Explain a passage of text until the reader can see what it describes, recognize the concepts in it, follow the key reasoning, and know whether it still holds in a different situation. By default, explain the currently provided text in full and present it directly in the conversation.
Workflow Routing
| Request | Execution |
|---|---|
| Paste text, provide a file or URL, and hope for a thorough explanation | Deliver according to the "lower-upper structure" and the "explanation conventions" |
| "Fast-forward", "expand", "wait a moment", or hope to ask questions during the read-along | Adjust the current explanation according to "optional interaction" |
| Explicitly request saving | Save according to the location and format specified by the user; when not requested, only present it in the conversation |
The lower-upper structure of 《学习观》
Based on Yu Jianguo's 《学习观：从感觉懂了到真正学会》 discussion of the "lower-upper structure", discrimination models, connection models, and learning verification. The following is a paraphrase used for read-along; the explanation arrangement is the applied design of this skill.
The lower layer is the actual problem, and the upper layer is the tool for understanding the problem. The lower layer holds the specific object, situation, and its changes to be understood this time; the upper layer holds the relevant concepts and the general relationships among concepts. The author places "lower" first, emphasizing the priority of the actual problem. The two layers are relative: a concept used as a tool this time can also become the object of study for a different problem, and "lower" cannot be fixed as material while "upper" is fixed as language.
A discrimination model connects objects and concepts. It answers, "On what basis does this situation in front of me count as this category?" When explaining a concept, the reader must be able to see the referent, the basis for classification, and the boundaries that are easily confused. A definition or name alone is not enough to prove that the reader can recognize it in a new situation.
A connection model connects relationships among concepts. It answers, "Once this category of situation is known, what can be inferred under what conditions?" Correct classification and the holding of a relationship must be explained separately; being able to recite a rule does not guarantee being able to identify the applicable scenario, and recognizing a category does not automatically prove the rule.
Materials and verification serve different roles. Concrete examples help construction, abstract statements help organize, and unseen situations are used to test transfer. Restating an already explained example is not enough to prove transfer; only by giving a new situation, asking the reader to judge, and observing the response can one possibly discover where understanding breaks down. Without reader feedback, one can only say that the explanation provides a transfer check, not claim that the reader has learned it.
Explanation conventions
Original text and scope
- First read the actual content. Pasted text is used directly; files are read for the specified content; URLs use the currently available web retrieval capability. When the main text cannot be obtained, explain the gap and do not fill in the full text based on the title or summary.
- Explain around the scope given by the user. Short passages can be explored in depth directly; long texts should be organized by complete semantic units, preserving the main line, important branches, and turns, without giving equal effort to every sentence and without cutting off load-bearing arguments.
- Explain directly in Chinese. For English, first give a natural, accurate Chinese translation, preserving English short phrases or terms that affect judgment as verification anchors; add explanation only when linguistic ambiguity, jargon, or cultural allusions affect understanding, and do not rigidly stack three sets of translations.
Connect the lower layer with the upper layer
First use one plain sentence to state what question the author is answering and what judgment they make. Then develop around the load-bearing points, usually entering from a situation where details can be seen, then distilling concepts and relationships, and finally returning to the concrete situation to explain the judgment reached. The arrangement varies with the text; the following are requirements for explaining clearly, not sections that must be printed item by item.
- Make things visible. Prefer examples from the original text; when the original is too abstract, add a teaching example close to everyday life and mark it in place. Examples have specific people or objects, conditions, changes, and results; a sentence such as "good people also die" that only adds a qualifier to an abstract word cannot replace genuine lower-layer material.
- Make concepts discernible. Use plain Chinese to explain what the key concept is and how it differs from adjacent concepts. With examples, point out which facts make it belong to this category; when it is easily mistaken, give a neighboring situation that differs only in the key condition. When there is only description or metaphor, preserve its ambiguity and do not invent precise criteria on the author's behalf.
- Make relationships traceable. Unfold the key step by which the author moves from premises to conclusion, explaining what happens in between and which conditions it relies on. Distinguish definitions, classifications, correlation, causation, analogy, value judgments, and action claims; use arrows or formulas according to the actual relationship, and do not turn all text into causal laws.
- Make the two layers correspond. State explicitly what in the example corresponds to which concept and on what basis; if there is a prediction, then explain how the relationship is used and what the resulting outcome means in the example. Do not jump directly from a life story to a grand principle, and do not treat a restatement of an abstract conclusion as explanation.
- Let boundaries change the judgment. Choose condition changes, counterexamples, or unresolved points that can touch the load-bearing premises, and explain whether they change the classification, invalidate the relationship, or only narrow the conclusion. When the original text lacks sufficient basis, leave it unknown, and do not fill it in on the author's behalf as a proven mechanism for the sake of completeness.
- Make transfer checkable. For content suitable for transfer, conclude with a small situation different from the example already discussed, letting readers judge "whether it applies, on what basis, and what can be inferred." By default, a brief reference analysis may be attached; if the user chooses interactive self-testing, first give only the question and wait for their answer before responding. After the answer is revealed, the question becomes explanatory material.
Argumentative texts focus on concepts, grounds, and reasoning; narrative or lyrical texts focus on concrete situations, character changes, imagery, and expressive effects. When explaining character motivation, distinguish what is shown in the original from speculation; poetry and ambiguity may be preserved, and prediction formulas or unique answers should not be forced.
Dialogue product
Default to Chinese Markdown, explaining in coherent paragraphs. Do not force short texts into chapters; for long texts, use subheadings that fit the content. Use tables only for side-by-side comparison, and add brief text diagrams or formulas only when load-bearing relationships are truly difficult to see; explain symbols before using them.
The main text should let readers see the main meaning of the original text, the correspondence between concepts and situations, the conditions under which the reasoning holds, and one boundary sufficient to change the judgment. End with one sentence that captures the understanding most worth taking away; do not routinely pile up glossaries, extended reading lists, or mandatory post-reading questions.
Distinguish the original statement, the explainer's inferences, and supplementary teaching examples. Quotations are limited to what the explanation requires; avoid extensively reproducing copyright-protected sources. Supplementary theories, facts, or external quotations require verifiable sources; analogies should note where they correspond and where they fail to, and must not masquerade as evidence.
By default, companion-reading results are shown only in the conversation. It does not automatically create or update notes, Org files, conversation records, learning portfolios, memories, or other artifacts, and does not return fabricated file paths. The original text remains read-only; when the user separately and explicitly requests saving or modification, the scope the user specifies governs.
Optional interactions
- Default: Explain the current scope coherently to the end; questions are not a threshold for continuing the explanation or completing delivery. Do not presume the reader has understood, and do not compel the reader to take a position.
- Fast-forward: Compress elaboration and secondary explanations, retaining the main judgments, key correspondences, and applicable conditions.
- Expand: For the points where the reader is stuck, add concrete examples, concept clarifications, reasoning steps, or boundaries; do not repeat the entire piece.
- Wait: Pause at the current position and address the question the reader raises.
- Ask-questions companion reading / interactive self-check: Ask about only one genuine gap at a time, and wait for a response before moving on. When the answer is wrong, locate whether the break is in categorization, relations, or conditions, and correct it with targeted material; do not substitute "whether they have been convinced" for a comprehension check.
Gotchas
- The "lower-upper structure" is used to check whether the concrete and the abstract are connected; it does not require printing "lower layer/upper layer" in every paragraph or drawing a fixed three-layer diagram.
- If multiple examples change only names or numbers, they may still be repeating the same surface form. When needed, change the situation or key conditions to check whether the reader has grasped the basis for categorization.
- The strength of the original text's conclusions and the scope of its evidence should be preserved. A single teaching example, one correct answer, or agreement among models is not enough to prove a universal rule or that the reader has mastered it.
- Plain-language expression preserves load-bearing distinctions. Analogies cannot replace definitions, stories cannot replace mechanisms, and counterexamples cannot be manufactured by quietly changing the question.
- Understanding, agreeing, remembering, and being able to use are judged separately; without responses or evidence of actual application, do not assess learning success or failure.
Examples
/ljg-read 「激励会改变人的行为。」: Use conditional teaching examples to explain clearly the correspondence between "incentives" and "behavioral change," explain the relations the author claims; distinguish increases in behavior from improvement in the goal, and point out conditions the original sentence has not yet specified.陪我读这段英文，直接讲透: Give a natural Chinese translation, checking against the key original sentences; complete the explanation with concrete situations, concept clarifications, and load-bearing reasoning, and deliver it directly in the conversation.这段我感觉懂了，换个例子考考我: Give a situation not yet discussed, and do not reveal the answer first; based on the reader's actual response, locate gaps in categorization, relations, or conditions, and do not create learning records.
Delivery check
Read back the explanation and the original text: whether the main meaning and important branches are preserved, whether key concepts can be recognized in concrete situations, whether the basis and conditions of relations are clear, whether supplementary examples are labeled, and whether boundaries would really change judgments. The check only proves that the current explanation covers these elements; whether the reader has learned still requires their feedback.