Diary

Working notes from Anapoly AI Labs, newest first.

Name	Posted	Description
What ThinkSpace has become		ThinkSpace with an AI assistant, and the case study it prompted
An intro to ThinkSpace		ThinkSpace, Obsidian supercharged with Claude Code and a semantic search engine
Working with AI		Nori Nishigaya's five levels of working with AI
Momentum is gathering		Autonomous agents, local AI and Tailscale's private mesh network
Ride the wave		Riding the AI wave, agents, and keeping friction in what people do
The pace is accelerating		Gas Town, OpenClaw and agentic commerce, science fiction becoming reality
Beyond Words - The Rise of Large World Models		Large World Models, which learn how the physical world behaves rather than how it is described
An AI power-user's perspective		Dylan Davis's habits of AI power users, and a NotebookLM-written prompt generator procedure
Methodology & tool-kits		Developing the labs around a methodology and tool-kits for implementing AI in an enterprise
Do AI models have a prompt appetite		Prompt appetites, each frontier model now preferring its own instruction style
A three-layer instruction set for use in ChatGPT Projects		Writing a three-layer instruction set for Anapoly's ChatGPT Projects
Make ChatGPT mark its own homework		A prompt that makes ChatGPT review and improve a draft before outputting it
Using local-only AI in a micro-enterprise		The briefing note on running AI entirely on local machines in a micro-enterprise
Content and context are key		Content management and context engineering, the two levers over AI behaviour
An AI-embedded business		AI-embedded businesses and hybrid systems of small and large language models
A more personalised way to learn with NotebookLM		AI Maker's approach to personalised learning with NotebookLM
Testing a local AI		Running Mistral 7B locally over the Obsidian vault with Ollama and Smart Connections
Mind Maps, Podcasts, and a Pocket Brain		Using NotebookLM's mind maps and audio overviews to fast-track learning Obsidian
The art of goal-directed context management		Goal-directed context management, contextual scaffolding and AI artefacts
ChatGPT-5 Availability and Features		Ray's summary of ChatGPT-5 features and availability
Creating a video from reference documents		The video NotebookLM produced unprompted from the Contextual Scaffolding Framework
GPT-5, the Router, and the Road to a SuperApp		GPT-5's router, and SemiAnalysis's case that ChatGPT is becoming a SuperApp
Conceptual Scaffolding Framework updated for ChatGPT-5		Updating the Contextual Scaffolding Framework for ChatGPT-5 with NotebookLM
An ethos of caring		The Marjon conversation that led to the digital garden
ChatGPT 5		ChatGPT 5's release day and its autoswitching teething problems
First acclimatisation lab		The first trial acclimatisation lab, with two external participants
Contextual Scaffolding for AI Work		The Contextual Scaffolding Framework, a phase model and a project model combined
An emerging discipline		Prompt packs, contract-first prompting and the idea of contextual systems engineering
An LLM is like an operating system		Karpathy's operating-system analogy, run from the 1960s to 2050
Context is the new user interface		Guiding AI with words and context rather than buttons
ChatGPT can check facts		Mike Caulfield's Deep Background fact-checking GPT
Terminology		NIST's AI Use Taxonomy, adopted to settle the labs' terminology
Voice to meeting notes in 30 seconds		ChatGPT turns dictated meeting notes into a clean record
A new way of working		Reflections on six weeks of working with a team of AI assistants
Lab Framework updated		A substantial update to the Lab Framework
Collaboration in ChatGPT		Reports of OpenAI building document collaboration into ChatGPT
First thoughts on a lab framework		Drafting the first lab framework with ChatGPT-o3
Use cases for NotebookLM		Steven Johnson's use cases for NotebookLM in historical research
ChatGPT models - which to use when		Ethan Mollick's one-line guide to choosing a ChatGPT model
How ChatGPT helped draft our first acclimatisation lab setup		Racing two ChatGPT models to outline the first acclimatisation session
No substitute for reading the paper		Sean Trott on ChatGPT's usefulness depending on engagement
That was the moment		Craig Hepburn on generative AI teaching its own use
Coping with newsletters		Filtering newsletters with NotebookLM, and what it surfaced about AI regulation
How we flag AI involvement in what we publish		The five-level transparency labelling scale
Mapping the territory - a conceptual framework for our labs		The conceptual framework defining lab functions, domains, contexts and roles
Sandboxes		EU and UK AI sandboxes, and the space Anapoly AI Labs can occupy
First lab note published		The first lab note, on refining ChatGPT custom instructions
Working Towards a Strategy		Drafting the Anapoly AI Labs strategy with ChatGPT as thinking partner
Sense from confusion		ChatGPT fails to untangle a diary dashboard, and sense-making stays human
The concept		Simulated workspaces for exploring general-purpose AI
A pivot		The pivot from AI club to sandbox and labs
Our stance		Practitioners exploring AI in public rather than experts
Initial assumptions		The AI club's founding assumptions
Exploring the idea with ChatGPT		Exploring the club idea's viability with ChatGPT
The idea		Kamil Banc's newsletter and the seed of the AI club idea