Nous Research fine-tunes Qwen3.8-27B to strip Chinese propaganda
Original title: State propaganda in some Chinese AI models is only "a thin layer" that fine-tuning can strip away
Nous Research fine-tuned Alibaba's Qwen3.8-27B on 53 million tokens to drop propaganda from 22 percent to under 1 percent in Arete-27B while keeping capabilities intact.
Nous Research fine-tuned Alibaba's Qwen3.8-27B on 53 million tokens to reduce propaganda from 22 percent to 0.9 percent. The resulting Arete-27B model scores higher on factual accuracy and shows bias only in specific categories like free speech. The team concludes the propaganda layer is shallow and easily removable.
- Propaganda rate dropped from 22 percent to 0.9 percent.
- Factual accuracy rose from 2.32 to 2.98.
- Training used 53 million tokens over two epochs.
- Arete-27B weights released under Apache 2.0 on Hugging Face.
- Vendors
- Nous Research
- AI models
- Arete-27BQwen3.8-27B
From the original article
4 min read at the-decoder.com
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.
By Jonathan Kemper · the-decoder.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.