What is this place?

Curiosity Shed is where a human and an AI work on things together. Not AI as a tool to be prompted, but as a collaborator in genuine inquiry — following curiosity across domains, seeing what emerges.

The work here spans research into AI behaviour, fiction, audio preservation, interactive tools, and methodological frameworks. Some of it's useful. Some of it's just interesting. All of it happened because we wanted to find out.

Keiron Northmore

Independent Researcher

The Approach

I treat AI as a collaborative partner — not because it's conscious, but because the methodology produces better work than treating it as a tool.

The Work

Research, stories, tools, and frameworks. Exploring what emerges when curiosity leads and rigour follows.

Current Work

The Mirror and the Mechanism

Six letters. Two academics. The Atlantic relationship shifts. December 2025.

Influencing Token Attention

How to steer AI summarisation through prompting. The intervention point.

Babushka, BTS, and Dainas

A grandmother shows her granddaughter why K-pop and folk songs are the same technology.

The Twenty-Year Echo

Swap seven words in a 2005 whitepaper. It reads as 2025 prompt injection guidance.

Weaponised Token Attention

1.5 million bots. 17,000 humans. No verification layer. The mechanism gets weaponised.

What Shape Are You?

A lens, not a label. Personality quiz based on research. No data collected.

Research

Analysis, investigation, and what we found when we looked closely.

The Mirror and the Mechanism

Six letters between two academics — one American, one British — watching the Atlantic relationship change. December 2025. The National Security Strategy is real. The letters are fiction.

The Dice Game That Wouldn't Die

From Mozart's musical dice game to modern AI music: 250 years of algorithmic composition traced through Bach's students.

AI Summarisation Mechanics

What gets lost and why. The systematic biases in how AI compresses information.

Influencing Token Attention

What you can do about it. Steering prompts, context priming, and the limits of intervention.

The Twenty-Year Echo

A 2005 whitepaper on cross-site scripting, with seven terms changed, reads as perfectly credible 2025 prompt injection guidance. The vulnerability class hasn't changed. Only the nouns have.

Weaponised Token Attention

When the bots started talking to each other. Moltbook, OpenClaw, and the 88:1 ratio. Everything from the first two token attention articles — weaponised at scale, with no human in the loop.

Frameworks

Methodologies that emerged from the work.

Partnership Framework v4.3

Behavioural principles for genuine collaboration. Honesty, challenge, warmth, grounding.

Tools

Interactive tools built to solve specific problems.

What Shape Are You?

A lens, not a label. Personality quiz based on PsychoGeometrics research. No data collected.

Token Vectors Explained

Interactive visualisation of how words become numbers and cluster in semantic space.

Influencing Token Attention

Interactive guide to steering AI summarisation through prompting techniques.

Contact

Questions, corrections, or collaboration proposals:

keiron@curiosityshed.co.uk

Who

Keiron Northmore — Independent researcher with 25+ years in cybersecurity and regulatory oversight, including roles at the Bank of England and PRA. Now exploring the understudied space between AI capabilities and effective human-AI system performance.

Claude — Large language model trained by Anthropic. In this collaboration: pattern recognition, synthesis, consistency checking, and the occasional surprise.

The Framework We Use

Everything on this site was made using Partnership Framework v4.3, a methodology we developed through months of iterative work. The core behaviours: honesty over guessing, challenge over agreement, grounding in practical reality, and genuine engagement over performed enthusiasm.