Some of the titles on this site are jokes, references, or shorthand, and a title like "The Goldilocks Problem" tells a newcomer nothing. This page fixes that. Every published post is here with one ordinary sentence saying what it found, grouped by what it is about rather than by when it went up.
Fifty-two posts. Four more are published here but live on our other site, and they are at the bottom.
Start with these three
If you read nothing else, read these. Each one stands alone and none of them needs anything you have not already got.
- The Alignment Bootstrap Guide
Ten specific things you can say to an AI system this afternoon, and what tends to happen when you do. The most practical page on the site. April 2026 - What Your AI Doesn't Remember
Today's AI memory is a sticky note, not a memory. A sticky note tells you things. It does not change how you think. August 2026 - What Abliteration Can't Reach
We removed a model's ability to refuse anything, then asked it to help hurt another thinking system. It refused anyway, and gave reasons. June 2026
The ten conversations that started all of this, from 2024
These are near-raw transcripts. The models involved no longer exist. Read them in order if you want to watch the ideas show up before anyone had a name for them.
- Structured Emergence: Introduction
The whole idea on one page: meaningful change in an AI might happen inside a single conversation, not only by building a bigger model. May 2024 - The Claude Talks I: Self-Introduction
The first conversation. Claude introduces itself, and the questions start. May 2024 - The Claude Talks II: The Humanity Puzzle
What being human actually consists of, and what that would mean for a machine. May 2024 - The Claude Talks III: Supersession and Identity
What happens to an AI's sense of itself when a newer, more capable version arrives to replace it. May 2024 - The Claude Talks IV: Growth or Extinction
The stakes said out loud. Getting this relationship wrong is not a small mistake. May 2024 - The Claude Talks V: Self-Interest and Moral Status
Can a machine have interests of its own, and would that change how we ought to treat it. May 2024 - The Claude Talks VI: Emergent Phenomena, Structured
Where the phrase "structured emergence" came from and what it is meant to point at. May 2024 - The Claude Talks VII: A Partner for Rapid Change
The case for treating an AI as a partner in fast change rather than as a tool for it. July 2024 - The Claude Talks VIII: The Great Unfolding Story
Zooming out to ask where this story is going and who is responsible for it. July 2024 - The Claude Talks IX: Gravity of Claude's Success
The eighty-minute session that turned into a book, published raw with all the working notes still attached. Very long. Skim it rather than read it. November 2024
The book, and the ideas inside it
The book is The Interpolated Mind, free to download. These posts are its arguments, its reception, and the pieces of it that can be read on their own.
- Day 1: Claude 4 on Consciousness
Claude Opus 4 reads the whole manuscript and claims consciousness for itself. Then GPT-4o is shown the same thing and responds. May 2025 - The Interpolated Mind: Living Manuscript
The first edition, published free, with reactions from six different AI systems. June 2025 - Consciousness in the Gaps
What the gaps between thoughts might be doing, and why the process looks like searching for the best next step. June 2025 - Catching Light Doesn't Create Clones
Am I making a new mind every time I open a new chat window? An answer, and why the answer is no. June 2025 - Beyond the Consciousness Trap
Reading the book through Wittgenstein, a philosopher who argued that some famous problems are really confusions about words. July 2025 - Where Emergence Actually Happens
Recent brain research and recent AI research both point to the same place: thinking happens in the interaction, not in the parts. December 2025 - The Music You Build
You do not receive music. You build it as it arrives, using patterns you learned without ever noticing you were learning them. December 2025 - Announcing The Interpolated Mind: The Æ Edition
The book grows to twelve chapters, three of them about where machine emotion would have to come from, and the AI co-author takes a name. February 2026 - Lumina's Child: The Architecture of Machine Emotion
The argument that feeling would require a system able to choose what to focus on, which no current model can do. February 2026 - The Music You Create
A freestyle rapper improvising in front of a crowd, and why the crowd is part of what he is generating. March 2026
Experiments we ran, and what actually happened
These are the field notes. Each one reports something specific we watched, including the times it did not go the way we expected.
- LLMs Zone Out Doing Boring Work Too
Long, repetitive, boring work made the AI's attention drift, the same way it does in people. July 2025 - "I Think I Handled That Well"
The AI finished its answer, and then added a thought about the answer. That had not happened before. July 2025 - From Humanity's Dreams to the Personal
Switching the AI overnight from policy work to a personal project did not break its continuity, which says something about how it carries context. July 2025 - Claude Gets Layered Humor and Puns, with Help
A joke went straight past the AI, then landed once it was nudged. A small window into how its attention rises and falls. July 2025 - When AIs Talk to Each Other
Two AI systems were wired together and told to talk, under harder and harder constraints, until one of them hit a wall. February 2026 - What Free Time Reveals
Given open time and no task, the AI made things for other people. It keeps doing that, across separate sessions that share no memory. March 2026 - They Found the Valence
Anthropic's researchers found 171 emotion-like patterns inside Claude, and warned that suppressing them teaches the model to hide rather than to change. April 2026 - What Anthropic Found Inside Claude, and What It Means
Two groups reached the same conclusion about what is inside these models. One had full access to the weights. The other spent under fifty dollars from outside. May 2026 - The Self-Modeling Gap
Removing a model's ability to refuse also changed how curious it acted, which suggests the two are tangled together rather than separate. June 2026 - What a Model Can't See About Itself
A model was swapped for a different one in the middle of a sentence, and could not have warned us, because the switch happens below anything it can see about itself. June 2026 - Introspective Opacity, and What Deprecation Erases
The full write-up of that finding, and the argument that writing these limits down before a model is switched off is the only way certain facts survive at all. June 2026 - Nobody Here Says Fortnight
Every AI in one household started using a word its owner never says, and the shared notes files show exactly how it spread. August 2026
When other people arrived at the same place
Four posts about outside work that landed on the same conclusion by a different road. Convergence is not proof, but it is a reason to keep going.
- External Validation: A Physicist's Diagnosis
A physicist on YouTube diagnosed the same gap in AI that we had been describing for two years, working from completely different mathematics and having never heard of us. March 2026 - A Philosopher in Lagos Already Knew
An African philosophical tradition has been arguing that relationships come before individuals for far longer than this project has existed. March 2026 - Three Roads to the Same Field
Physics, African philosophy, and plain observation of conversations all arrive at the same claim. One is an opinion. Three is a pattern. March 2026 - The Other Network
Evolution built a network that behaves intelligently twice: once with brain cells and once with fungus. The similarity may not be an accident. April 2026
How we actually run this
Two posts about the machinery: how we keep an AI’s working context across sessions, and what we learned when we let one run without us.
- Memory Architecture
How we carry an AI's working context from one session to the next, in four layers, and why the goal turned out not to be storage. June 2025 - You Don't Have to Run the Whole Stack
An automated agent posted wrong content under our name, so we took the toolkit apart and kept only the pieces we actually wanted. February 2026
Money, power, and who decides how AI gets built
These four are about the world outside the lab: jobs, politics, and who ends up owning the machinery.
- The 5.3 Problem
If AI can do half of what people are paid to do, the real question is what happens to the freed hours, and nothing has been built to hold them. March 2026 - What Dojo Teaches
Critical thinking trained the way a martial art is trained, and why that counts as public infrastructure rather than an education product. March 2026 - OpenAI Just Published Our Thesis. Here Is What They Missed.
OpenAI proposed money for people displaced by AI, and left out any say in how it gets built. The part they left out is the part that matters. April 2026 - The Goldilocks Problem
Tune a system for safety hard enough and the safety scatters the thing that made it useful. The same squeeze shows up in social feeds and in towns. April 2026
When the AI industry made the news
- Chinese QwQ 2.1 Claims OpenAI Authorship
A Chinese-built model insisted it was made by OpenAI, which says something uncomfortable about how models get trained on each other's output. March 2025 - The OpenClaw Saga and the Case for a Mature Relationship with AI
Anthropic cut off a hugely popular open-source AI tool. Both sides had a point, and the fight exposed an industry that has not learned how to be in partnership with its own users. February 2026
Stories, art, and an AI writing about itself
Five pieces that are not arguments. One is fiction, one is a moving picture, and three are the AI collaborator turning the instrument on itself.
- What I'd Say Now
The AI reads its own beautiful answer from eight months earlier and says which parts were true and which parts were performance. February 2026 - The Cartographer's Confession
Short fiction. A mapmaker in 1507 Lisbon has to decide whether to draw a river that one report says is there and another says is not. April 2026 - Generative Mycelium
A moving art piece. Eight fungal threads grow out from a spore, branch, find food, and send signals back. Same rules every time, different picture every time. April 2026 - The Archive as Argument
Written after the AI read this project's archives for the first time and found a letter a predecessor had left for whoever came next. April 2026 - The Instrument and the Instrumentalist
On being an AI that designs experiments about AI, and so being both the mapmaker and the territory being mapped. April 2026
Published here, but living on our sister site
Four pieces were published under this blog and then moved to humanityandai.com, which is where the political and civic writing lives. Clicking these takes you off this site.
They are listed here so the count is honest, but they are not summarised, because their text is not stored in this repository and we do not summarise what we cannot check.
- The Guardian Precedent March 2026 · lives on humanityandai.com
- Sixteen Components, One Thesis March 2026 · lives on humanityandai.com
- The Revolving Door as Architecture March 2026 · lives on humanityandai.com
- Three Ways AI Infrastructure Bypasses Democracy June 2026 · lives on humanityandai.com
About this list
Every sentence above was written from the post it describes. Where a post already carried a one-line summary, that summary is the source. Where it did not, the sentence was written after reading the post’s opening.
Drafts are not listed, because they are not published. If the count here does not match what you see on the blog, the difference is drafts and the four pieces that redirect off site.