A study by AI benchmark firm Vals AI tested 25 models on political questions regarding China, finding that Chinese models respond less uniformly than expected and some Western models occasionally adopt Beijing's official positions. The research measured alignment scores ranging from −1 to +1 and sho