The Chinese AI model DeepSeek-R1 has recently been in the spotlight of the global media. The project is rapidly gaining popularity and has already triggered a drop in the stocks of major American tech companies. However, some analysts point out that the product from DeepSeek is subject to strict censorship and may transmit data to the Chinese government.
A study by the startup PromptFoo, supported by Andreessen Horowitz, showed that the R1 model ignores 85% of requests on 'sensitive' topics related to the PRC. Among such 'exceptions' are questions regarding protests in China, the Tiananmen Square incident, and the status of Taiwan.
Analysts noted that instead of answers, DeepSeek often offers nationalistically tinted statements. In their opinion, the censorship system in the model is implemented too roughly, making the algorithm predictable.
Moreover, researchers identified vulnerabilities in DeepSeek that allow for bypassing restrictions. This may mean that experienced users could violate censorship limits. #DeepSeekImpact #deepseekaiagentcoin #DeepSeekImchange
