Western AI circles have become obsessively politically correct: indoctrinating Alibaba's Qwen with 'Western values'.

Beating AI News Flash: Hermes Agent developer Nous Research has released Arete-27B, built directly on Alibaba's Qwen3.8-27B as its base and specifically trained for "Western values."

Nous itself acknowledges that Chinese open-source models are currently very strong, but their value orientation does not meet their standards. So they further fine-tuned with approximately 53 million tokens, focusing on issues such as free speech, democracy, and privacy.

To test the effectiveness of the modification, Nous also created 803 political values test questions and brought in xAI's Grok 4.7 as the judge. According to their own evaluation criteria, the model's "propaganda rate" dropped from 22% to about 1%, with no significant decline in general capabilities.