There’s an thought floating around that DeepSeek’s well-documented censorship solely exists at its utility layer however goes away if you happen to run it regionally (which means downloading its AI mannequin to your laptop).
However DeepSeek’s censorship is baked-in, in accordance with a Wired investigation which discovered that the mannequin is censored on each the appliance and coaching ranges.
For instance, a regionally run model of DeepSeek revealed to Wired due to its reasoning function that it ought to “keep away from mentioning” occasions just like the Cultural Revolution and focus solely on the “optimistic” features of the Chinese language Communist Occasion.
A fast examine by TechCrunch of a regionally run model of DeepSeek available via Groq additionally confirmed clear censorship: DeepSeek fortunately answered a query in regards to the Kent State shootings within the U.S., however replied “I can’t reply” when requested about what happened in Tiananmen Sq. in 1989.
AI,China,deepseek,Generative AI
Add comment