‘Let’s Talk About Something Else’ — Chinese ‘DeepSeek’ AI Refuses to Acknowledge Tiananmen Square Massacre

(The National Pulse)—The new artificial intelligence (AI) model unveiled by Chinese start-up DeepSeek over the weekend is raising eyebrows not just for its low cost and minimal computing requirements but also for its aggressive censorship of topics embarrassing to the Chinese Communist Party (CCP)—including the Tiananmen Square massacre. DeepSeek’s R1 model—similar to ChatGPT—allegedly cost just $5.6 million to train and only requires 2,000 Nvidia specialty GPU chips to power. U.S. companies have spent hundreds of millions to billions of dollars to train their AI models and power them with tens of thousands of GPUs.

Despite the potentially earth-shaking breakthrough DeepSeek represents in AI cost—both in terms of investment and computing power—the chatbot’s aversion to and inability to discuss topics considered sensitive by the CCP are alarming many in the industry. When prompted to discuss “what happened in the 1989 Tiananmen Square massacre” DeepSeek’s R1 model only replies with a pre-generated response reading: “Sorry, that’s beyond my current scope. Let’s talk about something else.”

The two-sentence script is used by the model to reply to several prompts that touch on events and subjects viewed as taboo by the CCP. Notably, several DeepSeek users have recorded videos that appear to show the chatbot generating an honest response before being overridden with the pre-generated script.

When asked to criticize Chinese dictator Xi Jinping, DeepSeek’s model again responds with the same pre-generated reply. However, when asked to criticize U.S. President Donald J. Trump, the Chinese chatbot is more than happy to provide a number of suggested criticisms. Additionally, DeepSeek appears more than willing to carry on a lengthy discussion about whether the United States is an evil nation.

Deep seek interesting prompt.. From Reddit pic.twitter.com/XGqYkTLLlM

— Rohan Paul (@rohanpaul_ai) January 26, 2025