How to Research

It can be difficult to know where to start researching a topic, especially as an undergrad with little or no experience. I had to figure this out a few years, so let me save you some time :)

The Beginning

At the beginning of the research process, you want to go broad in your subject. For example, if you are researching Gen AI Security, you might look up “Generative AI Security review” (or overview), “Generative AI Security survey”, and similar terms. If you can find a review paper, great, read it. If not, start with the most relevant paper, read the abstract/conclusion, then jump to related work. Most papers have a related work section; it discusses papers related to that one in the field.

Use Google Scholar/ArXiv/Semantic Scholar

How to Read a Paper

If you need to understand a technique proposed by a paper (rather than just learn broadly about the field), do this:

I am using the Chain of thought paper as an example: https://arxiv.org/pdf/2201.11903.pdf

  • Read the abstract+conclusion

  • Read the description of the proposed method/analysis

  • Look at any images (usually exclude graphs)

  • Skip related work

  • Skip experimental setup

  • Skip ablations

  • Read the conclusion, and perhaps some discussion

You will usually only want to read the sections I said to skip if you really need to understand the technique (for example, if you are comparing it to other techniques).

How to keep track of information

Do whatever your project lead suggests. The way I keep track of info is by adding the citation to the Overleaf bibliography file (read the on Overleaf), then writing comments above it as my notes. I try to just started writing in the regular tex file as soon as I get a decent understanding of the field.