suggestion: use a vector-db like pinecone to overcome the token limits? about aider HOT 3 CLOSED

FYI,I just released some improvements to how aider manages the context window

https://github.com/paul-gauthier/aider/blob/main/docs/ctags.md

from aider.

The good news is that I haven't really found the context window to be a problem with GPT-4 and the diff based approach used by aider.

But ya, I am familiar with that technique. I've used it with other LLM projects, and have been thinking about how it might work for aider. It's not obvious that it is the best approach, but I will certainly experiment with it more.

I have previously done a fair bit of work experimenting with various approaches to work around the smaller gpt-3.5 context window. I shared some notes on HN at the time:

https://news.ycombinator.com/item?id=35441666

from aider.

Wow! That's an impressive solution and very "compressed". It amazes me, that ctags is totally sufficient. Great idea + great, that it works.! Contratulations! I like your blog-post about it, too!

from aider.