// AI Transparency;
AI Transparency Statement for Twisted Brackets
Listen to this statement using the player at the bottom of the page.
I am not hiding the fact that I use AI on Twisted Brackets. This page explains how.
The ideas, opinions, and structure of every post start with me. I decide what is worth writing, what angle to take, and what the article needs to cover. Generative AI helps with the step that is slowest for me: turning a solid outline into finished prose, and producing supporting images and graphics along the way. Every article is then reviewed, edited, and managed by me so the published piece is accurate and reflects what I actually mean.
About the audio voice
If you listen to the audio version of this statement, or to other audio posts on this site, you are not hearing my real voice. It is an AI voice I have chosen as my avatar. I deliberately do not train a clone of my own voice. My voice is biometric data I am not comfortable handing to a model provider, so a selected AI voice is the safer option for me.
How a post gets made
In short: I curate and outline; AI helps draft and visualise; I review and ship.
1. Curate, understand, and outline
Every post begins with a topic I care about and understand well enough to teach or argue for. I prepare the outline myself: the claims, the structure, the examples, and the point of view. That outline is the brief. AI does not invent the thesis.
2. Use AI to fill the gaps
Writing the connective tissue between outline points is the part that takes me the longest. That is where AI earns its place. I use large language models to draft sections and tighten explanations from my prompts, and multimodal / image-generation tools to produce explainer graphics that support the article. The goal is speed on the slow step, not a substitute for knowing the material.
I also use specialised skills I have built for this site, often run through AI agents that can work a section at a time. Those skills let me iterate on individual pieces of an article (a section, a diagram, a visual, a pass of edits) until that piece is right, before the whole article is finished. A lot of that iteration is prompt engineering in practice: refining instructions until the output matches what I meant.
3. Review, edit, and own the result
Nothing publishes as raw model output. Models can hallucinate: confident-sounding text that is wrong, shallow, or not grounded in the brief. I review and edit the full article for accuracy, tone, and intent. If a draft fails that bar, it gets fixed or cut. The published post is something I stand behind.
What AI is and is not used for
AI is used for:
- Drafting and refining article prose from my outline
- Filling gaps between outlined sections
- Producing explainer images, diagrams, and supporting graphics
- Iterating on individual sections through custom skills I have created
- Narrating audio versions in a chosen AI avatar voice (not a clone of my own)
AI is not used for:
- Choosing what I believe or what the article should argue
- Replacing my review of facts, claims, and technical detail
- Training or cloning my real voice for narration
- Tracking readers, personalising content, or other non-writing AI features on this site
Why this matters
Transparency beats theatre. Plenty of sites either pretend AI was never involved or wave it around as a gimmick. I would rather say plainly: the thinking and structure are mine; AI helps me get from outline to draft faster; the final article is reviewed and edited by me. And when you hear audio on this site, you are hearing an AI avatar voice I selected, not a replica of me.
Want the vocabulary?
If terms like generative AI, LLMs, prompts, agents, or hallucinations are unfamiliar, I keep a plain-language glossary in Cracking the AI Jargon. It covers the definitions above in more depth, plus related ideas like inference, tokens, and tool use.
If you have questions about how a specific post was made, get in touch.