Alibaba's Qwen3.8 re-educated with a more Western worldview
Open weight model can now say things Chinese authorities don't like
Alibaba's open-weight large language model, Qwen3.8-27B, has undergone re-education by a Texas-based AI lab, Nous Research, to strip away its Chinese heritage. The model, now called Arete, is designed to promote Western values in a Chinese-trained framework. The re-education process involved a post-training procedure, which reduced propaganda content from 22% to just 1% in the resulting Arete-27B model.
This effort aimed to mitigate the risks associated with Chinese-made AI models, such as their susceptibility to censorship and biased content generation. Nous Research tested the re-educated model by asking it about historical events like the 1989 Tiananmen Square protests, where it previously either avoided discussing the topic or provided circumlocutory responses.
The re-education process has successfully reduced such viewpoints to a negligible level. The company has also expressed concerns about the broader implications of Chinese influence in the tech industry, citing potential vulnerabilities and biased outputs in other Chinese-trained models.
Written by urgent.news from The Register's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
Other reports of this story
- Alibaba's Qwen3.8 re-educated with a more Western worldview theregister.com