一孔录Yikonglu uses zhiqi (智器, our word for AI) to make videos that read the classics. Zhiqi can invent commentary, invent sources and mispronounce characters, so we set out here which steps use zhiqi, which must be checked by a person, and what we do about mistakes — for readers to check us against, and for others doing similar work.
1. Base editions and commentaries
- Before a book begins, we fix its base edition and list of commentaries on the book's page. For the Analects, the text and Zhu Xi's commentary follow the Sibu Beiyao edition of the Lunyu Jizhu; older commentaries use public-domain woodblock editions (Huang Kan's Lunyu Yishu: the Zhibuzu Zhai edition and a 1750 Japanese edition); modern commentaries still in copyright (Yang Bojun, Qian Mu) are quoted only briefly, with page numbers. See Editions & licenses.
- We never use web texts of unknown origin as a base.
2. How each chapter is made
- Text: entered character by character from the base edition; variants go in a variants table; nothing is silently changed.
- Pronunciation: each episode has a reading table; polyphonic and rare characters are fixed from the commentaries and standard readings, checked before narration, and compared line by line with speech recognition afterwards; mismatches are re-recorded. All readings are in the pronunciation index.
- Quotations: every quotation in a video is logged with source and page and checked word for word against the original page before it goes on screen; anything we cannot find on the page is only paraphrased, never shown as a quote.
- Original pages: when we cite a commentator, the scanned page goes on screen with the quotation boxed in red; the chapter page on this site shows the same page.
- Renderings and readings: we write our own plain renderings and check them line by line against standard translations, rewriting any match of six characters or more; “How Yikonglu reads it” is our own reading, labelled as ours, never passed off as a commentator's.
- Visuals: illustrations and characters are our own; type is cut from the base edition's glyphs and machine-scanned for damage; fictional characters in the videos do not assert “history” beyond the text.
3. Where zhiqi works, and where people do
We do not pretend every step is done by hand. The actual division of work:
| Step | Zhiqi does | People do |
|---|---|---|
| Script | Drafts; a second pass reviews against a checklist (cuts filler, checks names and attributions) | Choose editions and commentaries, decide the reading, review and revise |
| Quotations and sources | Lists them and compares them word by word with the page | Spot-check; nothing unverified goes on screen |
| Pronunciation | Narrates from the reading table; speech recognition compares line by line | Decide readings; re-record anything that sounds wrong |
| Plain rendering | Drafts; checks overlap with standard translations | Final text |
| Illustration and animation | Generates images, cuts type, renders, scans automatically | Choose images and set the style |
| Finished video | Automatic checks (quotations, readings, length, visuals) | Every episode is watched by a person and released only after it passes; notes go back into the script and visuals |
All videos are labelled as containing “AI-generated content”, following each platform's rules and wording.
4. When we get it wrong
- Errors in content (text, quotations, sources, pronunciation, readings), whether reported by readers or found by us, all go on the public corrections page, stating what we said, what it now says and why, and crediting the reader by their chosen name.
- Published videos are not deleted; where we can, we correct the description, and a remade version is published separately and noted on the corrections page.
- Our contact address is on the about page.
5. Numbers we publish
Every week we update the counts of chapters covered, quotations checkable against original pages, commentaries with page images, and public corrections, plus whether searching a line of the text on each platform finds us.
We also publish the share of segments changed between first draft and final script: each episode's earliest script is compared with the final one segment by segment, and any difference in wording counts as a change. So far: 20 episodes, 658 segments, 152 changed — 23.1% (updated 2026-11-06).
| Episode | Segments | Changed | Share |
|---|---|---|---|
| 学而一 | 32 | 1 | 3.1% |
| 学而二 | 30 | 12 | 40.0% |
| 学而三 | 39 | 13 | 33.3% |
| 学而四 | 29 | 6 | 20.7% |
| 学而五 | 30 | 8 | 26.7% |
| 为政一 | 39 | 10 | 25.6% |
| 为政二 | 35 | 10 | 28.6% |
| 为政三 | 35 | 8 | 22.9% |
| 为政四 | 32 | 9 | 28.1% |
| 为政五 | 35 | 12 | 34.3% |
| 为政六 | 35 | 9 | 25.7% |
| 八佾一 | 35 | 13 | 37.1% |
| 八佾二 | 35 | 9 | 25.7% |
| 八佾三 | 35 | 4 | 11.4% |
| 八佾四 | 36 | 5 | 13.9% |
| 八佾五 | 33 | 6 | 18.2% |
| 八佾六 | 30 | 4 | 13.3% |
| 八佾七 | 26 | 3 | 11.5% |
| 里仁一 | 30 | 5 | 16.7% |
| 里仁二 | 27 | 5 | 18.5% |
6. About this page
What this page describes is our promise to readers. If the method changes, this page changes, with a revision note below. Text licensed CC BY-NC 4.0; videos, narration, characters and illustrations are not covered.
Revisions
- 2026-11-06 First version.
- 2026-11-06 We now call AI “zhiqi” (智器).
