There is an idea that floats around Dibsic is well documented Censorship It is only in its application layer but goes away if you operate it locally (this means downloading the AI on your computer).
But Dibsic control is bread, according to Wireless Which found that the model is subject to control over both the levels of application and training.
For example, a local version of Deepseek revealed to WIRED thanks to the thinking feature that should “avoid mentioning” events such as the cultural revolution and focus only on the “positive” aspects of the Chinese Communist Party.
Quick examination by techcrunch to make a local operation from Deepseek Available via Groq He also showed clear censorship: Deepseek answered happily to a question about Kent fire In the United States, but I answered “I cannot answer” when she asked her What happened? In Tiannmen Square in 1989.
https://techcrunch.com/wp-content/uploads/2025/01/GettyImages-2195739346.jpg?resize=1200,800
Source link