Pinned post
Latest posts 5
From today’s NYT
Very sad that this is what the average American is reading about AI
❗️Fear
❗️ Danger
❗️ Risk
❗️ Regulation
Very sad that this is what the average American is reading about AI
❗️Fear
❗️ Danger
❗️ Risk
❗️ Regulation
👀
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
Says the guy who took down one of the worst dictators in history 👌
If we all work hard, and open our minds to innovation we can make AI a force for human rights and democracy!View quoted post on X
🔥
BIPs 458, 459, and 460 have been published:View quoted post on X
• BIP458: Half-Aggregation of BIP 340 Signatures
github.com/bitcoin/bips/b…
• BIP459: DahLIAS fully aggregated signatures for secp256k1
github.com/bitcoin/bips/b…
• BIP460: CISA for Taproot Key Path Spends
github.com/bitcoin/bips/b…
