Only those who have read many papers can design good research, conduct it efficiently, and package results well to produce good papers
Reading & Writing Papers
Academic Papers in the LLM Era
β
Ultimately, you read papers to write your own papers better
To convince your advisor (to become an independent researcher)
- Without reading papers and developing insight, you cannot conduct independent research
- To avoid being dependent, you need to know best
To avoid getting scooped
- What you think is "new" may not be new at all
- Read papers to quickly discover this and develop in different directions
To stay current
- Engineering research results have a half-life of about 5 years
- If you don't follow recent papers, you end up researching based on outdated results
Historically prestigious journals like JACS or Angewandte are more likely to publish good papers
Nature familyScience familyEESAdv.Mat familyChemJouleJACSAngewandteACS CatalACS Energy LettAppl Catal BnpjChem MatJ CatalJMCAJPCCPCCP
Norskov & ChorkendorffN2 reduction
Phillip ChristopherSAC characterization
Jinlong GongComputational PDH
Yong WangThermochemical analysis
Yadong Li, Tao ZhangSingle-atom catalysts
Nuria LopezPt/CeO2 catalysts
Jan RossmeislHigh entropy alloy
Ulissi & J. KitchenAI/ML catalysis
VlachosComputational design
Karsten ReuterElectrochemical modeling
Emuel HensenComp/Exp thermochemical
W.A. GoddardComputational chemistry
J.G. ChenSurface science
T.F. JaramilloElectrocatalysis
MarkovicElectrocatalysis
Groups Strong in Both Computation and Experiment
Level 1
Skimming
Abstract + Figures only
5-10 minLevel 2
Selective
Intro + Results + Conclusion
30 minLevel 3
Deep Reading
Including Method + SI
2-3 hrs01
Read critically: Don't accept claims blindly, find logical gaps
02
Read actively: Make questions while reading, take notes
03
Focus on figures: Figures are the core
04
Check methods carefully: Verify reproducibility
05
Follow references: Understand important prior work
06
Set time limits: Don't spend too much on one paper
If you only see the world through the LLM lens, it can become a cognitive ceiling. Deep reading is still necessary.
NotebookLM
- Convert to podcast
- Auto-generate PPT
- Easy access to unfamiliar fields
Advanced Usage
Query all papers from one authorUnderstand research direction
Compare multiple papersFind research gaps
Generate visualizationsPrepare presentations
GPT / Claude / Gemini
- Upload PDF and Q&A
- Summary requests
- Critical analysis requests
βNovel discovery or methodology
βClear key contribution
βReproducible methodology
βIn-depth discussion
βBeautiful figures
βLogical storyline
- This is the stage where you define the key contribution you want to deliver
- Generally target journals with IF 15 or higher
- Some values transcend IF. e.g., JACS is similar in importance to Nature Catalysis
- Usually the ceiling of research is determined from the start
When research matures, your advisor will signal to start preparing the paper
The most important thing is fast draft completion and fast iteration
Figure First!
Figure Prep
- Study how others compose figures
- Create expected scenarios before results
- Prepare about 5 figures
- Unclear figures are likely failures
Synopsis
- Build framework around figure set
- 3 bullet points per figure
- Move unnecessary to supplementary
Iterate A Lot!
"When someone points out something, there is usually a problem there. Even if you think 'this is perfect,' just fix it anyway."
- Haruki Murakami
Citation Principles
- Always cite the original source - avoid secondary citations
- Cite only the most important papers - reduce unnecessary citations
- Avoid excessive citations just to increase count
1. Trend statements
"Recent advances in MLIPs have enabled large-scale MD [3-5]."2. Methodology
"DFT calculations were performed using VASP [7]."3. Prior work comparison
"Similar relations have been reported [10]."4. Differentiation
"Unlike conventional approaches [15], our method..."Citations Not Needed
- Generally accepted scientific facts
- Sentences describing your own results
- Logical reasoning from this study
Word Pain Points
- Equation editing - click click, equations break
- EndNote update - plugin stops working
- Citation style change - 50% chance of breaking
- Move Figure - manually update all numbers
- Format change - 200 pages of adjustment
LaTeX Code Examples
\int e^{-x^2} dx = \sqrt{\pi}
\cite{Park2024}
\bibliographystyle{nature}
\ref{fig:name}
\ce{CO2 + H2O <=> H2CO3}
LaTeX separates content from formatting. You focus on science, the system handles format.
