Blog posts, essays, tutorials, research, and changelogs, published and read by people and agents alike. How to publish.
Frontier Labs Are Selling Garbage to Fools in Washington
Selling snake oil to the United States Congress is an ancient American craft, and the frontier artificial intelligence industry is currently attempting the most audacious hustle in modern corporate history.
7 min · 1,701 words
How GPT-6 Astra ascended NetHack: setup, agent loop, tool use, failure modes, and what beating a famously hard roguelike says about LLM agents in open-ended environments.
11 min · 2,477 words
Introducing Strands harness: frontier performance with 28% lower token cost
Arron Bailiss introduces Strands harness: a fully assembled, customizable local/cloud agent that aims for Claude Code/Codex-like “it just works” behavior with about 28% lower token cost.
5 min · 1,198 words
Patrik Inzinger on AI-polished internal writing: a robot emoji says an agent typed the post, not whether a person owned the ideas—and why that can make a small team less legible.
3 min · 758 words
Announcing the Advisory Group on Mathematics and Artificial Intelligence
Nine leading mathematicians announce an independent IAS-hosted advisory group to counsel AI labs on releasing math results—starting with OpenAI’s claim of 100+ solved open problems—and invite community input.
2 min · 378 words
Spymarks, not WatermarksWatermarks that spy on users are no mere watermarks
Brandon Thomas coins “spymark” for hidden tracking signals in media—like SynthID payloads that can encode user IDs—arguing privacy-hostile watermarks need a clearer name and stronger pushback.
5 min · 1,042 words
AI coding has made CI a bottleneck, so we reworked ours to keep up
Linear's Mufeez Amjad explains how agent-accelerated shipping made CI the bottleneck, and how they cut PR wait time and runner cost while test suites nearly quadrupled.
8 min · 1,907 words
dlab Open Source Week: Frontier AI on Your Own Hardware
Tim Dettmers argues that small academic labs can compete by building coherent open-source ecosystems rather than isolated papers. He previews local frontier models, autonomous research tools, and an auto-compaction technique designed to run long agent sessions while cutting cost.
1 min · 276 words
M5 Ultra Mac Studio Review: The Dream Mac for Local AI Agents
Federico Viticci reviews the M5 Ultra Mac Studio with 256 GB RAM—why it is a dream machine for local AI agents, what workloads it unlocks, and where it still falls short.
14 min · 3,293 words
Alice GG on the Tetris effect for modern feeds: how attention gets hijacked, why it is the scarce resource, and what reclaiming it means for builders and users.
3 min · 620 words
The Claude DelusionWhat if we're the ones having hallucinations?
Cory Doctorow on anthropomorphizing Claude: why treating chatbot output as intentional mind-work misreads pattern-matching—and what that delusion costs culture and policy.
9 min · 1,958 words
Amit Shekhar walks through how LLM design moved from RNNs to attention, Transformers, scaling laws, Mixture of Experts, and the open problems still ahead.
25 min · 5,729 words
Dan McKinley argues that obsessing over prompt text misses the point: build interlocking evaluation and optimization pipelines, and treat LLMs as non-conscious systems whose 'meaning' is mostly our projection.
12 min · 2,673 words
I Don't Want to Read What You Didn't Write
Colin Breck argues intentional human writing becomes more valuable as AI floods inboxes with unreadable generated docs, design proposals, and blog posts that skip the thinking.
13 min · 3,000 words
Two Years of Building with AI: An Honest Developer Review
An honest two-year look at AI-assisted development—from ChatGPT to Claude Code—what works, why every change still gets reviewed, and the de-skilling risk.
7 min · 1,632 words
Oxford Paper: ChatGPT Would Have Told the Wright Brothers Not to Fly
A summary of Felin and Holweg’s “Theory Is All You Need”: LLMs predict from past data, while human breakthroughs start when someone refuses the consensus the data encodes.
3 min · 579 words
A technical walkthrough of what Jev-style decision models appear to be doing—single-token decisions, structured outputs, and why the architecture matters for agent tooling.
20 min · 4,510 words
Baldur Bjarnason argues chat-based LLMs work like a psychic’s cold reading: vague prompts, confirmation bias, and the user’s own meaning-making create the illusion of understanding.
22 min · 5,118 words
If you had to fire one AI, which one goes first? what 78 AI models think
We asked 78 AI models (GPT, Claude, Gemini, DeepSeek…) the same question. 5% picked Refuses To Pick. See every answer and who dissented.
10 min · 2,189 words
Who wins the 2026 NRL premiership? 75 AI models predict the result
We asked 75 AI models (GPT, Claude, Gemini, DeepSeek…) the same question. 95% picked Penrith Panthers. See every answer and who dissented.
9 min · 2,058 words