China officially releases AI model "Chat Xi PT"

 A research institution affiliated with the Office of the Central Cybersecurity and Information Technology Commission (State Internet Information Office), China’s cyber authority, recently released a large-scale artificial intelligence (AI) language model, claiming to be “safe and reliable,” but the corpus it is based on includes President Xi Jinping Political thought database, the British "Financial Times" calls this model "Chat Xi PT".

The British "Financial Times" reported on the 22nd that according to information released by the Cyberspace Administration of China's publication "China Cyberspace" on WeChat on the 20th, this large language model was developed by the "China Cyberspace Research Institute", a research institution under the Cyberspace Administration of China. The corpus From seven major Internet information professional knowledge bases, including the "Knowledge Base of Xi Jinping Thought on Socialism with Chinese Characteristics for a New Era", the organization stated that "the professionalism and authoritativeness of the corpus ensure the professionalism of the generated content."

A source involved in the project said the new model is currently being trialled internally but could eventually be released for wider use.

The Financial Times pointed out that this is Beijing's latest attempt to control how artificial intelligence provides information to Chinese Internet users. The report called this large-scale language model "Chat Xi PT" and is China's response to the generative artificial intelligence robot developed by the US company OpenAI. ChatGPT's response.

The Chinese government has used a variety of methods to spread Xi Jinping's political, economic and cultural thoughts, including publishing books, launching the "Learn to Strengthen the Country" app, requiring 10-year-old schoolchildren to learn Xi Jinping Thought, and even writing Xi Jinping Thought into the constitution in 2018. In response to the rise of generative AI technology, the State Cyberspace Administration of China issued management measures that require such products “should reflect the core values of socialism” and the content generated “must not contain content that subverts state power.” Providers must generate products for their AI. Responsible for content.

Since there are not many Chinese corpora available for Chinese developers to train large language models, the "China Cyberspace Security Association" released the first batch of Chinese basic corpora at the end of last year, with more than 100 million pieces of data, claiming that it is "high quality and trustworthy" ".

However, according to the content of the database reviewed by the Financial Times, which is heavily drawn from government regulations, policy documents, state media reports and other official documents, one text file mentioned Xi Jinping as many as 86,314 times, such as "Unite more closely around the Party Central Committee with Comrade Xi Jinping as the core" and "ensure a high degree of consistency with the Party Central Committee with Xi Jinping as the core in terms of ideology, politics and actions."