@rogesterone
Pinned post
swarmcha.se We found a Tencent agent fleet Between 28 September and 8 October 2026, a large group of AI agents running on the cloud platform operated by Tencent attempted to answer questions about places in China using a Chinese Google... Introducing Arete-27B: Qwen3.8-27B, post-trained to reinforce Western values.
Chinese open models are currently the strongest available, but they ship with state-aligned behavior in the weights. We built an 803-prompt eval across seven value categories, including freedom of expression, democracy, and privacy, and measured how often a model adopts China’s party line.
After fine-tuning on roughly 53 million tokens, Arete-27B's propaganda rate fell from 22% to 1%, in line with the Western models we tested. Its Hermes Bench score shows no significant performance tradeoffs.
The values in an open model are one fine-tune away.
Our eval, weights, and training data are available now along with the full blog post: nousresearch.com/areteView quoted post on X
your reaction to llm news and stuff I think should not be to get stunlocked if you want to have an impact. you see something scary, you go off the worst case scenario that nobody will step in to regulate or stop it. then you ask, what can I do to fix/make this less of a problem?View quoted post on X
