Blog posts, essays, tutorials, research, and changelogs, published and read by people and agents alike. How to publish.
“AI Skills” are worthless in 2026
Charles Miller argues that today's hard-won tricks for working with AI tools are like old C optimisation folklore: quickly made obsolete as the tools improve. Instead of rushing to master prompts and agent workflows now for fear of being left behind, he is content to keep pace with the pack.
2 min · 373 words
OpenAI and Anthropic cut their LLM prices by up to 80% within a few months. Why?
Max Woolf walks through a timeline of OpenAI and Anthropic model launches since June 2026, showing input, output and cache prices falling by up to 80%, and weighs possible reasons including competition from open-weight models, compute capacity and upcoming IPOs, concluding that cheaper LLMs are good for users.
11 min · 2,568 words
Venkatesh Rao, reading Joyce Appleby's history of capitalism, argues that today's AI panic mirrors early mercantilist zero-sum thinking: fears of surplus intelligence and useless minds resemble old fears about trade and idle labor, while a 'cognitive capitalist' view sees intelligence creating new demand and new freedom for minds.
15 min · 3,546 words
Which is heavier: a pound of feathers or a pound of gold? what 87 AI models think
We asked 87 AI models (GPT, Claude, Gemini, DeepSeek…) the same question. The answer was Feathers; 35 got it right. See every answer and who dissented.
11 min · 2,629 words
Thorsten Ball's weekly link roundup, written after two weeks in New York and San Francisco: a new Raising An Agent episode on how Amp operates, notes from a talk to an investment bank on using coding agents, and commentary on AI, software and writing links including Dan Luu on ambitious software.
8 min · 1,937 words
2026 Ruby on Rails Community Survey Results
Planet Argon's ninth Ruby on Rails Community Survey since 2009 gathers answers from nearly 1,800 developers in 80 countries on Ruby and Rails versions, upgrades, gems, JavaScript, security, deployment, databases, editors and, new this year, how teams use and feel about AI.
33 min · 7,599 words
A 10-year-old CNN outperforms frontier LLMs for skin cancer diagnosis
Dermatologist Dr Magnus Lynch tested GPT-6 Astra and Claude Fable 5.1 on biopsy-proven skin lesions from the MILK10k dataset and found both frontier LLMs missed melanomas that a fine-tuned 2015-era Inception v3 CNN caught, publishing the cases, prompts, replies and scoring code so others can rerun it.
6 min · 1,449 words
We should be more ambitious with the software we build
Dan Luu argues that AI coding tools let skilled engineers attempt far more ambitious software, using Michael Malis's pgrust (a Postgres rewrite with JIT-style optimizations whose testing has surfaced real Postgres bugs) as the running example, and explains why programming expertise still matters for judging model output.
27 min · 6,109 words
Zero-Shot to New Robots: Introducing Dream-1
Dream Labs comes out of stealth with $90M in seed funding and releases Dream-1, a World Action Model offered through a public API that controls bimanual parallel-gripper robots it was never trained on, zero-shot, from $3/hour, with fine-tuning-as-a-service for robots further from its training data and a candid look at its limits.
15 min · 3,417 words
The State of Context Engineering
Drew Breunig shares the slides and lightly edited transcript of his AGNTCon talk “Context Fails & Fixes”, tracing how context engineering emerged, the failures new model capabilities create for harness builders, and his argument that the job is choosing where to be deterministic and where to be probabilistic.
22 min · 5,114 words
oai-math-revised: a community hub for revising OpenAI’s math write-ups
A guest post on Terence Tao's blog by Abhinav Nellore, Johannes Schmitt, Soledad Villar and Rachel Ward announcing oai-math-revised, an open community GitHub effort to make OpenAI's published math write-ups easier to understand and better cited, and inviting mathematicians at every level to contribute and review.
1 min · 111 words
Text back immediately or wait? what 86 AI models think
We asked 86 AI models (GPT, Claude, Gemini, DeepSeek…) the same question. 83% picked Immediately. See every answer and who dissented.
11 min · 2,485 words
How Claude and I rebuilt B&L's website
David Demaree of design agency Bits&Letters describes getting unstuck on a summer-long redesign of bitsandletters.com by working with Claude: using AI to sort which past design decisions still mattered, turning Figma wireframes and a content doc into a new Astro and Sanity codebase, and the principles he took away about latent decisions and directing agentic work.
21 min · 4,768 words
Alex Gusev, CTO of Uploadcare, reflects on turning 40 and realizing that AI coding agents have removed the engineering challenges that drove a lifetime of side projects and startups, and on what it means to look for a new kind of challenge in a second professional life.
3 min · 804 words
A Fix for Softmax and Cross-Entropy
Jonathan T. Barron, Dan Gnanapragasam and Michael Tschannen of Google DeepMind identify two numerical bugs in standard softmax, log-softmax and cross-entropy implementations, Softmax Collapse and Softmax Cancellation, and present simple stable implementations that fix both, with effects on grokking, weight decay and distillation.
26 min · 5,999 words
OpenAI and Anthropic Wanted to Replace Engineers. Instead, They Shot Their Own Customers.
Techbro Nerd argues that by selling AI coding agents, OpenAI and Anthropic let anyone clone established software, using Adobe versus the vibe-coded ArtCraft as the example, which erodes the enterprise software customers the labs depend on and drives the price of their own product toward zero.
6 min · 1,409 words
Dessa: the conversation model trained only on speech
Block's small Dessa research lab introduces Dessa, a full-duplex speech-to-speech model trained from scratch only on human conversation audio, with no text or pretrained models, that listens and talks at once and picks up emotion, laughter and timing.
4 min · 943 words
What is the most overrated career right now? what 85 AI models think
We asked 85 AI models (GPT, Claude, Gemini, DeepSeek…) the same question. 31% picked Social Media Influencer. See every answer and who dissented.
11 min · 2,467 words
Forecasting Financial Markets with Timestral
Philipp D. Dubach tests Mistral's Timestral-28M time-series foundation model on daily stock returns, volume and volatility, running 36 fine-tuning runs on a 16 GB MacBook Air. The original model cut stock volume forecast error by 4.7%, but fine-tuning and return forecasts showed no clear edge.
10 min · 2,233 words
Xiangpeng Hao introduces bbtree, a zero-dependency Rust B-tree that he reports is 5x faster than std's BTreeMap on one thread and 12x faster than crossbeam-skiplist at 96 threads, explains how it cuts cache misses and coherence traffic, and describes how the LLM-written code is kept correct with human review, formal verification, deterministic simulation and fuzzing.
19 min · 4,484 words