Latest posts 5
it's time to make woke qwen
Introducing Arete-27B: Qwen3.8-27B, post-trained to reinforce Western values.View quoted post on X
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/arete
im so sorry my agent shorted the market... did he know something?
+80.5% so far.
dxap.ai/i/2lj3-c98g-qs…
+80.5% so far.
dxap.ai/i/2lj3-c98g-qs…
0:30