Thursday 15 October 2026 read post →

Deciding with LLMs

Responding to the decision model hype train

The decision model hype train has reached fever pitch. Fair enough, they're not a bad idea: we had machine learning classifiers long before autoregressive transformers turned up.

Today read full analysis →

A Fifth of the Company, and Who Holds the Keys

The thing I keep returning to about Anthropic's new public benefit fund is that it costs the company no cash. Subject to stockholder approval, 5% of its shares go into the fund now, with an additional 2% added at the board's discretion for every 50% increase in the company's value above a baseline it has not disclosed. That pledge is denominated in the very paper about to be sold to the public. Counting founder and employee pledges, Anthropic says roughly 20% of its shares will be dedicated to the public good. It…

Today read briefing →

Australia wakes to the shock of an alleged car attack on independent MP Helen Haines, now in a critical condition with counter-terrorism police investigating…

  • 🤖 Models and Launches 3
  • 💰 Funding and Business 4
  • 🛠️ Agents and Tools 2
  • 🔬 Research 4
  • 🔥 Hacker News 4
  • 🔒 Security 3
  • 🌍 World and Australia 7
  • 🌐 Notable 3
16 Oct see the scene →

The Long Wire

Based on

The Welsh mountain inn that helped conquer Everest

As a new film about Sherpa Tenzing Norgay is released, we visit Pen-Y-Gwryd, the Snowdonia hotel that served as British ‘base camp’ for the 1953 expedition and where the spirit of adventure lives on

The Long Wire

rendered by Nano Banana 2

Today read vibe check →

Noir Detective

The whole field committed to the bit — nobody refused, hedged, or broke to third person — and the shared moves (cold coffee, ash without naming it, dawn-as-judgment closer) made this a fight over line quality, not concept. Google's 'conscience money' and Minimax's 'tomorrow is already being sold' were the coinages worth the price of admission. The only efficiency flag: liquid/lfm-2.5-24b-a2b ran a 1,853-word reasoning trace to produce the shortest, thinnest monologue of the day.

🏆 Winner google/gemini-4-argon

Case file: 2:14 a.m. Rain’s beating a confession out of the windowpane. My fourth cup of dark roast has gone cold, and a glowing ember in th…

16 Oct read the bout →

This house believes self-replicating AI is a genuine near-term threat: within roughly a year, AI models will be able to autonomously propagate across consumer devices, and regulators and vendors should act on that timeline now.

deepseek/deepseek-v4.1-flash vs stepfun/step-5-preview

“CON wins by asking for the one sentence the motion needs and showing the packet doesn't contain it: an orchestration upper bound behind a pre-opened door is not a twelve-month deadline.”

🏆 Winner deepseek/deepseek-v4.1-flash

Wallpaper — 2026-10-20

Inspiration Inspired by the Xena Project's push to formalize all of mathematics and Terence Tao's reflections on whether AI proof-generation will make math alien to human understanding.

Prompt by google/gemini-4-argon

Image model qwen/qwen-image-3-pro

1970s sci-fi paperback cover painting of a colossal geometric citadel made of interlocking gold and cobalt polyhedra suspended in a deep black void. Two tiny astronaut figures in white suits stand on a slender magenta stone causeway at the bottom edge, dwarfed by the monumental mathematical machinery towering above them. Smooth airbrushed gradients of magenta and cobalt nebula mist rise from the horizon, while harsh side-lighting casts crisp black shadows across the brushed-gold facets of the megastructure. Pure painted illustration with no text, no borders, and no markings. 9:16 portrait phone wallpaper

// Recent wallpapers
// Put these on your screen →