[carimura]

I continually run codebases through different models to have them look for bad code smells like repeated code. That's been pretty effective. You do have to maintain over time or else you end up with a sloppy mess which I can only imagine compounds.

[sejje]

Do you think it matters that it's a different model?

Or is it more about the review process and a context reset?

[carimura]

I don't have a scientific answer to this but I'm just assuming a different model, and more specifically, the frontier latest models like Fable/5.6 would catch things the author model might skip over.

[bluGill]

I have personally not found value in a different model. The review itself often is good.

[t-writescode]

What do you think of traditional AST-based static analyzers (like SonarQube, Rubocop, etc.) that catch cyclomatic complexity and code duplication deterministically?

You can even put them on the build pipeline, even. No tokens spent at all!

[carimura]

I haven't used them but I'm sure they're good and I'm sure they're even using LLMs themselves now as a non-deterministic pass. For my use cases I'm just trying to keep vendors/costs to a minimum.