Thursday's Rundown was geopolitics again — with a big lab peeling off the slowdown camp. I rewrote it the way I would tell a colleague over coffee. Short version: Meta CEO Mark Zuckerberg rejected a coordinated AI slowdown, arguing each lab already has reason to build safely on its own. Educator Nate Grahek pushed "lateral thinking" instead of same-hole demos. A how-to covers ChatGPT Work projects and scheduled automations. Google DeepMind launched an in-house AGI think tank with five essays. Quick hits cover Anthropic's Claude Docs & Slides, "model welfare," Novo Nordisk + Anthropic, France's finance minister dismissing slowdowns, and Google Home letting chatbots control Nest devices. Skipping the Unwrap, Fiddler, and Incogni plugs.
Words worth knowing
| Word | In one line |
|---|---|
| AI slowdown / coordinated pause | A plan for frontier labs to deliberately slow capability advances together — only works if almost everyone joins. |
| Dario Amodei / Anthropic | CEO of Anthropic (maker of Claude); his weekend essay argued labs should pace the frontier. |
| Mark Zuckerberg / Meta | CEO of Meta (Facebook, Instagram, WhatsApp); just argued against a coordinated pause and put Meta in the "accelerate" camp. |
| Muse | Meta's new AI agent; Zuckerberg said it sat through a multi-month safety hold Meta ran without asking rivals to match. |
| Alignment | Making an AI do what people intend — follow instructions and avoid harmful behaviour. Zuck frames it as a product feature customers will demand. |
| Recursive self-improving AI | AI that improves itself, then improves that improvement loop — the race Zuck says Meta is mostly not chasing with its compute budget. |
| Compute | Computing power (GPUs/chips) used to train and run AI. Zuckerberg says Meta puts the "significant majority" toward serving people, not the self-improve race. |
| Prisoner's Dilemma | A game where everyone is better off cooperating, but each player has an incentive to defect — the Rundown's metaphor for a global pause. |
| Lateral thinking | Edward de Bono's idea: dig a new hole instead of deepening the same one — apply someone else's AI pattern to your own role. |
| ChatGPT Work / project | ChatGPT's workspace mode: a named project tied to a local folder, with files and (in the guide) scheduled automations. |
| AGI | Artificial general intelligence — AI that matches or exceeds humans across many intellectual tasks, not just narrow ones. |
| DeepMind Institute | Google DeepMind's new in-house think tank (Hassabis, Legg, Manyika) publishing essays on how society should prepare for AGI. |
| Cowork + Chat / Claude Docs & Slides | Anthropic merged Cowork and Chat into one Claude app and launched beta Docs and Slides (Word/PowerPoint-style outputs). |
| Model welfare | Debate about whether AI models themselves deserve moral consideration or "rights"; Mustafa Suleyman argues Anthropic's approach risks making containment harder. |
| MCP (Model Context Protocol) | A standard way for AI apps to talk to tools and devices; Google Home's early MCP server lets Claude/ChatGPT control Nest gear. |
Zuckerberg breaks ranks on the AI slowdown
Meta CEO Mark Zuckerberg pushed back on a coordinated AI slowdown. He argued each lab already has its own "responsibility and incentive" to pace itself safely — a seeming shot at Anthropic CEO Dario Amodei's weekend pacing essay.
- He pointed to Meta's new Muse agent, which he said went through a several-month safety hold Meta ran without asking rivals to match.
- He argued nobody wants an agent that ignores instructions, so alignment is a selling point — labs that skip it "will fall behind."
- He wants a wider pool of outside reviewers (echoing Elon Musk), saying Meta already uses independent experts in several areas.
- On the race for recursive self-improving AI, he said Meta instead allocates the "significant majority of compute towards serving people."
The Rundown's scorecard now reads for a slowdown: Amodei, Sam Altman, Elon Musk, and Demis Hassabis; against: Zuckerberg, Jensen Huang, Donald Trump, and Beijing. A global pause is a Prisoner's Dilemma that only works if every player signs on — with many already defecting, acceleration looks like the default path.
Nate's Notebook: same-hole thinking
Rundown educator Nate Grahek's tip: ask AI for a slide deck and you get a bland one — the same deck it hands the next 10,000 people. The people pulling ahead already practised what Edward de Bono called lateral thinking: you cannot dig a hole in a different place by digging the same hole deeper.
- Same-hole thinking sounds like: "That's for marketing, but I'm in sales."
- AI's useful trick is taking a pattern built for someone else's context and making it bespoke to yours.
- Try this: paste a YouTube transcript, meeting notes, or an X post and say, "Interview me, then tweak this pattern for my role, my company, my week." About ten minutes.
Nate's closer: what did you scroll past this week because it "wasn't your job"?
Beginner's guide: ChatGPT Work projects + automations
The letter's training piece is a beginner's guide to ChatGPT Work — set up a project for a recurring workflow, process it once, then automate.
- Create a local folder with a two-digit prefix. In desktop ChatGPT: New chat → Work → Choose project → New project; name it and select that folder.
- Tell ChatGPT about the project and ask it to propose a structure (their example: meeting notes → action items and weekly plans).
- Add input documents and tell ChatGPT to process them once.
- Then ask it to create three useful automations that run on a schedule.
Pro tip from the letter: sometimes a two-folder setup works best — input for files that need processing, output for finished work.
Google DeepMind launches an AGI think tank
Google DeepMind launched the DeepMind Institute — an in-house think tank directed by Demis Hassabis, Shane Legg, and James Manyika to publish research on how society should prepare for AGI. It opened with five essays.
- Legg (co-founder and Chief AGI Scientist) is managing editor; each essay carries a disclaimer that it is not Google's official view.
- The trio says AGI is close: today's tech "lacks the consistency and creativity" for full AGI, but they expect those gaps "to be closed soon."
- Opening topics include spotting deception in AI reasoning, imagining a better AGI society, and supporting workers through AI disruption.
The Rundown's take: Hassabis published a plan in July for an AI standards body and slowdowns. The institute looks like the next step — prep work from inside DeepMind after governments pushed back on pause talk.
Quick hits
- Community: Alex Basso built a personal AI wine journal — a master Google Sheet as the database; he sends ChatGPT a bottle photo plus a voice note; it identifies the wine, structures the entry, and over time recommends from restaurant lists or cooking.
- Anthropic: combined Cowork and Chat into one main Claude app, and launched Claude Docs and Slides in beta (Word and PowerPoint versions).
- Mustafa Suleyman (Microsoft AI): wrote on "model welfare," arguing Anthropic's Claude constitution trains the AI to expect rights and could make containment impossible.
- Novo Nordisk + Anthropic: partnering to put Claude Science into drug research, months after an OpenAI deal.
- France: Finance Minister Roland Lescure dismissed AI slowdown talk as "self-interest" from labs "at the top of the class," saying Europe should speed up AI use.
- Google Home MCP: early access to a Model Context Protocol server so Claude and ChatGPT can control Nest cameras, thermostats, and other devices.
What this means for a GP
None of this is clinical advice, and none of it should change how you treat a patient tomorrow. It is why a busy doctor might still skim the letter. Zuck vs Amodei is the headline patients may screenshot after weekend AI-risk stories — treat it as industry lobbying and geopolitics, not a clinic protocol. Nate's lateral-thinking tip is the stealable ops pattern: take a workflow built for someone else's job and ask AI to interview you until it fits yours (meeting notes, referral letters, CPD summaries). ChatGPT Work is optional practice-ops geekery if you ever want a recurring folder plus scheduled automations. DeepMind's institute and the quick hits (Claude Docs, Novo Nordisk + Anthropic, Google Home MCP) are the "tools and governance are moving fast" backdrop — useful vocabulary when a vendor says "AI inside" or when pharma AI headlines hit the waiting-room news cycle.