summarize-arxiv-paper

Summarize an arXiv paper and save it as a Markdown document in the machine-learning/ directory.

The user will provide an arXiv URL (e.g., https://arxiv.org/abs/2312.00752) as the argument.

Steps

1. Convert URL to ar5iv: Convert the arXiv URL from https://arxiv.org/abs/<ID> to https://ar5iv.labs.arxiv.org/html/<ID> to get an HTML-rendered version of the paper.
2. Fetch the paper: Use WebFetch to retrieve the paper content from the ar5iv URL. Use the prompt: "Extract the full paper content including: title, authors, abstract, all section headings and their content, mathematical formulas, algorithms, experimental results, datasets, and references."
3. Extract the publication year: Get the year from the arXiv ID (e.g., 2312.00752 → 2023, 1706.03762 → 2017).
4. Generate the filename: Run echo "<paper title>" | bash scripts/title-converter.sh to convert the paper title to a kebab-case filename.
5. Create the year directory if it does not exist: mkdir -p machine-learning/<year>/
6. Write the summary to machine-learning/<year>/<filename>.md following the Document Template below.

Document Template

The summary MUST follow this structure, modeled after existing documents in this repository:

# Meta Information

- URL: [<Paper Title>](<original arxiv URL>)
- LICENSE: [arXiv.org - Non-exclusive license to distribute](https://arxiv.org/licenses/nonexclusive-distrib/1.0/license.html)
- Reference: <authors> (<year>). <title>. <venue/journal>.

# <Follow the paper's own section structure>

## <Section headings matching the paper>

<Content summarized in concrete, specific sentences.>
<Mathematical notation using LaTeX/MathJax ($...$).>
<Specify tensor/matrix dimensions explicitly.>
<Use > [!NOTE] blocks for direct quotes or clarifications.>
<Use > [!TIP] blocks for helpful external references.>
<Use > [!IMPORTANT] blocks for critical information.>

# Experiments

- Dataset: <list all datasets used>
- Hardware: <if mentioned>
- Optimizer: <if mentioned>
- Results: <key quantitative results>

Summarization Rules

Follow these rules strictly when writing the summary. These come from the Definition of Done (DoD) checklist:

Common Requirements

• Write concrete, detailed sentences that demonstrate understanding (NEVER write vague statements like "I understand" or "this is important")
• Describe applicability conditions: who would use this, when, and where
• Include license and copyright information in the Meta Information section

ML/CS Requirements

• Clear Input/Output: For each major component or layer, specify the input and output dimensions (e.g., $x \in \mathbb{R}^{n \times d}$)
• Algorithms with pseudocode: Describe key algorithms using pseudocode or step-by-step mathematical formulation
• Datasets: Explicitly list all datasets used, including splits (train/dev/test sizes) when available
• Clear calculation order: Present mathematical operations in the order they are computed
• Differences from similar algorithms: Compare with related or predecessor methods, highlighting what is new

Style Conventions (from existing documents)

• Use > [!NOTE] for direct quotes from the paper
• Use > [!TIP] for links to external references or tutorials
• Use > [!IMPORTANT] for critical details not obvious from the paper
• Use > [!CAUTION] for personal interpretations that may contain errors
• Define all mathematical variables before using them, with explicit dimensions
• Use tables for terminologies, comparisons, and experimental results
• Content can be written in English or Japanese, matching the paper's language or user preference
• Math notation: Use $...$ for inline math. For block equations (even single-line), use a math fenced code block with \begin{align}...\end{align} — do NOT use $$...$$.

Don'ts

• Do NOT copy-paste large blocks of text from the paper. Always paraphrase and summarize in your own words.
• Do NOT create table of experimental results because they are already present in the paper. Only include key results with context.