{ "schemaVersion": "wenbu-knowledge-v1", "id": "guide-ai-divination", "slug": "ai-divination", "language": "en", "title": "Why an AI reading can feel accurate", "description": "Separate calculation, interpretation and personal association, and avoid confusing fluent language with evidence of predictive accuracy.", "sections": [ { "heading": "Feeling understood and checking a claim", "paragraphs": [ "Feeling understood by a sentence is a real experience. It is different from demonstrating that a system can predict future events. A description can be meaningful to you without establishing predictive validity.", "Broad statements can fit many people. We also tend to remember matches more readily than misses. Fluent, reassuring AI prose does not remove these ordinary tendencies." ] }, { "heading": "Keep three layers separate", "paragraphs": [ "The first layer is calculation: which stems and branches follow from a date and convention, or which cards were drawn. The second is a traditional symbolic association. The third is your personal connection between that symbol and an experience.", "All three can be discussed, but they are not the same type of evidence. Precise calendar or astronomical data does not validate a personality or future-event prediction. Carlson’s study is one historical example of controlled testing in astrology, not a comprehensive verdict on every tradition." ] }, { "heading": "Build a more careful reading habit", "paragraphs": [ "Look for visible inputs, conventions, sources and uncertainty. Ask whether a reading identifies its basis, allows disagreement and leaves decisions with you. A claim that cannot be contradicted is difficult to use as predictive evidence." ], "bullets": [ "Record misses as well as matches.", "Separate what happened, what you infer and what you hope.", "Do not pay or make major decisions in response to frightening predictions.", "When another person is involved, prefer direct communication to alleged mind-reading." ] }, { "heading": "How Wenbu uses a model", "paragraphs": [ "Code calculates charts and performs random draws. DeepSeek receives the result with the question and context you choose to submit. It is instructed to use the calculated result and avoid invented classical quotations. Its interpretation and any references still need checking.", "The interface records the requested model and the model reported by the service. An articulate answer can still be wrong. Concrete bug reports are welcome, with private information removed." ] }, { "heading": "Check the sentence that feels most personal", "paragraphs": [ "If an answer says “You value independence but also want to be understood,” consider how many people it could describe. Find a specific example and a counterexample from your experience. A sentence can be useful without demonstrating that the system has identified something uniquely true about you.", "Ask the Agent to separate a paragraph into inputs or calculated results, traditional meanings and inferences from your context. Have it identify any unsupported claim. The source-reading guide gives a fuller method." ] } ], "updated": "2026-09-30", "authorship": "Wenbu · AI-assisted editorial; no independent expert review claimed", "evidenceType": "editorial", "links": { "html": "https://wenbu.app/en/learn/ai-divination/", "markdown": "https://wenbu.app/knowledge/en/ai-divination.md", "json": "https://wenbu.app/knowledge/en/ai-divination.json" }, "outline": [ { "id": "quick-answer", "title": "The short answer" }, { "id": "section-0", "title": "Feeling understood and checking a claim" }, { "id": "section-1", "title": "Keep three layers separate" }, { "id": "reference-table", "title": "Five different things to evaluate" }, { "id": "section-2", "title": "Build a more careful reading habit" }, { "id": "section-3", "title": "How Wenbu uses a model" }, { "id": "section-4", "title": "Check the sentence that feels most personal" }, { "id": "worked-example", "title": "Worked example: checking a sentence that feels uncannily personal" }, { "id": "glossary", "title": "Key terms" }, { "id": "questions", "title": "Common questions" }, { "id": "sources", "title": "Sources and their scope" } ], "answer": "An AI reading may feel accurate because it uses details you supplied, offers a broadly applicable description or helps you organize a real concern. First define what you are evaluating: reproducible calculation, a supported statement, practical usefulness or a tested prediction. These are different questions, not one accuracy score.", "takeaways": [ "Check calculation, personal resonance and usefulness separately.", "Recognizing a concern you already shared is not the discovery of an undisclosed fact.", "An accuracy percentage needs defined outcomes, data, a baseline and a validation method." ], "figure": { "caption": "Four ways a reading can seem to fit", "description": "The cards select restated context, reproducible calculation, personal resonance and practical help, with an example of each. None presents a predictive test result. The complete table below adds predictive performance and explains what evidence each of the five dimensions requires." }, "table": { "title": "Five different things to evaluate", "columns": [ "Dimension", "Example", "What it establishes" ], "rows": [ [ "Restated context", "You mention overtime; the answer suggests rest.", "Use of supplied information, not independent discovery." ], [ "Reproducible calculation", "Identical dates and conventions produce identical pillars.", "Consistency under specified inputs, rules and software versions." ], [ "Personal resonance", "You value independence and also want understanding.", "Ask for examples and counterexamples, and how widely the description applies." ], [ "Practical help", "The answer helps you write a message you can use.", "Usefulness for a task, not prediction of the recipient’s response." ], [ "Predictive performance", "An event is said to occur within a fixed period.", "Needs predefined outcomes, complete records and a meaningful comparison baseline." ] ] }, "example": { "title": "Worked example: checking a sentence that feels uncannily personal", "intro": "Consider this fictional reply: “You carry many responsibilities independently, yet want someone to understand you. A turning point is coming.” Review each part before accepting it.", "steps": [ "Check your earlier messages. If you wrote that you were managing a project alone, the responsibility statement has an input source. Mark it as a summary rather than hidden insight.", "Examine the broad description. Independence and wanting understanding can coexist for many people. Note a fitting example and a counterexample, such as a time you actively sought collaboration.", "Ask what “turning point” means, by when, and what would count against it. Do not retrospectively turn a vague phrase into a precise prediction it never made.", "Request a revision: “Separate my supplied details from general associations. Remove the unsupported future claim and give me two questions to discuss about the project.”", "Keep the useful result and its limit: “I can ask which tasks a colleague could share. I have no evidence of an inevitable turning point.” Assess whether the questions help, without treating a later coincidence as validation of the entire system." ], "conclusion": "Forer’s work cautions against using personal agreement alone to validate a personality interpretation. It did not test modern DeepSeek models or establish a Wenbu accuracy rate." }, "faq": [ { "question": "Does a classical citation prove predictive accuracy?", "answer": "It may document a traditional meaning. It still needs verification as a citation and cannot, by itself, establish a personal prediction." }, { "question": "Why can rephrasing change the answer?", "answer": "Prompts and context influence generated interpretations. Compare their assumptions; different prose does not mean your underlying chart or future has changed." }, { "question": "Is there a validated personal-prediction accuracy rate for Wenbu?", "answer": "This guide has no validation study supporting such a rate and reports none. Public calculation conventions and product behavior can be checked separately." } ], "glossary": [ { "term": "Personal agreement", "definition": "Feeling that a description fits; this alone does not show that it distinguishes you from others." }, { "term": "Baseline", "definition": "A comparison method used to judge whether a prediction adds information." }, { "term": "Confabulation", "definition": "Generated content that is false, unsupported or inconsistent with the input, even when confidently expressed." } ], "sources": [ { "title": "Forer (1949) · The fallacy of personal validation", "url": "https://pubmed.ncbi.nlm.nih.gov/18110193/", "scope": "Bibliographic record for the original personal-validation study. It is historical research context, not a test of modern language models or Wenbu." }, { "title": "NIST AI 600-1 · Generative Artificial Intelligence Profile (2024)", "url": "https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.600-1.pdf", "scope": "Section 2.2 describes confident false output; MS-2.5-003 calls for checking sources and citations. This is not an accuracy evaluation of Wenbu or divination." }, { "title": "Wenbu · AI reading input and interpretation rules", "url": "https://github.com/Digidai/wenbu/blob/main/worker/ai.ts", "scope": "Supports supplying calculated results and user context for symbolic interpretation; prompt rules do not guarantee correct output." }, { "title": "Wenbu · Agent evidence and uncertainty rules", "url": "https://github.com/Digidai/wenbu/blob/main/worker/agent.ts", "scope": "Supports the requirements to distinguish calculation, tradition and inference, and to avoid invented predictions, sources or personal history." }, { "title": "Carlson (1985) · A double-blind test of astrology", "url": "https://www.nature.com/articles/318419a0", "scope": "Further reading; not validation of real-world predictions." }, { "title": "DeepSeek · model update and alias behavior", "url": "https://api-docs.deepseek.com/news/news260910/", "scope": "Further reading; not validation of real-world predictions." }, { "title": "lunar-typescript · calculation source and examples", "url": "https://github.com/6tail/lunar-typescript", "scope": "Further reading; not validation of real-world predictions." } ] }