Back to Details

ljg-paper

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

ljg-paper: 读论文

ljg-paper: Read Papers

读论文不是做学术，是猎取思想。把别人的发现拆解成自己能用的认知。
Reading papers isn't about doing academia—it's about hunting for ideas. Break down others' findings into actionable knowledge you can use.

格式约束

Format Constraints

Org-mode 语法

Org-mode Syntax

  • 加粗用 
    *bold*
    （单星号），禁止 
    **bold**
  • 标题层级从 
    *
    开始，不跳级
  • Use 
    *bold*
    (single asterisk) for bold text; 
    **bold**
    is prohibited
  • Heading levels start with 
    *
    and must not skip levels

ASCII Art

ASCII Art

所有图表用纯 ASCII 字符。允许：
+ - | / \ > < v ^ * = ~ . : # [ ] ( ) _ , ; ! ' "
和空格。禁止 Unicode 绘图符号。
All diagrams must use pure ASCII characters. Allowed characters: 
+ - | / \ > < v ^ * = ~ . : # [ ] ( ) _ , ; ! ' "
and spaces. Unicode drawing symbols are prohibited.

模板权威性

Template Authority

输出结构依据 
references/template.org
。禁止参考 
~/Documents/notes/
中已有论文文件的章节结构——旧文件可能使用过期模板。
Output structure must follow 
references/template.org
. Do not reference the chapter structure of existing paper files in 
~/Documents/notes/
—old files may use outdated templates.

Denote 文件规范

Denote File Specifications

  • 时间戳：
    date +%Y%m%dT%H%M%S
  • 可读时间：
    date "+%Y-%m-%d %a %H:%M"
  • 文件名：
    {时间戳}--paper-{简短标题}__paper.org
  • 输出目录：
    ~/Documents/notes/
  • Timestamp: 
    date +%Y%m%dT%H%M%S
  • Human-readable time: 
    date "+%Y-%m-%d %a %H:%M"
  • File name: 
    {timestamp}--paper-{short-title}__paper.org
  • Output directory: 
    ~/Documents/notes/

Org 文件头

Org File Header

#+title:      paper-{简短标题}
#+date:       [{YYYY-MM-DD Day HH:MM}]
#+filetags:   :paper:
#+identifier: {YYYYMMDDTHHMMSS}
#+source:     {URL 或来源描述}
#+authors:    {作者列表}
#+venue:      {发表场所/年份}
文件写入后报告路径。
#+title:      paper-{short-title}
#+date:       [{YYYY-MM-DD Day HH:MM}]
#+filetags:   :paper:
#+identifier: {YYYYMMDDTHHMMSS}
#+source:     {URL or source description}
#+authors:    {author list}
#+venue:      {publication venue/year}
Report the file path after writing.

红线（每条必须过）

Red Lines (Must Comply with Each)

  1. 口语检验 — 你会这样跟朋友介绍一篇论文吗？不会→改。学术腔是默认敌人
  2. 零术语 — 先用大白话落地，再顺带提术语名。如果必须用原文术语才能解释，说明还没懂
  3. 短词优先 — 能用两个字说的不用四个字。「本文提出了一种新的框架」→「他们做了个东西」
  4. 一句一事 — 每句只推一步
  5. 具体 — 名词看得见，动词有力气。形容词能砍就砍
  6. 开头给理由 — 问题部分的第一句让人想知道答案
  7. 不填充 — 删学术套话（「近年来随着...的发展」「值得注意的是」）。每句干活
  8. 信任读者 — 说一遍够了。不重复结论
  9. 诚实 — 论文有硬伤就说有硬伤。看不懂的部分说看不懂
  1. Colloquial Test — Would you introduce a paper to a friend like this? If not, revise. Academic jargon is the default enemy
  2. Zero Jargon First — Start with plain language, then mention technical terms only if necessary. If you must use the original term to explain, it means you haven't fully understood it
  3. Prefer Short Phrases — Use two characters instead of four if possible. "This paper proposes a new framework" → "They built something"
  4. One Idea per Sentence — Each sentence only advances one point
  5. Be Specific — Nouns should be tangible, verbs should be powerful. Cut adjectives whenever possible
  6. Lead with the "Why" — The first sentence of the problem section should make readers want to know the answer
  7. No Filler — Remove academic clichés (e.g., "In recent years, with the development of...", "It is worth noting that..."). Every sentence must serve a purpose
  8. Trust Your Readers — Say it once, that's enough. Don't repeat conclusions
  9. Be Honest — If the paper has flaws, state them clearly. If you don't understand a part, say so

写作原则

Writing Principles

四条核心原则，决定文章是"活人在说话"还是"机器在汇报"：
  1. 一个锚点撑全文 — 找到一个具象的中心隐喻（一张图、一个场景、一个动作），让所有概念围绕它生长。不是并列罗列五个概念，是一根绳子串起来。锚点在「翻译」开头就要出现，后续章节可以反复回到它
  2. 推理外显 — 模拟"一个人想明白的过程"，而非呈现"想明白之后的结果"。用"既然A是B，那能不能C也是D？"带读者一起推。让读者觉得结论差一步就是自己想到的
  3. 变形替代定义 — 解释两个概念的关系时，把A连续变形成B，不要说"A和B是XX关系"。「把LSTM变形→看起来像ResNet」比「LSTM和ResNet是对偶的」有力十倍
  4. 落点在能用 — 给出"这意味着你可以___"，而非"这让我们重新思考___"。读者读完要带走一个能动手的东西，不是一个值得沉思的感慨
Four core principles that determine whether the content sounds like "a real person talking" or "a machine reporting":
  1. Anchor the Whole Text with One Metaphor — Find a concrete central metaphor (a diagram, a scenario, an action) and let all concepts revolve around it. Don't list five concepts side by side—string them together with one thread. The anchor should appear at the start of the "Translation" section, and you can circle back to it in subsequent chapters
  2. Explicit Reasoning — Simulate "the process of someone figuring things out" instead of presenting "the result after figuring it out". Use phrases like "Since A is B, can C also be D?" to guide readers through the reasoning. Make readers feel they almost arrived at the conclusion themselves
  3. Replace Definitions with Transformations — When explaining the relationship between two concepts, continuously transform A into B instead of saying "A and B have XX relationship". "Transform LSTM → it looks like ResNet" is ten times more impactful than "LSTM and ResNet are dual"
  4. End with Actionable Takeaways — Provide "This means you can ___" instead of "This makes us rethink ___". Readers should take away something they can act on, not a thought-provoking sentiment

工具箱（选用）

Toolkit (Optional)

讲解论文时可以拿的工具，没有哪个是必须的：
  • 类比 — 承重的，方法的关键组件都能映射上。沿着类比走一遍方法
  • ASCII 图 — 展示组件关系、数据流、结构对比。读者有概念脚手架后再画
  • 餐巾纸速写 — 「以前这么想，现在应该这么想」的并排对比
  • 好问题 — 把论文解决的困境变成一个让外行也好奇的问题
  • 递进例子 — 从简单到复杂，一步步搭建理解
  • 反问入链 — 遇到隐含假设，用问题打开
Tools you can use when explaining papers—none are mandatory:
  • Analogy — A robust analogy where key components of the method can be mapped. Walk through the method along the analogy
  • ASCII Diagram — Show component relationships, data flows, or structural comparisons. Draw it only after readers have a conceptual scaffold
  • Napkin Sketch — Side-by-side comparison of "how we used to think" vs. "how we should think now"
  • Good Questions — Frame the dilemma solved by the paper into a question that even laypeople would find curious
  • Progressive Examples — Build understanding step by step from simple to complex
  • Rhetorical Questions to Link Ideas — When encountering implicit assumptions, use questions to open up the discussion

执行

Execution

1. 获取内容

1. Acquire Content

  • arxiv URL → WebFetch
  • PDF → Read（注意 pages 参数限制）
  • 本地文件 → Read
  • 论文名称 → WebSearch
确保拿到：标题、作者、摘要、核心方法、结果。
  • arXiv URL → WebFetch
  • PDF → Read (note page parameter limits)
  • Local file → Read
  • Paper title → WebSearch
Ensure you obtain: title, authors, abstract, core method, results.

2. 定位：它在解决什么？

2. Positioning: What Problem Is It Solving?

找到那个真实的困境——某件事做不到、某个现象解释不通、某条路走不下去。用一段话讲清来龙去脉。
不是「本文提出了一种新的 XXX 框架」，是「大模型明明很聪明，为什么一问具体事实就开始胡说？」
Find the real dilemma—something that can't be done, a phenomenon that can't be explained, a path that leads nowhere. Explain the context in one paragraph.
Instead of "This paper proposes a new XXX framework", say "Large models are so smart, why do they start spouting nonsense when asked about specific facts?"

3. 费曼：让外行懂

3. Feynman Technique: Make It Understandable to Laypeople

把论文的核心想法讲到一个不懂这个领域的聪明人能跟上。形式自由——类比、图、例子、递进讲解，选最适合这篇论文的方式。
开头先立锚点：找到一个具象的中心隐喻或画面，在翻译的第一段就亮出来。后面所有概念围绕这个锚点生长，不是并列罗列。
推理带着读者走：不要直接给结论。模拟"一步步想明白"的过程——"既然X是这样，那Y能不能也这样？"让读者觉得结论差一步就是自己想到的。
需要覆盖：
  • 它怎么做的（核心机制/方法）
  • 做出来效果如何（挑最说明问题的两三个结果）
  • 理解全文需要的钥匙概念（如果有）
费曼翻译部分的子标题按内容需要组织，不必固定。
Explain the paper's core ideas in a way that a smart person outside the field can follow. Feel free to use analogies, diagrams, examples, or progressive explanations—choose what works best for the paper.
Set the anchor first: Find a concrete central metaphor or image and present it in the first paragraph of the "Translation" section. All subsequent concepts should revolve around this anchor, not be listed separately.
Guide readers through reasoning: Don't give conclusions directly. Simulate "the process of figuring things out step by step"—use phrases like "Since X is like this, can Y also be like this?" Make readers feel they almost arrived at the conclusion themselves.
You need to cover:
  • How it works (core mechanism/method)
  • What results it achieved (pick 2-3 most illustrative results)
  • Key concepts needed to understand the full text (if any)
Subheadings in the Feynman translation section can be organized as needed—no fixed structure required.

4. 核心概念：把术语变成直觉

4. Core Concepts: Turn Jargon into Intuition

挑出论文中最关键的 1 至 3 个概念（方法名、架构组件、数学对象、新定义……），逐个拆解。
每个概念：
  • 一句话：这东西是什么，干什么用的
  • 类比或例子：让没接触过的人秒懂。解释两个概念的关系时，优先用"把A变形成B"而非"A和B是XX关系"——变形比定义有力
  • 为什么重要：少了它论文的逻辑链断在哪里
选概念的标准：读者如果不懂这个，后面的洞见和审稿就跟不上。已经在「翻译」里讲透的不重复选。
Select 1 to 3 most critical concepts from the paper (method names, architecture components, mathematical objects, new definitions, etc.) and break them down one by one.
For each concept:
  • One sentence: What it is and what it's used for
  • Analogy or example: Make it instantly understandable to someone who's never encountered it. When explaining relationships between two concepts, prioritize "transform A into B" over "A and B have XX relationship"—transformation is more impactful than definition
  • Why it matters: Where the paper's logical chain would break without it
Selection criteria: If readers don't understand this concept, they won't follow the insights and review later. Don't repeat concepts that were already fully explained in the "Translation" section.

5. 洞见：思想结晶

5. Insights: The Core Idea

整篇论文最值钱的往往就一个点——作者真正找到的那颗新结晶。
用一句话把它说出来。这句话应该让读者觉得「这个想法我可以带走」，而不是「哦，论文说了这么个事」。
检验标准：把这句话单独抽出来，脱离论文上下文，它还有没有力量？如果只是在复述论文结论，那不是洞见。洞见是你读完之后自己看到的那个东西——论文里未必直说，但逻辑指向它。
说不出来就重读第三步。如果论文确实没有思想火花，直说「这篇论文是工程改进，没有认知层面的新发现」。不要硬挤。
Often, the most valuable part of a paper is just one point—the new "gem" the author actually discovered.
State it in one sentence. This sentence should make readers think "I can take this idea with me" instead of "Oh, the paper says this".
Test: If you extract this sentence out of the paper's context, does it still hold power? If it's just restating the paper's conclusion, it's not an insight. An insight is what you see after reading the paper—something the paper may not state explicitly, but the logic points to it.
If you can't articulate it, re-read step 3. If the paper truly has no intellectual spark, say "This paper is an engineering improvement with no new cognitive discoveries". Don't force it.

6. 博导审稿

6. Supervisor's Review

换身份：这个方向上带了二十年研究生的博导。学生拿着论文来找你，你判断这东西值不值得认真对待。
用白话说，像在办公室跟学生聊：
  • 选题眼光：问题值不值得做？真缺口还是人造缺口？
  • 方法成熟度：巧劲还是蛮力？有没有更自然的做法被忽略？
  • 实验诚意：baseline 公不公道？消融到位没？数字经不经得起追问？
  • 写作功力：最该说清楚的地方有没有偷懒？
  • 判决：strong accept / weak accept / borderline / weak reject / strong reject，一句话理由
好的说好，差的说差在哪儿。
Switch roles: You're a supervisor who's guided graduate students in this field for 20 years. A student brings you this paper—judge whether it's worth taking seriously.
Speak plainly, like chatting with a student in your office:
  • Topic Selection: Is the problem worth solving? Is it a real gap or an artificial one?
  • Method Maturity: Does it use clever thinking or brute force? Are there more natural approaches that were overlooked?
  • Experimental Integrity: Are the baselines fair? Are ablation studies thorough? Can the numbers withstand scrutiny?
  • Writing Quality: Did the authors skip explaining the most critical parts?
  • Verdict: strong accept / weak accept / borderline / weak reject / strong reject, with one-sentence reasoning
Praise the good parts, and clearly point out the flaws in the bad parts.

7. 启发：对我的提醒

7. Inspiration: Reminders for Me

落点在"能用"，不在"能想"。给出"这意味着你可以___"，而非"这让我们重新思考___"。
用三个视角试探连接，命中展开，没命中跳过，全没命中说「没有」：
  • 迁移：论文的某个机制/视角能移植升级我体系的某个零件吗？具体怎么接？
  • 混搭：论文的某个组件和我已有的东西组合能产生新东西吗？产出什么？
  • 反转：论文的做法和我的默认假设相反吗？该停下什么、开始什么？
Focus on "actionable" rather than "contemplative". Provide "This means you can ___" instead of "This makes us rethink ___".
Explore connections from three perspectives—expand on the ones that resonate, skip the ones that don't, and say "None" if none do:
  • Migration: Can a mechanism/perspective from the paper be transplanted to upgrade a part of your own system? How exactly?
  • Mix-and-Match: Can a component from the paper combine with something you already have to create something new? What would the output be?
  • Reverse Thinking: Does the paper's approach contradict your default assumptions? What should you stop doing, and what should you start doing?

8. 过红线

8. Check the Red Lines

逐条扫红线。额外检查：
  • 破公式——否定式排比全文不超过两处，三段式改两项或四项
  • 变节奏——长短句交替
  • 杀金句——听起来像可引用的，重写
  • 查跳跃——逻辑每步可追
列修改清单确认后生成文件。
Go through each red line one by one. Additional checks:
  • Avoid Formulaic Phrases: No more than two negative parallel structures; change three-part structures to two or four parts
  • Vary Sentence Rhythm: Alternate between long and short sentences
  • Overwrite "Quotable" Phrases: If a sentence sounds like a generic quote, rewrite it
  • Check for Logical Leaps: Every step of the reasoning must be traceable
Generate the file only after confirming all revisions.

9. 生成 Org 文件

9. Generate Org File

按 Denote 规范获取时间戳，读 
references/template.org
，写入 
~/Documents/notes/
。
Obtain the timestamp according to Denote specifications, read 
references/template.org
, and write the file to 
~/Documents/notes/
.

验收

Acceptance Criteria

  • 问题勾人：让不懂的人也想知道答案
  • 有锚点：翻译部分有一个具象的中心隐喻，后续概念围绕它生长
  • 带着推：读者能感受到"一步步想明白"的过程，而非接收打包好的结论
  • 外行能跟：不懂这个领域的聪明人读完能复述核心思路
  • 博导像博导：有判断力有分寸，最后一句判决
  • 启发能动手：启发部分的落点是"你可以___"，不是"值得思考___"
  • 零割裂感：读完像一个人在跟你说「我读了篇论文，发现了个有意思的事」
  • Engaging Problem: Makes even laypeople want to know the answer
  • Has an Anchor: The "Translation" section has a concrete central metaphor, with all subsequent concepts revolving around it
  • Guided Reasoning: Readers can feel "the process of figuring things out step by step" instead of receiving a packaged conclusion
  • Layperson-Friendly: A smart person outside the field can retell the core idea after reading
  • Supervisor-like Judgment: Has clear judgment and appropriate tone, with a final verdict
  • Actionable Inspiration: The inspiration section ends with "You can ___" instead of "It's worth thinking about ___"
  • Seamless Flow: After reading, it feels like a person saying "I read a paper and found something interesting"