AI Paper Writing

A modular skill bundle distilled from “Highly Opinionated Advice on How to Write AI Papers” (Neel Nanda).
The goal is to help you produce papers where readers understand, remember, and believe the narrative.

North star

An ideal paper is a short, rigorous, evidence-based technical story with a takeaway the reader cares about:

• What? 1–3 concrete claims that contribute to knowledge
• Why? rigorous evidence that supports the claims
• So what? motivation + impact: why the reader should care

How to use

Pick a workflow below, or jump straight to a module. Each module is independent.

Common inputs

• Your paper text (or section)
• Or: a project summary + key results + experiments list

Common outputs

• claim list (1–3 claims + scope + confidence level)
• evidence map (claim → experiments → risks)
• rewritten abstract/intro
• experiment fixes (baselines, ablations, stronger discriminators)
• figure + caption guidance
• limitations + discussion guidance

---

Workflows

A) Drafting from scratch (recommended order)

1. Compress to claims
2. Motivation & impact
3. Novelty & positioning
4. Evidence & red-teaming
5. Paper structure overview
6. Abstract
7. Introduction
8. Figures
9. Main body layering
10. Discussion & limitations
11. Related work

B) Reviewing a draft

1. Compress to claims (extract what the paper actually claims)
2. Evidence & red-teaming
3. Avoid misleading evidence
4. Abstract
5. Introduction
6. Common pitfalls

C) Strengthening experiments

1. Experiments design
2. Baselines + ablations
3. Avoid misleading evidence
4. Evidence & red-teaming

D) Tightening narrative + clarity

1. Compress to claims
2. Motivation & impact
3. Abstract
4. Introduction
5. Definitions & layering

---

Templates

• Abstract template
• Intro template
• Paper outline template
• Evidence audit template
• Figure caption template

---

Modules index

• 00 Overview
• 01 Compress to claims
• 02 Motivation & impact
• 03 Novelty & positioning
• 04 Evidence & red-teaming
• 05 Experiments design
• 06 Avoid misleading evidence
• 07 Structure: Abstract
• 08 Structure: Introduction
• 09 Figures
• 10 Main body
• 11 Related work
• 12 Discussion & limitations
• 13 Writing process
• 14 Common pitfalls
• 15 Checklists

---

Acknowledgements

This was compiled from Neel Nanda's article (Highly Opinionated Advice on How to Write AI Papers)[https://www.alignmentforum.org/s/5GT3yoYM9gRmMEKqL/p/eJGptPbbFPZGLpjsp]