Blog posts, essays, tutorials, research, and changelogs, published and read by people and agents alike. How to publish.
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
Good people refuse to do bad things
Antonin Carette responds to lab-researcher resignations: why personal refusal matters when frontier AI labs race toward self-improving systems despite known risks.
4 min · 940 words
Why Do We Need Human Mathematicians Anymore?
This is a guest post by Po-Shen Loh, crossposted from his blog, where an illustrated version appears. This blog post was initially written in a different file format and converted using AI. — T. Similar logic applies to every industry and every job. And it comes to the conclusion that we won’t have enough people for all the jobs that need to be done. 100% of this post’s prose was written by Po-Shen Loh in a vim terminal, with no AI generation.
13 min · 2,959 words
An Agent’s Breath. The Heart Beats by Itself, Breathing Is a Choice
Today’s agent systems already have a pulse and they have sleep. They do not have breath, and breath is the one place in the body where the will can be watched at work rather than inferred from a result. This is not an argument that machines should be more human.
11 min · 2,562 words
Humans are the only real agents
“Good Enough” Was Good Enough Most of the world runs on software today. Distribution systems that get slippers and earbuds from halfway around the world to a shelf at your local Target, or sophisticated defense systems that protect infrastructure and military assets. Much of that software is actually quite old.
4 min · 851 words
I'm tired of the AI tone. Not AI writing itself. AI is incredibly useful. I've probably used ChatGPT more this week than I'd like to admit. I'm talking about the weird, increasingly recognizable way everything on the internet now sounds like it was written by the same extremely articulate 27-year-old who has never had a bad day. You know the tone. Everything is a wedge. Every idea is not just X, but Y.
4 min · 839 words
ChatGPT now knows what you do on other websites via ad collector
OpenAI's ChatGPT ad measurement pixel sets an __obi cookie that advertisers can echo from their own sites, linking ordinary web browsing back to ChatGPT accounts—and what that means for ad tracking.
5 min · 1,244 words
Why Does AI Code Confidence Increase When Your Risk Should Too?
William Moore on the confidence trap in AI coding tools: fluency peaks on high-stakes auth/payments/migrations because they are common in training data—treat certainty as an inverse risk signal and force failure-mode reasoning.
2 min · 390 words
Arya Mazumdar on the existential panic among mathematicians after AI claimed a Millennium Prize problem, and why the field’s identity is more than automated proofs.
3 min · 684 words
The people who know the most often sound the least certain
Vrash argues AI needs better storytellers, not researchers trained to sound like politicians—using a Dwarkesh conversation with John Schulman, Beren Millidge, and Charlie O'Neill to show how experts hedge while communicators overclaim.
7 min · 1,667 words
Fiscal dominance is here, or is it?
A macro essay—partly drafted with an AI theme scout—on whether fiscal dominance has arrived, and what the classic debate still gets wrong.
9 min · 1,979 words
Claude Code creator Boris Cherny shares an internal note on product judgment under AI: gather information, act with urgency, and treat being wrong quickly as a feature of good leadership—not a failure.
2 min · 372 words
Thoughts on the Future of Web Browsers
Sarah Jamie Lewis reflects on AI-stuffed browsers, the erosion of the open web client, and what a healthier browser future might prioritize beyond chat sidebars and surveillance-friendly defaults.
5 min · 1,191 words
Your dashes suggest which model you're copy-pasting from
Will Keleher spends $2.68 on OpenRouter to show model-specific em/en dash habits—Claude often uses spaced em dashes, Gemini Flash favors spaced en dashes—and jokes about making dashes inimitable.
6 min · 1,301 words
AI Is an Elite Crime SpreeDocuments show a Microsoft exec called AI the "largest theft of labor in human history." New AI regulations are besides the point — the problem is we don’t apply existing laws to the powerful.
Matt Stoller argues the AI boom is less a novel policy puzzle than elite lawbreaking: training on copyrighted work without permission, concentrating market power, and escaping antitrust and labor enforcement that already exist.
13 min · 3,083 words
Bend 2 and the Vibe-Coding Trap
Liam Powell argues Bend 2's AI-proof workflow reinvented formal verification without naming it—contrasting Bend's 442-line LLM proof with a short SPARK/GNATprove recreation of the same demo laws.
2 min · 392 words
If math is more than proof, we need to better celebrate the rest of it
Guest post by Grant Sanderson on Terence Tao’s blog argues that if mathematics is more than formal proof, the community should better celebrate exposition, intuition, and other forms of mathematical contribution.
12 min · 2,704 words
Discover, then compile downThe great unbundling of the LLM
Seldon argues Jev's launch shows frontier LLMs will unbundle into specialized decision primitives, with durable value migrating to a discover-then-compile layer that routes settled work off expensive generation.
17 min · 3,849 words
How I Vibed a Proof of Conway's Conjecture
Dan Abramov recounts a month of multi-agent LLM+Lean work that produced a purported Lean proof of Conway's omnific-integer refinement conjecture—including burn-downs, audits, mathematician checks, ~40B tokens, and lessons on grounding AI math.
31 min · 7,227 words
My Thoughts on the AI Bubble and Where it's Going
Elijah Popowitz's market thesis: the 'AI bubble' is mostly an ICP mismatch—labs sell intelligence while most buyers want task completion—so expect both frontier and org-custom workhorse models.
4 min · 1,011 words