I actually tried building this, and the RAG was such a cumbersome process. But now OpenAI just abstracted the RAG away. Damn.

No need to slice sentences, pick embedding model, worry about embedding DB. Upload then chat. Insane.

I ran this

find . -type f $ -name "*.md" -o -name "*.txt" $ -exec cat {} + > everything.md

on the farcasterxyz/protocol GitHub repo

then upload the everything.md

Every devtools company gonna have their own GPT for docs

exactly my first thought

docs site will go to this Assisstant frontend rather than a Vitepress site, matter of time

"Creative Technology" 🤌🏽

@ google creative lab

Owner of /makerspace - come build with us!

How many chars/tokens was your doc and how much room to spare in context window did u have? And what is approx cost per question? (This is very interesting - I had same problem as u mentioned above trying to roll my own)

context is ~60kb
RAG is done by OpenAI, and it costs $20/gb/day for the context stored
I have tried rolling my own LLM+RAG and it was a massive pain, this is way better