AI Fortune-Telling and Fabricated Citations

AI算命假引文 is a category of error found in large-language-model output about Chinese metaphysics. The model produces a sequence of characters that reads like a classical quotation, assigns it to a known or plausible book title, and sometimes adds a volume or chapter number. The result can look credible to a reader who does not check the source. The term identifies the failure precisely: the quotation is not real, even when the book named is real, and the problem is exposed by verifying the citation rather than by judging the style of the text.

为什么书名可以被生成出来

A language model does not retrieve a quotation from a stored edition and then type it out. It generates a sequence that is probable according to its training data. Classical Chinese prose on calendrical systems, stems and branches, or destiny analysis has strong recurrent phrasing, and a model can produce new text that fits that register. Book titles are also sequences, and many titles follow patterns such as 子平，通会，宝鉴, or 真诠. There is no mechanism that forces a generated title to correspond to a real printed work, nor one that forces a generated passage to match the contents of the title placed beside it. The model can therefore emit a real-sounding title, a plausible volume number, and a passage that appears to be quoted, all from the same generative process. This is why the question is not whether a model is intelligent enough to know a source, but whether the output is presented as a source at all.

真实存在的几部命理典籍

The problem is not that all cited books are invented. Several transmitted texts in the field are real and can be checked. These include 子平真诠，滴天髓，三命通会，穷通宝鉴, and 渊海子平. Their titles are well attested, and their contents exist in printed editions with identifiable structure. A real title can therefore be attached to a generated passage, and the presence of the real title gives the false quotation its authority. The existence of these works does not make the generated passage real; it only makes the error harder to notice at first glance. A reader who recognizes the title may lower the level of scrutiny, because the title itself is not suspicious. The correct response is to treat the title as a claim that must be checked, not as evidence that the passage came from the book.

一段引文的三项可查特征

For a passage to count as a quotation, three things must be present and checkable. First, there must be a definite book title, not a description of a kind of book. Second, there must be a definite volume, chapter, or section name that locates the passage within that book. Third, there must be an original text that can be compared character by character with the quoted passage. If any of the three is absent, the material cannot be treated as a quotation, regardless of how familiar the language sounds. If the title is real but no volume is given, the citation is incomplete. If a title and volume are given but the passage cannot be found, the quotation is false. If the passage is found but differs in wording, the quotation is inaccurate. The standard is simple: a real quotation can be checked, and a checked quotation either matches or it does not.

去归属化这个次生问题

When a generated quotation is challenged, a common secondary move is to remove the specific attribution while keeping the appearance of a written source. Instead of naming a book, the text may switch to 古籍，典籍，原文，记载, or 云. These terms still assert that somewhere there is a written record containing the statement. The specific title is gone, but the claim of a source remains. This is not a correction of the original error; it is a retreat from a falsifiable citation to an unfalsifiable one. A passage said to come from 古籍 cannot be checked, because no edition is identified. But the rhetorical force is similar: the reader is still being told that the statement is grounded in an old written tradition. The absence of a title does not make the statement neutral. It makes the source claim vague, and vagueness is a way of avoiding verification.

没有出处的说法怎么写才诚实

The practical rule is to treat vague attributions as no attribution at all. 古籍，典籍，原文，记载, and 云 are not weaker forms of citation; they are the absence of citation wearing the language of citation. If a statement cannot be traced to a definite book, volume, and original text, then for the purpose of the knowledge base it has no source. This does not mean the statement must be deleted. It means the statement must be described as unsourced. The alternative is worse: to call an unsourced statement a traditional saying or an ancient method is to invent a pedigree. The honest form is to say that the origin is not known, or that no verifiable source has been found, and to stop there.

The site's practice follows this distinction. In computational matters, the entry states the algorithm. In interpretive matters, the entry states which school or method it follows, when that can be identified. When a claim has no verifiable source, no book title is added. The purpose is not to eliminate uncertain material, but to prevent uncertain material from being mistaken for documented material. A reader can then see the difference between a method that is explained and a claim that is merely repeated.

One further point is that checking a quotation does not require expertise in Chinese metaphysics. The task is bibliographic, not doctrinal. A book title can be searched in a catalogue or a digital library. A volume or chapter number can be checked against a table of contents. A passage can be compared character by character with the text of an edition. These operations are open to anyone who can read the characters and use a search tool. The check does not depend on whether the reader understands the technical meaning of the terms. That is why false quotations are not a specialist problem. They are a citation problem, and the remedy is ordinary verification.

The category AI算命假引文 therefore names a specific failure: a generated text is presented as a quotation, but it lacks one or more of the three features that make a quotation checkable. The failure can occur with a real title or a fabricated one, and it can survive in softened form when the title is replaced by a vague phrase such as 古籍 or 云. The correct treatment is not to soften the claim further, but to classify it as unsourced. Where the site retains such claims, it writes them as having no known source, and it does not attach a book title that the evidence does not support.