At the moment, this is something of a thought dump. I'll come back to this and rework it into some kind of structure.
None of the content posted here is generated by AI. I used to use ChatGPT to read through my posts before I published them. I did this for things like spelling and grammar checks, and also for certain posts to improve for SEO. I no longer do that. Mostly because it kept trying to suggest changes, and I would spend longer thinking about how to encorporate those changes than I had forming the original thought that made up the post. All of that, and more, is why in early 2026 I stripped back my website completely - and a big part of that was making sure the only words you read here are mine. I also stopped caring about SEO for this site.
I do use AI for some of my work. Most often either Microsoft Copilot, or Anthropic's Claude. Like a lot of people in the charity space, I have conflicting thoughts about AI and if/how/when to use it. It's on my todo list to externalise that. Watch this space. Though I have spoken about the sector's mass-exodus to Claude, and why I don't think it's the virtuous act people think it is.
That said, as mentioned, I do use AI for certain things. At some point I'll list out my internal logic for what those things are.
In the meantime, as well as "big-tech" AI, I also make use of smaller, self-hosted AI tools. This includes an an Ollama instance, which I use in conjunction with n8n for a number of projects, including:
- Creating a local LLM (SLM, technically), of my Obsidian notes
- Analysing my own notes to do things like infer connections between thoughts, resurface old but relevant content, and other things of that nature
I mostly do this as a point of interest, and as an exercise in helping me understand the intricacies of how RAG models work. But also because it was fun. And I liked the idea of an AI model that is hosted and owned by me, trained only on my thoughts.