Pinned post
This is who you're being mean to when you're mean to me online btwView quoted post on X
Guy
Verified account This account is verified. Learn more@nosilverv
In my emotional era…
811 Following 35K Followers 132K Posts
Latest posts 5
"Appeal To Social Consensus Face"
1:16
I find it hilarious that Daniela does Appeal To Social Consensus Face on the concept that _anyone_ at *Anthropic* (!) is a transhumanistView quoted post on X
When loyalty matters most selecting exclusively, or even primarily, for competence is an own-goal: you don't want someone who might turn around and betray you and you ESPECIALLY don't want that if they're competent. Hence: politics.
Everybody always thinks they're not understood because they lack logos, pathos, and ethos, in that order when, in reality, the order is probably reversed.
Open source propaganda wars LES GOOOO
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
