2 ms·
This looks really interesting. Have you spent much time investigating how to provide code context to the LLM without needing to manually highlight all the relev
by anotherpaulg 3y ago
This looks really interesting. Have you spent much time investigating how to provide code context to the LLM without needing to manually highlight all the relevant code?
That’s been a major focus of my open source AI coding assistant project [0]. It seems super promising to exploit the semantic structure of code to help LLMs understand complex codebases [1].
Also, there’s a small discord [2] where a few of us are sharing learnings about building these coding assistants. Please join if you’d like to compare notes.
[0] https://github.com/paul-gauthier/aider https://github.com/paul-gauthier/aider
[1] https://aider.chat/docs/ctags.html https://aider.chat/docs/ctags.html
[2] https://discord.gg/FTYDTRKZ https://discord.gg/FTYDTRKZ
- sestinj 3y agoWe've tried embedding search multiple times, and are also looking into the LSP. We actually removed the former for the time being because it reduces control/transparency when files are automatically being selected. Soon however we plan to release a Continue add-on that lets you type '@magic' or '@search' or '@embed' or something to deliberately include the results of a vector search in your context. Looks like cool work you're doing, would love to learn more—thanks for the invite!