• Table of Contents

<aside> ✍️

Aside from @s.hadfield ‘s wiki page on Everything Induction Related, I’m writing this document because I’ve realized that I performed rather poorly when it comes to literature review (see Retro From My First Three Months ) and this process really has to change. Synthesizing new ideas can effectively be boiled down to three phases: Collection, Distillation, and Connection. The main thing I’ve realized is that trying to build a perfect filing cabinet before you have any information to put in it is a classic schema-on-write trap given it’s also quite counter-intuitive to introduce structure when you only have a small sample of data. In essence:

Structure should emerge from your ideas and thoughts, not be imposed before you've explored them.

So, I’ve spent most of this break refactoring the way I handle idea collection and idea distillation.

</aside>

The Simple Zettelkasten

“Zettelkasten” gets thrown around a lot and there are thousands of opinionated setups depending on the choice of tools. I've done some reading and research, but to avoid being distracted by shiny new tools and plugins, I'm designing a minimalistic workflow with my current toolset. The point of Zettelkasten is not to collect as many notes as possible, but to augment/add values to existing ideas, arguments, and discussions. We want to add value from what we’ve just learned to what we already know.

More info → More potential connections to new info
More connected info → More cues to trigger the right memoryThree Types of Notes

The New Process

First, there are three types of notes I’ll be taking:

• Literature Notes: Highlights and annotations captured while reading.
• Permanent Notes: Distilled notes from my reading, stored in .
• Fleeting Notes: These are random shower thoughts that popped in my mind and will be stored in with the tagged fleeting.

The Collect → Distill→ Connect process works as follows:

1. Collect: Capture ”Literature Notes” as I read. I highlight whatever I think is interesting or relevant, or anything that sparks some thought. A quick highlighting scheme:

   • main research question(s)
   • previous pain points or just pain points in general
   • what they do well/ what they’re trying to sell
   • technical implementation/ objective formulation
   • mathematical proofs / equations I need to understand
   • experimental results
   • what they fall short on/ gap (usually specified at the end)
   • datasets

2. Distill: Transform literature notes into a 3-5 sentence “Permanent Notes” in my own words without excessive use of AI. These 3 sentences should be comprehensive enough to answer the following:

   1. Identify the problem: What current gap in existing methods motivates this research?
   2. Identify the solution: What mechanisms/methods did they use? What are my thoughts on it? What do I like about it and what criticisms do I have?
   3. Identify the weakness: What limitations and open problems are left on the table that we can attack and poke holes at?

3. Connect: Once a permanent note is completed, add tags and categories in Zotero. Do not just throw a label on things but rather ask “How will I retrieve this information?”

   1. In which circumstances do I want to stumble upon this note?
   2. When and how will I use this idea?

   Finally, update these connections in .

On Tool Usage…

I realized this quite late, but it reinforces the first main point: structure will emerge after we digest more information. Zotero is not built to de-duplicate or keep items “normalized”. instead, it should be treated like a library catalogue rather than personal folders. Treating it like personal folders makes categorization difficult because some literature spans multiple collections. It is better to accept that items will belong to multiple categories and view them as soft-linked across overlapping collections.

Given that the Unfiled Items in Zotero already serves as a data dump of everything, each collection can and will contain duplicated items if we’ve thoroughly reviewed and filed them. The most effective way then is to search for specific articles is by looking up tags and create collections once we’ve distilled the information. To summarize:

• Tags are for topics and categories.
• Collections are for context or projects.

Alternatively, there should be an Inbox containing all papers that need to be read. Ideally, this inbox should be limited to the set number of papers I have the capacity to read this week. Once we go through the Zettelkasten process by the end of the week, they should be moved out of the inbox and filed appropriately.