A Codex skill that turns AI papers into bilingual, interactive courses.
Point it at a paper. Get back a polished English/Chinese HTML course that teaches what the paper claims, how the method works, what the experiments actually prove, and what a skeptical builder should test next.
AI papers are dense in a very particular way. The hard part is rarely just "what does this paragraph say?" The hard part is building the right mental model:
- What problem is this paper really solving?
- What changed compared with RAG, fine-tuning, agents, long context, or previous systems?
- Which equation, table, or ablation is actually load-bearing?
- What does the paper prove, and what does it merely suggest?
- If I asked an AI agent to implement this, what should I tell it to build first?
Paper2Course turns that reading process into a course. It starts from the paper's motivation, walks through the mechanism visually, translates formulas and tables into plain language, tests understanding with applied quizzes, and ends with a high-level synthesis plus FAQ.
The generated course is a static website. No backend. No build server. Open it directly or host it anywhere static pages work.
It includes:
- Bilingual pages —
index.htmlfor English andindex.zh.htmlfor Chinese, with a nav switch - Scroll-based modules with progress dots and keyboard navigation
- Formula / table / algorithm -> plain language translations
- Applied quizzes that test judgment rather than memorized terms
- Method and data-flow animations
- Research-meeting style chats between model components, datasets, losses, baselines, or evaluators
- Glossary tooltips for technical terms
- Final synthesis and FAQ to resolve the big conceptual confusions
- A warm closing note that reinforces confident, skeptical paper reading
Paper2Course is for practical AI builders, product-minded researchers, and vibe coders who want to read AI papers well enough to use them.
It is especially useful if you want to:
- understand a paper without getting trapped in notation
- compare a method against RAG, fine-tuning, long-context models, agents, or systems tricks
- identify whether an experiment really supports the paper's claim
- prompt AI coding/research agents with sharper technical instructions
- decide what would be expensive, fragile, or worth testing first
Use $paper2course to turn this arXiv paper into an interactive course:
https://arxiv.org/pdf/2603.23516
把这篇 AI paper 做成中英双语互动课程：./paper.pdf
Use $paper2course on this paper plus official repo and explain what I should trust, test, and be skeptical about.
Clone this repository into your Codex skills directory:
git clone https://github.com/clearchen666/paper2course ~/.codex/skills/paper2courseRestart Codex so the skill is discovered.
You can also ask Codex to install it for you:
Install the skill from https://github.com/clearchen666/paper2course into ~/.codex/skills/paper2course
After installation, try:
Use $paper2course to turn this AI paper into an interactive course: <paper URL or PDF path>
Each course is generated as a directory:
paper-course-name/
styles.css
main.js
_base.html
_base.zh.html
_footer.html
build.sh
notes/
analysis.md
modules/
01-problem.html
...
modules-zh/
01-problem.html
...
index.html
index.zh.html
Open index.html for English or index.zh.html for Chinese. The two pages share the same interaction engine and course structure.
A useful paper walkthrough should tell you what to believe, what to doubt, and what to try. Paper2Course treats every paper as a set of claims, mechanisms, and evidence, not as a pile of paragraphs to summarize.
If the paper's argument depends on an equation, algorithm, figure, table, or ablation, the course translates that artifact into plain language. The goal is not to remove technical detail; it is to make the technical detail usable.
The final module includes a synthesis and FAQ so the learner leaves with a stable mental model. For example: what is trained once, what changes per user or corpus, where memory is stored, how the method differs from RAG, and what should be tested first.
Paper2Course is adapted from Zara Zhang's excellent codebase-to-course.
The original project established the interactive course format, warm developer-notebook visual direction, scroll-based modules, quiz patterns, group chat/data-flow animations, glossary tooltips, and code-to-English translation blocks. Paper2Course extends that foundation from codebases to AI academic papers, adding paper-specific analysis, bilingual English/Chinese output, final synthesis/FAQ sections, and research-reading pedagogy.
All credit for the original concept and course interaction foundation goes to Zara Zhang and codebase-to-course.
This repository intentionally does not add a separate open-source license at this time. The upstream codebase-to-course repository did not declare a license when this project was created, so this repo preserves explicit attribution and avoids making broader license claims.