The DecoderNov 7, 2026 20:32

Nous Research Fine-Tunes Qwen3.8-27B to Reduce Pro-Chinese Bias

Nous Research retrained Alibaba's Qwen3.8-27B with 53 million tokens of supervised fine-tuning to create Arete-27B. According to its tests, the model's propaganda rate fell from 22 percent to below one percent while general capabilities remained intact, suggesting the political bias was added as a relatively thin, late-stage layer.

AI Brief

The work provides a concrete example of targeted fine-tuning substantially reducing political bias without reported capability loss.

A snow-covered, pointed rocky ridge against a dark night sky, rendered in a bluish, grainy style, with the text “ARETE-27B” in white serif font above it.

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 layer" that fine-tuning can strip away appeared first on The Decoder.

Be the first to rate this article.

Nous Research Fine-Tunes Qwen3.8-27B to Reduce Pro-Chinese Bias

Comments & Questions (0)

No comments yet