Description

A Memorandum on the Core Principles of AI Rate Reduction: The core recognition logic of AI detection tools (such as CNKI AIGC detection, GPTZero, Originality.ai, etc.) is based on the following statistical characteristics: 1. Perplexity: AI-generated text has low overall perplexity because it selects the "next word with the highest probability." Human writing tends to choose more "less common but appropriate" words, resulting in high perplexity. → Rewriting Strategy: S4 Lexical De-standardization. 2. Burstiness: AI-generated sentences have highly uniform lengths, resulting in low burstiness. Human writing has highly variable sentence lengths, sometimes extremely short and sometimes extremely long, resulting in high burstiness. → Rewriting Strategy: S1 Sentence Length Mutation Injection. 3. Transition Word Density: AI over-relies on explicit transition words ("in addition," "however," "therefore," "in conclusion"), resulting in abnormally high density. Humans use more implicit semantic connections. → Rewriting Strategy: S2 Template Word Removal. 4. **Structural Templates:** AI tends to use a three-part structure ("general-specific-general"), parallel items of equal length, and a numbered structure ("first-second-last"). Human structures are more flexible and varied. → Rewriting Strategy: S3 – Disrupt the structure. 5. **Lexical Diversity (TTR):** AI has a high word repetition rate in long texts, resulting in a low TTR. Humans naturally have a wider vocabulary. → Rewriting Strategy: S4 – Destandardize vocabulary. 6. **Linearity of Information Presentation:** AI tends to present information in a logical, linear fashion with uniform information density. Humans use non-linear techniques such as flashbacks, digressions, and contrasting jumps. → Rewriting Strategy: S6 – Reorganize information density. Special Constraints of Textbook Style: Textbook AI reduction cannot be as casually colloquial as ordinary online articles. It must adhere to the following: • **Academic Rigor Bottom Line:** No use of internet slang, overly colloquial expressions, or emoticons. • **Terminology Accuracy Bottom Line:** All subject-specific terminology must be preserved in its original form; "synonyms" for inaccurate statements are prohibited. • Logical Integrity: The logic of causal relationships, classification systems, and operational steps must not be altered during rewriting. • Citation Standards: The numbers of cited documents, regulations, and standards must not be falsified. • Instructional Applicability: The rewritten text must still be suitable for student reading and comprehension; readability should not be sacrificed to reduce AI scores.