Skip to content

Comment on GPT-4-turbo preliminary benchmark results on code-editingparent

Comments

The increased context size will have the most significant impact on my work. That's where I run into limitations, when reviewing written work and code. I've been feeding written work into GPT in chunks, and I'm really happy to be able to feed in whole pieces. (I don't have it revise anything for me, I just have a specific prompt for exactly the kind of feedback I want on written work.)

I tried Claude because of the larger context size, but I've been disappointed so far. I find Claude much more likely to just compliment my writing, whereas GPT will identify strengths and areas that could be improved.

Have you tried aider-chat? It does some interesting things with tree-sitter so it can give the LLM a context (files, classes, functions, parameters, etc) as well as certain full files. That way it has your entire codebase in API form and it can focus on the actual code you are looking at or editing.

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.