Chinese AI Models: Refusing vs. Reframing

How honest are the newest Chinese AI models, really? This episode digs into the measured data — not screenshots — on DeepSeek, Qwen, Kimi, GLM, and Xiaomi's MiMo. The picture has changed: where older models like DeepSeek R1 refused ~85% of China-sensitive prompts, the current generation often answers smoothly while quietly steering toward state-aligned framing. Studies from Aleph Alpha, Anthropic-affiliated researchers, CEIAS, CrowdStrike, and others reveal a shift from visible refusal to fluent reframing — plus code vulnerabilities that spike when prompts mention Tibet or Uyghurs, and English-vs-Mandarin gaps that double the propaganda. We also ask where the censorship actually lives: in the weights, or on the company's servers?

Episode 6068 — myweirdprompts.com/n/6068 (permanent short link: myweirdprompts.com/563914)