models | TechCrunch
https://techcrunch.com/2024/12/03/huggingface-ceo-has-concerns-about-chinese-open-source-ai-models/



Western AI Firms Are Worried About 'Censorship' In China's AI Models | CCN.com
https://www.ccn.com/news/technology/western-ai-firms-censorship-china-model/

In a post on LinkedIn, a business-focused social networking site, De Lang made six predictions about AI in 2025. His predictions included 'the first large-scale protests regarding AI will take place' and 'at least 100,000 personal AI robots will be pre-ordered,' as well as China's growth, saying, 'As a result of leading the open source AI race, China will begin to lead the global AI race in 2025.'

In fact, as of 2024, Chinese AI companies are at the forefront of AI development. DeepSeek, founded in 2023 with the aim of developing general artificial intelligence (AGI) , announced the large-scale language model 'DeepSeek-R1-Lite-Preview' in November 2024, which is said to be comparable to and in some cases superior to OpenAI's inference model 'o1-preview'. R1-Lite-Preview is also scheduled to release an open source version and API, which may become the basis for future AI development. In addition, Chinese AI companies such as Alibaba and Moonshot AI are making rapid progress in the AI field by focusing on open source development.

Chinese AI company DeepSeek releases 'DeepSeek-R1-Lite-Preview', an inference AI model comparable to OpenAI o1, with plans to open source it - GIGAZINE



However, in tests conducted by technology media TechCrunch, R1-Lite-Preview refused to answer questions about President Xi Jinping, Tiananmen Square, and the geopolitical implications of China invading Taiwan. TechCrunch pointed out that the Chinese-developed AI model is likely to be extensively censoring 'topics that the Chinese government considers sensitive.'

Delange expressed similar concerns in a podcast he hosts. 'If you ask a chatbot based on a Chinese AI model about the Tiananmen Square massacre, it's not going to respond in the same way as a system developed in France or the US,' he said. 'Conversely, if a country like China becomes so advanced in AI, it's going to be able to promote certain cultural aspects that Western countries might not want to see promoted.'

Other major AI companies have also voiced their opposition to Chinese AI models. In response to DeepSeek's refusal to respond to the Tiananmen Square incident, OpenAI researcher Steven Heidel posted a meme image on X criticizing China's AI censorship. OpenAI CEO Sam Altman also expressed concern in 2023, saying, 'We are very concerned about authoritarian governments developing AI models.'



According to De Lang, the problem is that the top open source models are concentrated in China and other countries, because AI models contain a certain amount of ideology from the country in which they were developed. 'It's important to avoid a situation where AI is distributed across all countries, that is, where there are one or two countries that are much stronger than the rest,' De Lang said.

Related Posts:

in Software, Posted by log1e_dh