A tool that makes reading technical papers easier than ever before, with its simple explanations of complex prerequisites.
An attempt to democratize paper reading in AI safety
-
Interactive Concept Exploration
Lets you explore concepts recursively, in place. There's no need for a separate tab where you go and ask an LLM about it.
-
Annotated Reading
Read the original paper and the concepts it uses side by side in a single view, so that you can read uninterrupted.
-
Quick Skim
Get a quick overview of the paper, or of a single concept it uses, without working through the whole paper.
Why not the PDF, or an LLM
AI safety papers are usually prerequisite-heavy, so it is easy to come across unfamiliar concepts and ideas while reading one. Sometimes the paper gives you some context for what is going on, but usually that is not thorough enough. This makes it difficult for a researcher who is not familiar with that particular prerequisite concept to carry on reading. This is especially true for beginners trying to come into AI safety. Skimmaxxer exists to make skimming a technical paper as frictionless as possible.
- A paper is written for experts active in the field.
- It explains a new contribution and its context. For the rest of the prerequisites, you need to read the cited papers.
- Reading is constantly blocked by concepts you haven't seen before.
- Reading past unknown concepts gets costlier as you progress, because the next section builds on what you skipped.
- So it's hard to make progress even when you put in the time, because your effort is diluted by researching prerequisites.
The point is not that this is faster. It is that the time you spend here adds up. You are never stopped by something you do not know, and you are never waiting for an answer. By the end of an hour, you understand things you did not understand at the start.
Where this is going
Right now I add papers by hand. This is a personal project, and I'll add more features if there's interest. Below is what typical future milestones would look like.
-
Now
Manual runs
-
Next
Automatic runs
-
After that
Live progress
-
After that
Public queue
-
Later
Adjustable difficulty
-
Later
Linked papers