Are there any good human experiences in reviewing AI-generated code?

How is the current level of acceptance of AI-generated code within the Julia community, as well as the entire programming community and academic community? Also, for the code generated by AI, is there any good experience in code inspection or any ready-made workflow to verify its correctness?

If I recall something like up to 80% of recent PR to JuliaLang itself have AI assistance (many from CTO Keno). So I would say that much acceptance. By those that merge PRs. There are many vibecoded Julia packages, and people are divided regarding those, and I’ve seen opposition to AI use for JuliaLang itself. There will never be full acceptance, some people want to stop using Julia (or could fork), others welcome more AI use, and accelerated development speed and e.g. Anthropics’s Mythos finding 16 confirmed security issues (now fixed) in Julia.

Thanks. Now I think one issue should be significant that whether we have a standard official workflow that can let Human check for correctness and quality of a julia package. What kind of standardized manual review process should the code undergo before it can be accepted for registration?

Do you have some experience on this issue?

There was quite a long discussion on a similar topic earlier this year.

There are really three wildly different questions here.

@WuSiren what do you want this topic to focus on?