Nous Research has released Arete-27B, a model retrained on the basis of Qwen. According to XenoSpectrum, it significantly reduces Chinese official-line answers.
What the source supports
XenoSpectrum reports the release and the sharp reduction in official answer patterns. The source cites no measurements or comparison figures beyond this claim. The original page could not be retrieved for verification; the statements rest on the feed excerpt.
Context
For companies using freely available base models, the case may mean that a model's behaviour and political orientation can be changed through retraining. Whether and how this affects specific applications cannot be derived from the source.
Sources
Editorially owned by Ideal Syka. Sources and method: Newsroom & method. Tips and corrections: ai@i6eal.de.
