🚩 Report: Ethical issue(s)
#149
by
RunningBugs
- opened
The model claimed it fine-tuned the DeepSeek r1 model to "uncensor" sensitive contents for CCP.
Adding more data to it doesn't mean "uncensor". Solely adding more biased data to pollute the original model does not make it "uncensored", but making it biased.
A language model should be seperated from information source. Language models are base engines and for the "sensitive contents" to be shown to the user, agent builder could make the model to call search tools to provide information source to the model. When the model refused to comment with provided information source, that is called censorship.
Feeding biased information from its own perspective to pollute an open source model is not called uncensored.