Have Devin do the code reading.
I did it because I wanted to look into Polis question presentation order mechanism.
I also did Polis clustering implementation and [Confidence Scoring Using Matrix Decomposition in Community Notes
prompt notes.
Polis question presentation order mechanism
prompt: I would like to read this source code and create an explanation of how the order in which the questions are presented to the user is determined. Maybe it's not random, maybe it's the order that maximizes the amount of information or something. I'd gather the relevant code and put it together in Markdown. As for what you don't know, let them know you don't know so they can ask a more knowledgeable AI about it.
Polis clustering implementation
prompt: please create an additional new Markdown report.
My question is about clustering. If I understand correctly, after k-means with 100 cases, you do k-means with 2~5 cases and choose one by silhouette coefficient.
Q1: How does it actually work? Explain, citing code.
Q2: What are you doing to reduce computational costs when data is added?
Q3: Is it possible to replace the latter half of k-means with hierarchical clustering?
Confidence Scoring Using Matrix Decomposition in Community Notes
prompt: read this source code and make a Markdown explaining how the note confidence scores are given. I know that you use matrix factorization. What exactly is the matrix from what data and how is it decomposed?
I let o1 Pro read the completed report and give feedback on my impressions.
---
This page is auto-translated from /nishio/Devinにコードリーディングさせる using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I'm very happy to spread my thought to non-Japanese readers.