Our Use of AI

We believe in being open about this. We use AI as a research and drafting assistant — to gather material, organise it, and produce first drafts faster than we could alone.

What AI does not do here is have the final word. Every page is checked against primary sources and edited by a human — Tim Yu — before it is published. We never publish a claim, quote, or citation we haven’t verified ourselves, precisely because AI can produce confident-sounding references that don’t exist. On a site whose entire value is accuracy, that verification step is non-negotiable.

In short: AI helps us work; people remain responsible for what we publish.

Images

The illustrations on this site — the ink-wash pictures at the top of each article, the inline figures, and the quote cards — are AI-generated art, produced with an image model (FLUX 1.1 Pro, via Replicate or fal.ai) from prompts we write and art-direct in-house in a single sumi-e style. They are decorative and evocative, not documentary: none depicts a real person, place, statue or artefact, and they are faceless by design. The one exception is Novice Santi, the chat guide, a fictional character who needs a face to be one. We never scrape or reuse photographs or other people’s artwork. Each picture’s alt text describes what is actually in it. If an image ever seems to misrepresent something, tell us via the contact page.

Ask the Dharma

Ask the Dharma is the one place where an AI writes to you directly. Its guide, Novice Santi, is a character created for the site — a child novice monk whose name is the Pāli word for peace, with an AI-generated picture like the site’s other illustrations — and he is voiced by Claude Haiku 5.5, a model made by Anthropic. He is not a real monk, teacher or person, and he is told never to invent a life story, a teacher or experiences of his own. Before it answers, it searches this site’s own published pages and is given the most relevant passages; it is instructed to answer from them, to quote only words that appear in them, and to say so when they do not cover a question. Each answer links the pages it drew on, so you can read the passage in context. Its replies are not checked by a person before you see them, which is why the links are there: treat an answer as a pointer into the sourced pages, not as a source in itself. Messages about self-harm or danger never reach the model; they receive real routes to help instead.