Nous Fine-Tunes Qwen3.8-27B to Remove Propaganda Content, Releases Arete-27B
Understand the story
Nous Research fine-tuned Alibaba's Qwen3.8-27B model to remove propaganda content from the model's answers. On November 7, 2026, Nous Research released the open-source model Arete-27B: it was trained with supervised fine-tuning (SFT) on 53 million tokens of data for two epochs, without reinforcement learning. According to The Decoder, after fine-tuning, the proportion of propaganda content in the model's answers dropped from 22% to 0.9%, and the factual accuracy score rose from 2.32 to 2.98.
AI overview of reporting · Updated 3 h前
Reporting timeline
Explore different aspects of the story through its reporting.
- The DecoderState 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 (two epochs of SFT, no reinforcement learning) and released the open-source model Arete-27B, cutting propaganda content from 22% to 0.9% and raising factual accuracy scores from 2.32 to 2.98.
Story heat trend
Current heat9·Comparable peak10（11/7 21:00）·Comparable change over 24 hours –
The trend compares the same continuously observed sources, which may cover fewer sources than the current heat score. Hover or select the chart to inspect each hour; use the left and right arrow keys to navigate.