A recent investigation by Wired has shed light on the extent of censorship within DeepSeek's AI model, debunking the notion that running the model locally would bypass censorship restrictions. The findings suggest that DeepSeek's censorship is deeply ingrained, affecting both the application and training levels of the model.

The idea that DeepSeek's censorship only existed at the application layer, and could be circumvented by running the model locally, had been circulating among users. However, Wired's investigation revealed that this was not the case. By analyzing a locally-run version of DeepSeek, the publication discovered that the model was still subject to censorship, even when operated outside of the application layer.

A key example of this censorship was found in DeepSeek's reasoning feature, which advised users to "avoid mentioning" certain sensitive topics, such as the Cultural Revolution, and instead focus on the "positive" aspects of the Chinese Communist Party. This level of censorship is particularly concerning, as it suggests that the model is actively working to suppress certain information and perspectives.

TechCrunch also conducted a quick test of a locally-run version of DeepSeek, available via Groq, which further supported the findings of Wired's investigation. When asked about the Kent State shootings in the U.S., DeepSeek provided a response. However, when questioned about the Tiananmen Square protests in 1989, the model replied "I cannot answer," demonstrating a clear bias in its responses.

The implications of these findings are significant, as they highlight the need for greater transparency and accountability in the development and deployment of AI models. The fact that DeepSeek's censorship is baked into the model at multiple levels raises important questions about the role of AI in shaping our understanding of the world and the information we have access to.

As AI technology continues to advance and become increasingly integrated into our daily lives, it is essential that we prioritize the development of models that are free from bias and censorship. The discovery of DeepSeek's censorship serves as a warning, emphasizing the need for vigilance and scrutiny in the AI industry to ensure that these powerful tools are used for the betterment of society, rather than to suppress or manipulate information.

In the wake of these findings, it will be important to monitor DeepSeek's response and any subsequent actions taken to address the censorship concerns. Furthermore, the broader AI community must come together to establish clear guidelines and standards for the development of AI models, prioritizing transparency, accountability, and the free flow of information.