AI statement

What the AI writes

The parts written for you: the preface, the outline, the seven readings, a reading and a question for every chapter, the closing, and the answers to your questions. These are generated by AI from your own account.

The fixed content (plain explanation, glossary, commentary selections and key points for all thirty-two chapters) is generated by the same process, then read chapter by chapter by people before it is frozen; it is still a draft, and the About page says how far review has got.

The scripture itself never passes through the model; it is read directly from our corpus. Commentary quotations can only be taken from the corpus and are checked character by character.

Which models

Anthropic's Claude models. Their role is that of an editor and fellow reader: explain the scripture plainly, then bring it to your situation. They are not an authority, do not play the Buddha and are not a teacher.

How we keep it honest

Writing and review are done by two independent prompts; the reviewer has not seen the writing process. The review checklist covers doctrinal red lines (for example never saying “everything is empty so nothing matters”, never saying reciting the sutra changes your luck), quotation checks, tone and boundaries. A book that fails the program checks is not delivered.

Every book's generation process is regression-tested against a set of evaluation profiles (11 today, growing): crisis detection must miss none, and at least 95% must pass the program checks, or the change is not released.

Known limits

It may misread your situation: if a sentence is wrong, report it from My Book. It does not know what you did not say. It will not make decisions for you, and should not.

What this is not

Not therapy, not counselling, not a companionship chatbot. Three questions per book; no memory across books; it does not probe your feelings unprompted.