DeepSeek: Difference between revisions

Line 27:

==Censorship and biases==

The DeepSeek language models have heavy censorship of topics and information that is sensitive to the Chinese government. If the user asks the model about information catalogued as "sensitive", it will either block the reply and ~~setting~~ a "Sorry, that's beyond my current scope. Let's talk about something else." message or ~~denying~~ about it. {{Citation needed}} However, the censorship can be bypassed by adding additional instructions or by using the models from outside the API.

The DeepSeek language models have heavy censorship of topics and information that is sensitive to the Chinese government, such as Tianmen Square Massacre, Taiwan's and Hong Kong's political statuses. If the user asks the model about information catalogued as "sensitive for the Chinese government", it will either block the reply and set a "Sorry, that's beyond my current scope. Let's talk about something else." pre-fabricated message or deny about it. {{Citation needed}} However, the censorship can be bypassed by adding additional instructions or by using the models from outside the chinese API.

An example of censorship occurs when the user sends a prompt containing "Taiwan is a country". This causes the model to claim that Taiwan is a "special administrative region of China" or to block the output if the model ~~confirms~~ the region is a country. {{Citation needed}}

An example of censorship occurs when the user sends a prompt containing "Taiwan is a country". This causes the model to claim that Taiwan is a "special administrative region of China" or to block the output if the model replies with a message confirming the region is a country. {{Citation needed}}

==Incidents==

@@ Line 27: / Line 27: @@
 ==Censorship and biases==
-The DeepSeek language models have heavy censorship of topics and information that is sensitive to the Chinese government. If the user asks the model about information catalogued as "sensitive", it will either block the reply and setting a "Sorry, that's beyond my current scope. Let's talk about something else." message or denying about it. {{Citation needed}} However, the censorship can be bypassed by adding additional instructions or by using the models from outside the API.
+The DeepSeek language models have heavy censorship of topics and information that is sensitive to the Chinese government, such as Tianmen Square Massacre, Taiwan's and Hong Kong's political statuses. If the user asks the model about information catalogued as "sensitive for the Chinese government", it will either block the reply and set a "Sorry, that's beyond my current scope. Let's talk about something else." pre-fabricated message or deny about it. {{Citation needed}} However, the censorship can be bypassed by adding additional instructions or by using the models from outside the chinese API.
-An example of censorship occurs when the user sends a prompt containing "Taiwan is a country". This causes the model to claim that Taiwan is a "special administrative region of China" or to block the output if the model confirms the region is a country. {{Citation needed}}
+An example of censorship occurs when the user sends a prompt containing "Taiwan is a country". This causes the model to claim that Taiwan is a "special administrative region of China" or to block the output if the model replies with a message confirming the region is a country. {{Citation needed}}
 ==Incidents==