karpathy

LLM Wiki

A pattern for building personal knowledge bases using LLMs.

This is an idea file, it is designed to be copy pasted to your own LLM Agent (e.g. OpenAI Codex, Claude Code, OpenCode / Pi, or etc.). Its goal is to communicate the high level idea, but your agent will build out the specifics in collaboration with you.

The core idea

Most people's experience with LLMs and documents looks like RAG: you upload a collection of files, the LLM retrieves relevant chunks at query time, and generates an answer. This works, but the LLM is rediscovering knowledge from scratch on every question. There's no accumulation. Ask a subtle question that requires synthesizing five documents, and the LLM has to find and piece together the relevant fragments every time. Nothing is built up. NotebookLM, ChatGPT file uploads, and most RAG systems work this way.

thatisuday

thread	goroutine
OS threads are managed by kernal and has hardware dependencies.	goroutines are managed by go runtime and has no hardware dependencies.
OS threads generally have fixed stack size of 1-2MB	goroutines typically have 8KB (2KB since Go 1.4) of stack size in newer versions of go
Stack size is determined during compile time and can not grow	Stack size of go is managed in run-time and can grow up to 1GB which is possible by allocating and freeing heap storage
There is no easy communication medium between threads. There is huge latency between inter-thread communication.	goroutine use channels to communicate with other goroutines with low latency (read more).
Threads have identity. There is TID which identifies each thread in a process.	goroutine do not have any identity. go implemented this because go does not have TLS([Thread Local Storage](https://msdn.microsoft.com/en-us/library/win

sloanlance

jq: JSONL ↔︎ JSON conversion

Prerequisites

• jq — https://jqlang.org/ — "like sed for JSON data"

  There are several options available for installing jq. I prefer to use Homebrew: brew install jq

• JSONL → JSON