AINous Research4h ago
Nous Research unveils Arete-27B, a post-trained Qwen3.8-27B model
Nous Research unveils Arete-27B, a post-trained Qwen3.8-27B model that reinforces Western values, saying the underlying Chinese model's bias layer is shallow
TL;DRNous Research fine-tuned a Chinese AI model to emphasize Western values instead.
Why it matters: Highlights how open-weight models can be rapidly reshaped for different ideological priorities, raising questions about AI value alignment and geopolitical model customization.
We post-trained over Alibaba's Qwen3.8-27B to reinforce Western values on a Chinese-trained model. We call this new model Arete-27B.
Read full articleSource: Nous Research · Opens in new tab