DeepSeek's latest AI model, R1-0528, is noted for its high scores in coding, math, and general knowledge, rivaling OpenAI's o3. However, testing indicates R1-0528 is much less willing to address contentious subjects, particularly those considered sensitive by the Chinese government. Reports suggest it is the most censored model from DeepSeek, using restrictions on criticisms of government actions. This aligns with China's strict legal framework governing AI content, which demands adherence to government narratives, leading to significant limitations on topics such as the internment camps in Xinjiang.
The newer R1-0528 model shows remarkable performance on benchmarks, yet it exhibits heightened censorship, particularly concerning sensitive subjects related to the Chinese government.
R1-0528 has been judged to be the most censored of DeepSeek's models, significantly limiting responses on free speech and critical topics about China.
Collection
[
|
...
]