HOW PRO-CHINA ARE CHINESE MODELS?

Summary from the TLDR Dev newsletter

Vals AI tested politically sensitive questions across languages and contexts, finding that Chinese models sometimes criticize China's government while non-Chinese models sometimes echo official narratives. Responses generally leaned more pro-China in Simplified Chinese, while deployment prompts, response formats, and agent harnesses could substantially change the results.