PaperMap

PaperMap is an open-source, static-first library of interactive AI paper explainers.

The goal is simple: turn dense research papers into visually rich, beginner-friendly, and research-accurate learning experiences.

Current status: 37 live interactive paper explainers, organized into a 7-category curriculum — from the original Transformer to multi-agent societies — plus 7 supplementary guides.

• Homepage

• Attention Is All You Need · 2017
• BERT: Pre-training of Deep Bidirectional Transformers · 2018
• Language Models are Unsupervised Multitask Learners (GPT-2) · 2019
• Language Models are Few-Shot Learners (GPT-3) · 2020
• Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer (T5) · 2020
• Scaling Laws for Neural Language Models · 2020
• Training Compute-Optimal Large Language Models (Chinchilla) · 2022

Study tip: read Scaling Laws and Chinchilla together — the second paper corrects the first.

• Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks (RAG) · 2020
• LoRA: Low-Rank Adaptation of Large Language Models · 2021
• Chain-of-Thought Prompting Elicits Reasoning in LLMs · 2022
• Self-Consistency Improves Chain of Thought Reasoning · 2022
• ReAct: Synergizing Reasoning and Acting in Language Models · 2022

• TruthfulQA: Measuring How Models Mimic Human Falsehoods · 2021
• SelfCheckGPT: Zero-Resource Black-Box Hallucination Detection · 2023
• HaluEval: A Large-Scale Hallucination Evaluation Benchmark · 2023
• A Survey on Hallucination in Large Language Models · 2023
• FActScore: Fine-Grained Atomic Evaluation of Factual Precision · 2023
• RAGTruth: A Hallucination Corpus for Trustworthy Retrieval-Augmented Language Models · 2024

• HELM: Holistic Evaluation of Language Models · 2022
• Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena · 2023
• JudgeBench: A Benchmark for Evaluating LLM-based Judges · 2024
• AgentBench: Evaluating LLMs as Agents · 2024

• AgentDojo: A Dynamic Environment to Evaluate Prompt Injection Attacks and Defenses · 2024
• InjecAgent: Benchmarking Indirect Prompt Injections · 2024
• Agent-SafetyBench: Evaluating the Safety of LLM Agents · 2024
• Agent Security Bench: Formalizing and Benchmarking Attacks and Defenses · 2024

• Generative Agents: Interactive Simulacra of Human Behavior · 2023
• MemGPT: Towards LLMs as Operating Systems · 2023
• A Survey on the Memory Mechanism of LLM-based Agents · 2024
• LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory · 2024
• A-MEM: Agentic Memory for LLM Agents · 2025
• LoCoMo-Plus: Evaluating Long-Term Memory Beyond Factual Recall · 2026

• CAMEL: Communicative Agents for "Mind" Exploration · 2023
• MetaGPT: Meta Programming for Multi-Agent Collaborative Framework · 2023
• Magentic-One: A Generalist Multi-Agent System · 2024
• Multi-Agent Collaboration Mechanisms: A Survey of LLMs · 2025
• MultiAgentBench: Evaluating Collaboration and Competition of LLM Agents · 2025

• ColBERTv2: Effective and Efficient Retrieval · 2022
• Training Language Models to Follow Instructions with Human Feedback (InstructGPT) · 2022
• Direct Preference Optimization (DPO) · 2023
• In-context Learning and Induction Heads · 2022
• Toolformer: Language Models Can Teach Themselves to Use Tools · 2023
• LLaMA: Open and Efficient Foundation Language Models · 2023
• DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via RL · 2025

• Interactive explanations instead of static summaries — 100+ playable demos and animated visualizations
• Expert-level Deep Dive sections on the core papers, not just paper summaries
• Research-accurate content grounded in original papers
• Shared visual language across all paper pages
• A structured curriculum — not a random pile of papers
• Pure HTML, CSS, and JavaScript with no framework overhead
• Fast load times, responsive on desktop and mobile

PaperMap/
|- index.html                  (homepage: 7-category curriculum library, 37 papers + supplementary)
|- 404.html
|- PROMPT.md                   (paper section guide + contribution system)
|- assets/
|  |- favicon.svg
|- paper/                      (44 paper pages + 2 legacy redirects)
|  |- ... one .html per paper, underscore naming ...
|  |- Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks.html   (legacy redirect)
|  |- LoRA Low-Rank Adaptation of Large Language Models.html                 (legacy redirect)
|- robots.txt
|- sitemap.xml
|- LICENSE
`- README.md

Because this project is pure static HTML, CSS, and JavaScript, you can run it with any static file server:

# Using Python
python -m http.server 8080

# Using Node.js
npx serve .

Then open http://localhost:8080 in your browser.

The Paper Section is a curriculum of 37 papers organized into 7 categories (LLM foundations → retrieval, reasoning & adaptation → hallucination → evaluation → agent safety → agent memory → multi-agent systems), plus 7 supplementary explainers kept beyond the core. Every explainer follows one shared page template, one design system, and one contribution workflow; all 37 core papers carry expert-level Deep Dive sections with animated demos.

Before adding or updating a paper, read PROMPT.md. It is the complete, contributor-friendly guide to the Paper Section: how papers are organized, the required page structure, naming and linking conventions, design/style rules, interactive demo requirements, verification steps, and the pull request checklist.

1. Pick an unmapped landmark paper that fills a logical gap in the curriculum.
2. Create a new HTML file inside paper/ using underscore naming (for example BERT.html).
3. Use an existing paper page (e.g., paper/BERT.html) as your style and structure template — copy the <style> block verbatim.
4. Keep the content research-accurate, include 2–3 interactive demos and a 5-question quiz.
5. Add your paper card to the correct category in index.html, update the footer links and paper count.
6. Add the URL to sitemap.xml and cross-link 2–3 related guides.
7. Verify responsive behavior on desktop and mobile, and run the checks in PROMPT.md §10.
8. Include complete metadata in the head section.

Contributions are warmly welcomed! We are actively building out the curriculum to cover the most impactful papers in AI history.

If you want an easy, guided process, use the workflow below.

1. Open Claude or ChatGPT.
2. Copy the entire contents of PROMPT.md.
3. Paste it into the chat and specify which paper you want to implement.
4. The AI will output a complete, standalone, production-ready HTML file adhering to PaperMap's design system.
5. Download the generated HTML file.
6. Fork this GitHub repository.
7. Upload your HTML file into the paper/ folder.
8. Update index.html so your paper appears on the homepage in the correct category.
9. Verify the page with PROMPT.md's checklist (§10) — fix any console errors or broken links.
10. Open a Pull Request.

Congratulations, you are now part of the PaperMap community.

1. Paper file added inside paper/ with underscore naming.
2. Homepage card added in the correct category, footer links and paper count updated.
3. Links tested locally (no 404s, no %20 URLs).
4. Desktop and mobile layout checked.
5. Metadata updated (title, description, canonical, social tags).
6. 2–3 interactive demos working with no console errors.
7. 5-question quiz included with explanations.

MIT License — see LICENSE for details.