Worthwhile Reads & Interesting Stuff 17

Hello Everyone to this Issue of Worthwhile Reads & Interesting Stuff! Let’s get straight into it.

LLMs & AI

• Introducing Gemma 3n, the new Google LLM with improved efficiency and reasoning, and a detailed dev‑guide for integrating it into your stack.
• V‑JEPA 2: Self‑Supervised Video Models Enable Understanding, Prediction and Planning presents a model for video that self‑supervises to predict future frames and plan tasks.
• OpenAI charges by the minute, so speed up your audio. The discussion highlights how per‑minute billing of OpenAI audio APIs pushes users to optimize for speed.
• MCP: An (Accidentally) Universal Plugin System explores a simple plugin architecture that surprisingly adapts across platforms and languages.
• Wasm‑agents: AI agents running in your browser: Mozilla shows how to run lightweight AI agents entirely client‑side via WebAssembly.
• SmolLM3: smol, multilingual, long‑context reasoner, a lightweight multilingual LLM with extended context capacity, aiming for on‑device reasoning.

Programming & Development

• Writing Code Was Never The Bottleneck argues that communication, architecture, and alignment are the true constraints, not code.

Security & Other

• New Linux udisks flaw lets attackers get root on major Linux distros: critical vulnerability in udisks can be exploited to escalate privileges to root.
• How I Scanned all of GitHub’s “Oops Commits” for Leaked Secrets: a walkthrough of scanning public commits for inadvertently exposed API keys and credentials.
• Your Brain on ChatGPT: MIT Media Lab explores cognitive effects of interacting with ChatGPT over time.
• The Fed says this is a cube of $1 million. They’re off by half a million.: Calvin Shan debunks the Fed’s visual estimate, showing the real volume is 50% larger than claimed.

This was this week’s list of worthwhile reads and interesting stuff, thanks for reading!