From a8a04bfd0d494ab8b89c02c8e44bda11a237e9ba Mon Sep 17 00:00:00 2001 From: Paul Gauthier Date: Mon, 22 May 2023 15:53:51 -0700 Subject: [PATCH] copy --- docs/ctags.md | 16 +++++++++------- 1 file changed, 9 insertions(+), 7 deletions(-) diff --git a/docs/ctags.md b/docs/ctags.md index be1dc1a53..aec046535 100644 --- a/docs/ctags.md +++ b/docs/ctags.md @@ -1,13 +1,15 @@ # Improving GPT-4's codebase understanding with ctags -While GPT-4 excels at coding tasks, it struggles with modifying -code in larger code bases. -Many common -types of code changes require knowledge and context from multiple files -scattered throughout a large repo. -This is a challenge for GPT-4, which can only fit 8k-tokens -worth of code in its context window. +GPT-4 is extremely useful for "self-contained" coding tasks, +like generating new code or modifying a pure function without dependencies. +But it's difficult to use GPT-4 to modify code with dependencies +that are scattered across many files of a codebase. +To modify such code, GPT needs to see all of these dependencies. +And depending on the assigned coding task, it may not be obvious +how to even determine which parts of the repo are relevent a priori. +Further, all of this context needs to fit within GPT-4's 8k-token +context window. To address this issue, `aider` has a new experimental feature that utilizes `ctags` to provide