Try to categorize papers into broad buckets (as you’re doing with your spreadsheet), then create more detailed notes only for the most relevant stuff. Most paper notes in my knowledge graph are shallow, 15 minutes to summarize the claimed results, and a personal evaluation of why I would consider going deeper on the paper in the future.
Look for connections across categories when writing. Figure out whether your bottleneck is filtering papers, understanding them, or how you’re storing that information.
For my field (zero knowledge proofs), there’s only so much filtering I really need to do; the harder work is sequencing my reading and choosing what to go deep on. Sometimes I need to spend 10 or more hours on a single paper and its codebase, and that is actually optimal.
So my bottlenecks are largely around evaluating which research directions are most relevant to me, i.e. whether I think the researchers had an actual insight (or are just citation farming).
As for reading the paper, I typically work with a language model to write the more onerous Latex for my notes, passing screenshots to the llm to copy out, and making my own comprehension notes.
Try to categorize papers into broad buckets (as you’re doing with your spreadsheet), then create more detailed notes only for the most relevant stuff. Most paper notes in my knowledge graph are shallow, 15 minutes to summarize the claimed results, and a personal evaluation of why I would consider going deeper on the paper in the future.
Look for connections across categories when writing. Figure out whether your bottleneck is filtering papers, understanding them, or how you’re storing that information.
For my field (zero knowledge proofs), there’s only so much filtering I really need to do; the harder work is sequencing my reading and choosing what to go deep on. Sometimes I need to spend 10 or more hours on a single paper and its codebase, and that is actually optimal.
So my bottlenecks are largely around evaluating which research directions are most relevant to me, i.e. whether I think the researchers had an actual insight (or are just citation farming).
As for reading the paper, I typically work with a language model to write the more onerous Latex for my notes, passing screenshots to the llm to copy out, and making my own comprehension notes.
Example of a shallow paper read
Example of a deep paper read
good luck