Archive — US — techcrunch
DeepSeek’s updated R1 AI model is more censored, test finds
DeepSeek's R1-0528 AI model excels in coding and general knowledge but shows significant censorship of topics sensitive to the Chinese government. Testing indicates it's the most restricted version regarding criticism of China, complying with strict local laws on content. Comparatively, it performs well against benchmarks but avoids contentious subjects like Xinjiang's internment camps.
Key points
- DeepSeek's updated R1-0528 model excels in benchmarks for coding and general knowledge.
- Testing shows R1-0528 is more censored than previous versions regarding contentious topics.
- It complies with Chinese laws that restrict discussion on politically sensitive matters.
- While it sometimes criticizes Chinese policies, it tends to align with the government's official viewpoint.
Why it matters
- This highlights ongoing concerns around AI censorship in China and the balance between technology and state control.
- The trend reflects broader issues in AI development regarding ethical considerations and compliance with local laws.
- Models like R1-0528 showcase the impact of government regulations on technological advancements and free speech.
The developments surrounding DeepSeek’s R1-0528 underscore the ongoing tension between AI innovation and governmental censorship, raising critical questions about the future of free expression in technology.