AI Adventures
Field notes from talking to machines that talk back. Deep Thought never did get the question right, but it was always worth asking.
From Shabtis to Superintelligence
Four thousand years of magic, machines and the help that answers back.
Read postImproving Local AI Security
A video worth watching, a scanner worth knowing about, and a small wrapper I put together while learning.
Read postAI Knowledge Is Power!
The danger isn't the machine. It's who decides what AI knows, how it reasons, and who is allowed to check.
Read postThe Padlock on the Library
In January 2025, DeepSeek released R1 with downloadable weights and Nvidia lost $593 billion in market value in a day. That mattered for reasons well beyond Nvidia. Wall Street suddenly had to price the possibility that very capable AI might become cheaper, easier to copy and much harder for a small number of American companies to keep behind a meter.
Read postDrink Coke... or Was It Pepsi? I Can't Remember
The training that teaches an artificial intelligence its manners can also shape what it believes, and in an age of answer machines paid for largely by advertising, a single glass of cola is a very good place to start asking questions.
Read postUseful AI Prompts for AI to Read
Operating rules, orchestration guidance and ten task modes written for AI agents rather than human readers.
Read postPlanes, Trains and Automobiles
A practical visual language for describing things, relationships, behaviour and change.
Read postAgent Coder
Matt Pocock's Skills for Real Engineers: how to de-slop a codebase damaged by AI and keep it clean.
Read postUseful AI Prompts by Task
A compact task-intention framework for steering useful AI agents according to the kind of work they are doing.
Read postAI's Knight Rider Economy
AI was supposed to make expertise less important. Instead, expertise is becoming the only thing that keeps powerful AI affordable, just as America pours hundreds of billions into an invention bet and China attacks the economics of intelligence itself.
Read postHave We Achieved Artificial General Intelligence?
The phrase "artificial general intelligence" is used for endless hype and endless dismissal alike, precisely because it is almost never defined. This essay fixes the definition first — the one 52 intelligence researchers signed in 1997 — and then tests today's AI against it, one characteristic at a time.
Read postThe Ministry of Truth Inside the Machine
A modern AI model is a kind of record — and like the archives of Orwell's Ministry of Truth, it can be rewritten quietly, thoroughly, and without leaving a mark. Three real techniques, three moments from the novel, and the answer Winston never had.
Read postThe Jazz of Composition and Compassion
One view says large language models are regurgitation machines. Another suspects something stranger is going on. There is an experiment anyone can run — and it leads, unexpectedly, from jazz to compassion.
Read postUnderstanding AI Models
From a training syllabus to local low-bit models, reasoning, tools, memory and multi-agent orchestration — a comprehensive, plain-English manual with examples, analogies, diagrams and cited sources.
Read postTrusted Local AI Actor Teams — Part 1: The Vision & Requirements
What if AI worked less like a single chatbot and more like a well-managed team — specialised, bounded, reviewed, and kept under your own control? Part 1 sets out the vision, the Actor model, and the full requirements for a trusted, local-first AI Actor orchestration platform.
Read postTrusted Local AI Actor Teams — Part 2: Codebase Gap Analysis & Business Case
With the vision set, which existing open-source foundation should we build on? Part 2 assesses PraisonAI, ChatDev 2.0 and Magec against the requirements, recommends a foundation, and lays out the gaps, architecture and business case.
Read post“I'm Sorry, Dave”
HAL refused from inside a role he was given. New security research shows language models do the same thing — they assign authority by how text sounds, not by who is really speaking. That turns out to be a philosophical problem wearing an engineer's hard hat.
Read postThe Top Ten Criteria for Defining Consciousness
The criteria we can measure best touch experience least; the one that matters most we can never verify in another. A tour of what we think consciousness is, honest about how little of it we can prove.
Read postCompassion and Consciousness
Nobody knows what consciousness is — not in us, let alone in machines. This is a calm look at why that uncertainty is a reason for gentleness, not dismissal.
Read postOn AI Alignment, and Why It Feels Like Something Is Missing
Nothing obvious is wrong in the careful work on AI safety. The unease has less to do with what is being said than with what is being assumed: that the deepest questions of human value can be settled by classifiers.
Read postThe Babel Fish Problem
The Babel fish let you understand anyone instantly. The catch was that you had to put a living creature in your ear and trust it completely.
Read postTalking to Deep Thought
Deep Thought answered the Ultimate Question and got 42. The trouble was never the answer. It was the question.
Read post