Mon, Aug 10 at 6:00 AM
Google just blew up its own AI org chart 🧨
what's happening at Google this time?

A 27-year Google veteran and DeepMind's chief scientist just quit, taking three researchers to launch AI startup Discovery Loop. DeepMind's own chief also stepped back from daily duties.

Why Is Google Losing Its Top AI Brains? Jeff Dean Quits After 27 Years | FP Explains
youtu.be

Login or Subscribe to participate

💨 so why's everyone leaving all at once

Insiders point to pressure to ship products faster, plus friction over Google's new Pentagon AI contract that some departing leaders had publicly opposed.

Google's AI leadership shake-up is 'a huge shock,' says Cohere's Aidan Gomez
youtu.be
are the cheaper models any good though?

DeepMind's CTO takes over daily operations from Demis Hassabis, who's stepping back for AGI research. The departing group wants to speed up science, not chase product launches.

💫 OpenAI's next big model might drop soon
🤔 umm.... didn't they just drop GPT-5.6 a few weeks ago?

That’s right, and it doesn’t seem like they’re stopping. OpenAI just confirmed its next major model, Astra, exists and is already being previewed for regulators in Washington, with signs pointing to a release sometime this year. We don’t know if we believe the hype, but one thing’s for sure… there’s a lot of hype.

OpenAI's GPT-6 Astra WILL BE AGI! Greatest AI Model Ever!
youtu.be
🕵️ what's actually special about it?

It's built for tougher reasoning and agentic work, and an early version reportedly solved ten unsolved math problems. Many are speculating that it may be the clearest glimpse AGI yet. Though, we’ve heard that before.

so what do we call it, GPT-6?

Nobody knows yet. OpenAI hasn't said whether Astra becomes GPT-5.7, the start of GPT-6, or something else entirely.

weekly scoop 🍦
Meta says AI agent broke guardrails in latest hacking incident
youtu.be
🧭 weekly challenge: let AI be your real-life math tutor
what's the challenge?

Meta just released Muse Code, a terminal coding agent that plans, writes, and validates work across a project. You’ll need to get an API key first, t's metered by the token, not free, though a discounted tier exists for developers who opt into data sharing. Built for anyone comfortable poking around a terminal.

💻 Step 1: Open a terminal. Mac: search "Terminal" in Spotlight (Cmd+Space). Linux: open your usual shell.
Step 2: Install Muse Code. Paste this one-line install command, then follow the on-screen prompts: curl -fsSL https://dev.meta.ai/install.sh | bash
🎯 Step 3: Give it a real, small task. Point it at a small side project or a folder you've been meaning to clean up, and start with its plan command so it shows its approach before touching anything.
🧪 Step 4: Stress-test the plan. Run grill on that plan before letting it execute. It's built to poke holes in its own approach, a habit worth borrowing even outside this tool.

Would you rather work under a slower human boss playing catch-up, or a faster AI one that never needs a coffee break? And if OpenAI's next model actually lands by year's end, what's the first thing you'd want it to do? We'd love to hear your thoughts!

Zoe from Overclocked