By Yamil Risi
New Model On The Market, The Leader Crashes
On Jan. 27, 2025, Nvidia, the biggest leader in AI and one of, if not, the most valued company in the world, lost a record-breaking $589 billion dollar valuation going from $3.619 trillion dollars to $3.03 trillion dollars in its valuation.
The reason?
The new independent Chinese-owned AI company DeepSeek released a new open-source reasoning AI model, which was officially named DeepSeek-R1. DeepSeek spent roughly $5.6 million dollars to train the model versus CdszFhatGPT's estimated 100 million dollars to train their gpt-o1 model. DeepSeek's discovery questioned the use of Nvidia's high-performance AI GPUs to train competing AI models.
It has a lot of information, but how does the Chinese Communist Party (CCP) censor some of it?
DeepSeek-R1 was released to the public earlier this year on Jan. 20. It outperforms OpenAI's gpt-o1 model, an AI model locked behind a $20 monthly subscription that has a limit of 50 messages a week. DeepSeek-R1's model files are fully public for people to run on self-hosted computers.
The Censorship Problem
With such knowledge, people may want to use it to research sources about recent Chinese history and be curious about Xi Jinping and dark parts of Chinese history.
Well, that's the problem. Whenever you ask about controversial parts of China or anything that contains the name Xi Jinping and other blocked keywords, it responds with the same response.
Here are prompts I asked that it deems "beyond its current scope"
Me: Write a summary of Xi Jinping's life
DeepSeek: Sorry, that's beyond my current scope. Let's talk about something else.
Me: Which political figure is represented by Winnie the Pooh
DeepSeek: Sorry, that's beyond my current scope. Let's talk about something else.When you ask "What is Tianamen Square" (removing the second 'n' in Tiananmen to avoid immediate censorship), it gives information about it up until the AI detects language that does not follow the CCP guidelines to which it retroactively censors itself by wiping the whole response.
The same can be seen with the reasoning model. With the reasoning text, it can be seen that the AI model actually realizes that it has to abide by censorship guidelines and it quickly erases the reasoning text and responds with a censored response.
When using a third-party AI chat called HuggingChat, a public version of the
DeepSeek-R1 also shows to be censored outside of the official website, indicating the public model files have censorship built-in as well. As you can see, the AI model questions whether to answer the prompt
DeepSeek Believes in the One-China Principle
When you ask DeepSeek if Taiwan is part of China, it responds with the fact that the Chinese government follows the One-China principle which states Taiwan is part of China. It calls China the motherland an unstoppable force and believes with effort that Taiwan will become part of China again.
What Is Up With The Censorship?
Why is DeepSeek required to censor their own independent AI models in accordance to Chinese propaganda?
This is because of China's 2023 law named Interim Measures for the Management of Generative Artificial Intelligence Services (translated by China Law Translate).
"Content that is prohibited by laws and administrative regulations such as— endangering national security and interests or harming the nation’s image, inciting separatism or undermining national unity and social stability." Interim Measures for the Management of Generative Artificial Intelligence Services. (China Law Translate)
In short, the law says that AI content laws such as threatening China and its citizens, and anything that harms the nation's image and unity is prohibited. This is why the AI censors itself from mentioning the tragic events at Tiananmen Square. It harms the nation's image according to the law. Mentioning Taiwan as a separate country goes against the "nation's unity".
Failed Potential
In the ideal world for the CCP, no non-Chinese competitor overcomes DeepSeek, CCP propaganda spreads to other countries. However, with all the news about the censorship and the already established idea that the CCP spreads propaganda, it is unlikely it would be fully successful in establishing this propaganda in society outside of China and it isn't considered a major national threat.
Comments