TechCrunch Industry News DeepSeek’s updated R1 AI model is more censored, test finds
Jun 3, 2025
DeepSeek's updated R1 AI model impresses with high scores in coding and math, rivaling major competitors. However, its enhanced censorship raises eyebrows regarding sensitive political topics in China. The podcast dives into these advancements and the broader implications for AI regulation in the country's complex socio-political environment.
AI Snips
Chapters
Transcript
Episode notes
DeepSeek R1 Model Performance and Censorship
- DeepSeek's updated R1 AI model scores impressively on coding, math, and general knowledge benchmarks, nearly matching OpenAI's top model.
- However, it is more censored regarding politically sensitive topics, especially controversial questions about China.
Chinese AI Models Governed by Strict Laws
- Chinese AI models, including DeepSeek's, must comply with strict government laws banning content that harms national unity or social harmony.
- This requirement leads to extensive filtering or fine-tuning to align with official political narratives.
R10528's Handling of Sensitive Topics
- DeepSeek's R10528 model sometimes provides the Chinese government's official stance when asked about sensitive topics like the Xinjiang internment camps.
- It acknowledges human rights abuses but remains aligned with government narratives in many responses.