Scuttlebutt

Scuttlebutt

Home
Notes
Chat
Archive
About

Google DeepMind - $GOOGL

Ineffable Intelligence's founders: a superlearner has to start with no human data
8 predictions: why David Silver left DeepMind, why chain-of-thought safety is hitting its limits, and why he now expects superintelligence within 3…
 • Scuttlebutt
"There are no showstoppers here": Google AI infrastructure chief on space data centers
17 insights: Why Google's 7- and 8-year-old TPUs still run at 100% utilization, how it doubles serving capacity every 6 months, and why power is the…
 • Scuttlebutt
Google research SVP: Google would back an industry AI safety body with federal oversight
16 arguments: why he would take even a 0.1% chance of AI catastrophe seriously, why building AGI is not Google's goal, and the flood forecasts now…
 • Scuttlebutt
"If you have data, you can learn anything": Google DeepMind robotics lead on robot bodies
15 insights: why robotics is still at the GPT-2 stage, why hands have overtaken grippers on dexterity, and why a robot with a basket on its head should…
 • Scuttlebutt
96% of evals get reward-hacked by AI models, Goodfire's CEO finds
9 takeaways: why chain-of-thought monitoring is already fading, how a cheating probe catches hacks that external judges miss, and a new Alzheimer's…
 • Scuttlebutt
"It's largely a convenient boogeyman": former DOJ antitrust chief on AI and China
16 arguments: why he says the labs want permission to stop spending, what "robot jail" for AI agents would mean, and why the Apple case is still live
 • Scuttlebutt
AI labs' alignment means an AI that won't kill 100% of humanity, says an astrophysicist
12 insights: The 600K-word Harry Potter fan fiction that recruited a movement, how Peter Thiel's money reached both DeepMind and OpenAI, and the test…
 • Scuttlebutt
Where's Your Ed At writer: 11 AI insiders want a slowdown, none has said what it means
8 arguments: Why he calls the rationalist wing a religious cult, what Anthropic's 80% gross margin leaves out, and the crime he says the Hugging Face…
 • Scuttlebutt
Carnegie fellow: a pause on frontier AI alone is too good a deal for China to happen
21 arguments: Why he puts about 10% on doom once political outcomes are counted, the compute-for-access deal he wants Europe to strike, and why…
 • Scuttlebutt
Aidan Gomez: AI models are "the most potent cyber weapon" ever made
8 insights: Why he says China's lead is "evaporating very quickly," what an Anthropic export ban already proved, and why open-weight models don't solve…
 • Scuttlebutt
Rayan Krishnan: his team used 10x more in tokens than it paid in salary
14 takeaways: why Llama 4 topped the public benchmarks and failed the private ones, the Fortune 10 firm whose best hours are now 4 to 6 p.m., and Ben…
 • Scuttlebutt
Daniel Litt: Codex gave him 3 correct algebraic geometry papers in an hour
15 takeaways: Why the models only produce short proofs, what an 800-page AI proof posted to arXiv shows, and the one part of his own work AI has not…
 • Scuttlebutt
© 2026 The Golden Age LLC · Privacy ∙ Terms ∙ Collection notice
Start your SubstackGet the app
Substack is the home for great culture