AI Writing De-Slop Tool

The Problem It Solves: The "Plastic Feel" of AI Text

You've likely encountered paragraphs that are grammatically perfect and logically coherent, yet feel exhausting to read—your brain keeps predicting the next sentence only to be "hit" by the safest, most boring word, leading to distracted attention and zero recall after finishing. This is the core symptom of "AI flavor": the uncanny valley of writing. It stems from the probabilistic nature of large language models—they always choose the safest, most average word combinations. This results in a series of signature issues:

• Syntactic Traps: High-frequency repetition of patterns like "不是A而是B" (It's not A, but B) and its variant "不是X。不是Y。只是Z。" (It's not X. It's not Y. It's just Z.), "值得注意的是" (It's worth noting), "让我们来看看" (Let's look at).
• Straw Man Logic: First erecting a viewpoint nobody actually holds (the straw man), then refuting it to feign depth while being hollow.
• Synonym Substitution for Contrast: "不是在演疯，而是在演失控感" (Not acting crazy, but acting out a loss of control)—where "crazy" and "loss of control" are essentially the same thing, forcing a false transition.
• Other Hallmarks: Dense parallelism, vague metaphors ("The sky is like a greasy shroud"), false precision ("He took a 49cm stride"), and high-frequency warning words like "说白了" (simply put) or "意味着什么" (what it means).

This writing de-slop tool doesn't aim to fix grammar errors but to combat this "plastic feel" and the resulting reading fatigue, restoring human warmth and specificity to the text.

How It Works: An Engineering-Driven Rewriting Pipeline

This skill isn't a simple "find and replace" tool; it's a complete processing pipeline that integrates multiple open-source projects (like stop-slop, Humanizer-zh) and deep research findings. Its workflow is clear:

1. Scanning & Diagnosis: It first performs a rapid scan, marking 11 types of issues (e.g., banned words, the three poisons of "不是而是"-straw man/synonym substitution/forced contrast, AI hallucinations, punctuation violations) and automatically diagnoses the text as Mild, Moderate, or Severe based on metrics like banned word density and consecutive parallel paragraphs.
2. Three-Pass Progressive Rewriting:
   • Pass 1: De-generalization. For "Mild" cases or as a base step. Externalizes abstract psychological descriptions ("He was nervous" → "His hands were shaking"), breaks up continuous parallelism, and repairs forbidden punctuation.
   • Pass 2: De-formalization. For "Moderate" cases. The core is handling constructions like "不是A而是B"—after judging the specific poison (straw man, synonym substitution, or forced contrast), it applies strategies like deletion, merging, or converting to a progressive "不仅是A更是B" (Not only A but more so B). It also trims connectors and clears piled-up modifiers.
   • Pass 3: Restoring Natural Flow. For "Severe" cases or final enhancement. It creates an "unexpected" reading experience through alternating long and short sentences, paragraph breaks, colloquial expressions, and sensory details, while allowing some imperfect "human" traces.
3. Four-Layer Self-Check System: After rewriting, the text undergoes checks from L1 (Hard Rules) to L4 (Human Feel Final Review). L1 scans ensure no banned words or punctuation violations remain; L4 ultimately judges from a reader's perspective: "Does this read like it was written by a human?"

Applicable Scope & Considerations

• This is an Aid, Not a "Soul Injector". The tool can significantly improve the mechanical feel of text, but the core viewpoint, unique perspective, and genuine emotion ultimately depend on the author. It is best suited for optimizing drafts initially generated or polished by AI.
• "De-slopping" Does Not Mean "Dumbing Down". Its goal is to make text more natural and specific, not to oversimplify it or make it so colloquial that it loses its logic. Professionalism and depth can be preserved through natural narration.
• Output Requires Human Review. Despite the self-check system, "whether it's natural enough" is ultimately a subjective judgment. It's recommended to use the tool's output as an optimized starting point, which the author then gives a final polish to ensure it aligns with the overall style.