Two OpenAI infrastructure and workflow posts anchor today’s list: how the company built a Git-compatible backend that can handle the scale of development Codex drives, and how engineers are applying ChatGPT and Codex to CAD, PCB design, and simulation. Meta contributes a deep dive on moving ads ML features to production at scale, and Simon Willison reproduces a paper finding showing an on-laptop Qwen model can read and interpret neural network weights. Beyond AI, a Linux security tool’s update channel was hijacked to backdoor customers, a password manager is telling every user to rotate credentials, and Dimitri Fontaine turns ten years of logical-replication progress into a working multi-writer Postgres setup.
From the labs
- Scaling Git at OpenAI — OpenAI
- Accelerating engineering for the physical world — OpenAI
- F3 Lightning: How Meta Accelerates Ads ML Features to Production at Scale — Meta Engineering
Worth your time
- Qwen 3.8 27B on my laptop can interpret model weights, too — Simon Willison
Beyond AI
- cPGuard security advisory: Update channel compromise — OPSSHIELD
- mSecure Warns Users to Change Passwords, Reset 2FA and Passkeys After Suspicious Activity — The Hacker News
- Postgres logical replication beyond one writer — Dimitri Fontaine