Research Guide
How to read papers efficiently, tools to use, prompts that work, and how to write. Distilled from practice.
Why Visualizations + Examples?
A 2014 study in the Postgraduate Medical Journal found that 87% of people connect with multiple learning styles β especially when auditory learning is combined with other methods. The VARK model identifies four primary styles: Visual, Auditory, Read/Write, and Kinesthetic.
The 70-20-10 rule also suggests that 70% of learning comes from practice, 20% from social interaction, and only 10% from traditional training. That's why PaperTrace uses interactive examples (70%), connection to related papers (20%), and structured explanations (10%) β not just passive reading.
Source: Fleming & Mills (1992) VARK model; Postgraduate Medical Journal (2014); Lombardo (1996) 70-20-10 model
How to Read a Paper
The 3-pass method (Keshav 2007): don't read linearly. Three focused passes, each with a different goal.
- Read title, abstract, intro, section headers, conclusion
- Look at all figures β they usually tell the whole story
- Answer: What is the problem? What is the solution? Are there experiments?
- Decide: worth reading further? (most papers: no)
- Read with a pencil β annotate everything
- Mark equations you don't understand β don't get stuck
- Note the key contributions (usually 3-5 things)
- Read related work to understand positioning
- Every claim, every equation β verify you can re-derive it
- Find every assumption and question it
- Think: how would I have done this differently?
- Identify future work directions
Prompts for Reading Papers with AI
I'm reading a paper. Here's the abstract: [PASTE ABSTRACT] Give me: 1. The problem being solved (1-2 sentences) 2. The core method/insight (1-3 bullet points) 3. Key results (1-3 metrics) 4. What I need to know first (prerequisites) 5. How it relates to [RELATED WORK]
Explain this formula from a paper step by step: [PASTE FORMULA] Context: [WHAT THE PAPER IS ABOUT] For each variable/operator: - What does it represent intuitively? - What values can it take? - Why is it there? Then give me a concrete numerical example.
I've read this paper: [TITLE] Core method: [BRIEF DESCRIPTION] Please analyze: 1. What assumptions does it make? Are they justified? 2. What are the weakest points of the experiments? 3. What baselines are missing? 4. What does it not handle that I should be aware of? 5. What follow-up experiments would be most revealing?