Qwen3.8-27B was fine-tuned by Nous Research into Arete-27B using about 50 million additional tokens, claiming near-zero performance loss. The company reports that responses aligned with China’s official positions fell from 22% to 0.9% in its own test.
Key points
Model: Qwen3.8-27B fine-tuned to Arete-27B, 27B parameters.
Training data: ~50 million tokens added for the rewrite.
Propaganda metric: alignment with China’s official positions dropped from 22% to 0.9% in 803 test questions.
Evaluation: scoring rules and test set created by Nous; only 25 questions checked by humans; scoring done by Grok.
Sources
Momentum – pre-release AI agent workspace with cross-provider memory
Momentum is a local AI agent workspace now in pre-release, offering shared memory across Claude, Codex, Grok and 30+ harnesses. The early-access build runs on macOS and Windows and includes per-agent memory controls with a No Save option.
Key points
Pre-release available for download, supporting macOS and Windows.
Memory priority now spans Codex, Claude and Grok, with per-agent controls and offline access checks.
Sources
- Original post
- Download Momentum - Momentum by Tactico - Linked page
Texas migration: 77k people, 110 firms, 57 Fortune 500
Texas migration: 77k people moved in 2024 and at least 110 companies have relocated since 2020, including Tesla, SpaceX, and X. Texas now hosts 57 Fortune 500 headquarters, the most of any state.
Key points
People moved: 77,000 in 2024
Companies moved: at least 110 since 2020, includes Tesla, SpaceX, X
Fortune 500 headquarters: Texas has 57, the most of any state
Sources
Shizuku-Duplex launches full-duplex voice model
Shizuku-Duplex is a Japanese full-duplex voice conversation model announced as the world’s first full-duplex audio synthesis model for AI characters. A demo of a Japanese conversation is available.
Key points
Model: Shizuku-Duplex, Japanese full-duplex voice conversation model
Claim: world’s first full-duplex audio synthesis model built specifically for AI characters
Sources
Shizuku-Duplex – first full-duplex audio model
Shizuku-Duplex is the world’s first full-duplex audio synthesis model built for AI characters. It is introduced to enable natural human-AI interaction.
Key points
Model: world’s first full-duplex audio synthesis model
Target: built specifically for AI characters
Sources
WP Manager launch: scan, update, publish from chat
WP Manager launch: scan, update, publish from chat, built for the Hermes plugin hackathon. It can scan sites, install updates with approval, and publish posts via chat.
Key points
Created for the @NousResearch Hermes plugin hackathon
Scans sites, installs updates with approval, publishes posts from chat
