dejavucoder

fair enough

we live in exciting and scary times...

you may notice Kimi, GLM have also started telling how their model is able to optimise it's own inference pipeline

https://www.kimi.com/blog/kimi-k3

fair argument

this is true. in one of the later problems (cholesky decomposition), the organizer ran the submissions on a tiny training run to validate... and also provided code for same for our reference. most of the top solutions hit 4/8 or so. not very numerically stable.

i found out that as i learnt more domain wise, i was (obviously) able to steer better. doing a re-write can also remove lots of slop and context rot (and subsequently make it easier for both human and LLM to make solution more numerically stable, less reward hackish)

submission #2 by gau nernst is most numerically stable

1. labs have lots of inference capacity 2. they will have domain experts working on this so their efficiency is gonna be exponentially more (can direct LLM better, save money, reach same results faster)

hello author here.

yes, it gives labs edge and leads to self-recursive improvement loops.

also i was myself able to finish 7th in a later competition with 2-3 other approaches which are variants of the method discussed in this blog.

in general, having a harness as thin as possible with some problem specific instructions while controlling for context rot is the key.

point i am trying to make is there are a lot of optimisation surface areas possible.

author here!

welcome! check out my featured section

It's just a you can tell claude to make to write notes to

Can probably give access to tools like ast-grep to Claude. Will help it see all references. I still agree some dynamic references might still be left. Only way is to prompt well enough. Since I tested this on a Ruby on Rails codebase, I dealt with this.

lol yeah

thanks!

Fair point.

For context, I was using Claude Code on a Ruby + Typescript large open source codebase. 50M+ tokens. They had specs and e2e tests so yeah I did have feedback when I was done with a feature - I could run specs and Claude Code could form a loop. I would usually advise it to fix specs one by one. --fail-fast to find errors fast.

Prior to Claude Code, I have been using Cursor for an year or so.

Sonnet is particularly good at NextJS and Typescript stuff. I also ran this on a medium sized Python codebase and some ML related work too (ranging from langchain to Pytorch lol)

I don't do a lot of prompting, just enough to describe my problem clearly. I try my best to identify the relevant context or direct the model to find it fast.

I made new claude.md files.

I use Claude Code 50% of times with Cursor now due to the diff and tab. The extension is just a bit buggy sometimes otherwise I would use it much more. I hit some node related bugs today while searching stuff with it (forgot to report to Anthropic lol). Other bugs include a scroll stuttering.

Almost feels like a game as you level up!

Thanks, I will check this out

An attempt to build cursor's @codebase feature - RAG on codebases

Two part blog series + 1 blog on my personal blog to attempt speedup

https://sankalp.bearblog.dev/speeding-codeqa/

*Uses LLMs to do so

AutoLatex is a browser extension that simplifies LaTeX equation generation for researchers and students. It uses LLMs to convert natural language and images into markdown and LaTeX, with instant rendering so you can edit on the spot.

In short, you can do the following

text/image to equations in markdown+latex

live rendering of markdown+latex

drag and drop screenshot

just a simple chat mode

customizable models and default prompts

I love Linus' blogs

thanks, will check out on einops more.

thanks

just keep doing the bit. nothing matters other than the bit. you have been doing it for so long, keep doing the bit. it's funny, like no one gets it at this point but that's the bit. keep doing the bit. nothing matters more than the bit.

I draw connection between the above meme and Karma Yoga as proposed in the Holy Bhagavad Gita. Please give a read.

Blog post noting down my learnings and references used to make my side project - codeQA.

codeQA is a question-answering system for codebases that uses semantic code search with embeddings. It indexes codebases by chunking them into methods and classes using abstract syntax trees (AST), generating LLM-based comments for the chunks, and embedding them. The system then uses techniques like hypothetical document embeddings (HyDE), BM25 keyword search, and cross-encoder based reranking to improve the retrieval of relevant code snippets in response to natural language queries about the codebase.

I use tree-sitter (Part 1) for extracting the abstract syntax tree and building the codebase index. Elaborate discussion on cross-encoder vs bi-encoder

I use LanceDB for vectorDB, cohere rank v3 for re-ranking and various LLMs to generate LLM comments, summary and chat.

a claude haiku based tiny app to get recommendations based on any three favourite things . e.g you can mention 2 favourite books and 1 song and get movie recommendations based on those in the form of a mermaid diagram (or a markmap mindmap diagram)

A Udemy free guide to learning React in 2021/2022 using up-to-date resources with focus on React Hooks and functional components. This will save you from frustration and possibly wasting your time on outdated material.