human-write
SKILL.md
Human Write
請 AI 幫忙潤稿嗎?請避開以下詞!
基於 2025 年兩篇研究,分析 2,750 萬篇 PubMed 文獻,找出 AI 偏好的英文詞彙。
觸發時機
- 用戶請你潤飾英文文章
- 用戶請你寫英文學術/專業文章
- 用戶想檢查文章裡有沒有 AI 味的英文詞
- 用戶說 "human-write" 或 "scan AI words"
- 中文寫作請參考
chinese-patterns.md
Top 10 最嚴重的 AI 詞
基於 2025 年兩篇研究,分析 2,750 萬篇 PubMed 文獻,找出 AI 偏好的英文詞彙。
Top 10 最嚴重的 AI 詞
寫完或潤完稿,先搜這 10 個。出現任何一個都要三思:
| 排名 | 詞彙 | 為什麼可疑 | 替換方向 |
|---|---|---|---|
| 1 | delve | 2024 Z-score 最高,AI 最愛用 | explore, examine, look into |
| 2 | underscore | AI 用來替代 emphasize | emphasize, highlight, stress |
| 3 | meticulous | AI 最愛的形容詞 | careful, thorough, precise |
| 4 | intricate | AI 用來替代 complex | complex, detailed, elaborate |
| 5 | tapestry | AI 愛用的比喻詞 | mix, combination, array |
| 6 | commendable | AI 的誇獎用語 | impressive, notable, solid |
| 7 | showcase | AI 用來替代 show/demonstrate | show, demonstrate, present |
| 8 | pivotal | AI 用來替代 important/key | key, important, central |
| 9 | bolster | AI 用來替代 support/strengthen | support, strengthen, back up |
| 10 | surpass | AI 用來替代 exceed/outperform | exceed, beat, outperform |
完整 AI 偏好詞彙表
| AI 偏好 | 更自然的替代 |
|---|---|
| delve (into) | explore, look into, examine |
| underscore | emphasize, highlight, stress |
| showcase | show, demonstrate, present |
| bolster | support, strengthen, back up |
| elucidate | explain, clarify, make clear |
| garner | gain, attract, earn, get |
| leverage | use, take advantage of, build on |
| unveil | reveal, introduce, present |
| scrutinize | examine, inspect, review |
| surpass | exceed, beat, outperform |
| boast | have, feature, offer |
| foster | encourage, promote, support |
| facilitate | help, enable, make easier |
| harness | use, apply, put to use |
| navigate | handle, manage, deal with |
| streamline | simplify, improve, speed up |
| illuminate | show, reveal, shed light on |
| spearhead | lead, drive, champion |
| embark (on) | start, begin, launch |
| endeavor | try, attempt, work toward |
| encompass | include, cover, span |
| catalyze | trigger, spark, drive |
| augment | add to, increase, supplement |
| delineate | describe, outline, define |
| juxtapose | compare, contrast, set side by side |
| transcend | go beyond, rise above, exceed |
| AI 偏好 | 更自然的替代 |
|---|---|
| meticulous | careful, thorough, precise |
| intricate | complex, detailed, elaborate |
| pivotal | key, important, central, crucial |
| commendable | impressive, notable, solid, good |
| groundbreaking | new, innovative, first-of-its-kind |
| transformative | significant, game-changing, major |
| nuanced | subtle, complex, layered |
| comprehensive | full, complete, thorough |
| robust | strong, solid, reliable |
| multifaceted | complex, varied, diverse |
| noteworthy | notable, worth noting, significant |
| paramount | most important, top, critical |
| holistic | overall, whole, complete |
| overarching | main, broad, general |
| unprecedented | never before seen, first, new |
| salient | key, main, important |
| burgeoning | growing, rising, expanding |
| AI 偏好 | 更自然的替代 |
|---|---|
| primarily | mainly, mostly, largely |
| notably | especially, in particular |
| predominantly | mostly, mainly, largely |
| compellingly | convincingly, strongly |
| profoundly | deeply, greatly, strongly |
| moreover | also, besides, in addition |
| furthermore | also, and, in addition |
| importantly | (often deletable — just state it) |
| crucially | (often deletable — just state it) |
| remarkably | surprisingly, strikingly |
| substantially | greatly, significantly, a lot |
| inherently | by nature, naturally, fundamentally |
| AI 偏好 | 更自然的替代 |
|---|---|
| tapestry | mix, combination, range |
| realm | area, field, domain |
| landscape | field, area, scene, picture |
| synergy | teamwork, cooperation, combined effect |
| paradigm | model, approach, framework |
| nexus | link, connection, center |
| interplay | interaction, relationship |
| cornerstone | foundation, basis, key part |
| underpinning | basis, foundation, support |
| discourse | discussion, debate, conversation |
| trajectory | path, direction, trend |
| myriad | many, numerous, a wide range of |
| plethora | many, plenty, a lot of |
使用方式
模式一:潤稿時自動避開
幫用戶潤飾英文時,主動避開以上所有詞彙。如果原文沒有這些詞,潤稿後也不該出現。
規則:
- 不加入原文沒有的 AI 詞 — 潤稿是改善,不是加料
- 替換原文已有的 AI 詞 — 用更自然的替代詞,但先問用戶
- 保留用戶的聲音 — 如果用戶本來就愛用某個詞,別幫他換掉
模式二:掃描檢測
用內建腳本掃描文章,產生 AI 詞彙密度報告:
python3 ~/.claude/skills/human-write/scan-ai-words.py <file>
# 或從 stdin
cat essay.txt | python3 ~/.claude/skills/human-write/scan-ai-words.py
輸出格式
═══ AI Word Scan Report ═══
Score: 3.2 / 10 (moderate AI flavor)
Found 8 AI-flavored words in 1,200 words (0.67% density)
TIER 1 (top flagged):
delve ×2 (lines 5, 23)
meticulous ×1 (line 14)
TIER 2 (commonly flagged):
comprehensive ×3 (lines 8, 31, 45)
furthermore ×2 (lines 12, 38)
Suggestions:
Line 5: "delve into the mechanisms" → try "explore the mechanisms"
Line 14: "meticulous analysis" → try "careful analysis"
...
評分標準
| 分數 | 密度 | 解讀 |
|---|---|---|
| 0-2 | < 0.2% | 自然,幾乎無 AI 痕跡 |
| 3-4 | 0.2%-0.5% | 輕微,稍作替換即可 |
| 5-6 | 0.5%-1.0% | 中度,建議逐一檢查 |
| 7-8 | 1.0%-2.0% | 明顯,需要大量改寫 |
| 9-10 | > 2.0% | 強烈 AI 味,重寫吧 |
研究背景
兩篇關鍵論文
- Matsui K. (2025). Delving Into PubMed Records: How AI-Influenced Vocabulary has Transformed Medical Writing since ChatGPT. Perspectives on Medical Education, 14(1), 882-890.
- Kobak, D., et al. (2025). Delving into LLM-assisted writing in biomedical publications through excess vocabulary. Science Advances, 11(27), eadt3813.
核心發現
- 分析 2,750 萬篇 PubMed 紀錄 (2000-2024)
- 135 個 AI 偏好詞彙中,103 個在 2024 年出現統計上顯著暴增
- 趨勢從 2020 年就開始(ChatGPT 發布前兩年)
- 對照組 84 個標準學術用語大多穩定
雙相模式
- 第一階段 (2020-2022):DeepL、Grammarly 等早期 AI 工具開始影響詞彙選擇
- 第二階段 (2023-至今):ChatGPT 爆發,AI 詞彙被大規模放大 (β = 0.655, p < 0.001)
非英語母語研究者
- 來自中國、韓國、台灣等地的論文,AI 詞彙使用比例約 20%
- 英語國家約 5%
- 這不代表「不當使用」,但代表需要更注意審稿後編輯
重要提醒
- 詞彙「出現」不代表一定是 AI 寫的
- 關鍵是密度:人類偶爾用 delve 很正常,每段都用就不正常
- AI 沒有「創造」這些用法,但大大「放大」了它們
- 同時,一些傳統學術用語反而減少了:purpose of, hypothesis, results suggest
注意事項
- 這個清單基於醫學文獻研究,但 AI 偏好詞彙在各領域通用
- 不是說這些詞「不能用」,而是用之前要確認是自己的選擇
- 非英語母語者用 AI 潤稿完全合理,只要最後自己編輯一遍
- 人類文章有「凸點、凹洞、偏執與破綻」,這是特徵不是缺點
- AI 是草稿機,不是代筆人
Weekly Installs
18
Repository
htlin222/dotfilesGitHub Stars
75
First Seen
Feb 23, 2026
Security Audits
Installed on
opencode17
gemini-cli17
github-copilot17
codex17
kimi-cli17
amp17