Nous Research unveils Arete-27B, a post-trained model built to reduce Chinese AI propaganda
This story is only covered by news sources that have yet to be evaluated by the independent media monitoring agencies we use to assess the quality and reliability of news outlets on our platform. Learn more here.8 Articles
8 Articles
Arete-27B's Reported Framing Fix Is Nous Grading Nous » Saipien
A lever you cannot certify yet If a local model writes your country briefs, compliance summaries, or sales research, its framing is already a product decision. You either chose that point of view or you inherited one. Nous Research says it has a control for one inherited version of that problem: Arete-27B, a supervised fine-tune […]
Nous Research has fine-tuned Alibaba's Qwen3.8-27B and developed the Arete-27B. The propaganda rate dropped from 22 percent to 0.9 percent, overall talent was preserved.
Nous Research ensures that a fine tuning of Qwen3.8-27B reduced from 22% to 0.9% the proportion of answers that its test classified as Chinese state propaganda, with no noticeable changes in an internal evaluation of general capabilities. The experiment, published along with weights and data, opens questions about what really measures that bias and how far the result can generalize.
State propaganda in some Chinese AI models is only "a thin layer" that fine-tuning can strip away
Nous Research retrained Alibaba's Qwen3.8-27B to produce Arete-27B, largely stripping out its pro-Chinese bias. The propaganda rate dropped from 22 percent to less than one percent, while general capabilities remained intact in tests. The method relied on 53 million tokens of supervised fine-tuning. The political bias is "applied late, it is applied thin," the team writes. The article State propaganda in some Chinese AI models is only "a thin la…
Nous Research has trained with Arete-27B Alibaba's Qwen3.8-27B in such a way that the pro-Chinese coinage largely disappears: The propaganda rate decreased from 22 to less than one percent, the general skills remained according to tests. For this, 53 million tokens of monitored fine tuning were sufficient. The political imprint was only a "thin, late applied layer". The article Fine tuning against propaganda: US start-up trains Alibaba's Qwen mo…
Coverage Details
Bias Distribution
- There is no tracked Bias information for the sources covering this story.
Factuality
To view factuality data please Upgrade to Premium
