← Back to context

Comment by IshKebab

17 hours ago

> Is the code bad? It sounds like they had no idea he was using AI. That seems to imply there was nothing wrong with the code as-is. Why not judge it on it's merits?

Anyone that has used AI at all knows this isn't how it works. AI is extremely good at producing plausible-but-wrong outputs. It's literally optimised for plausibility, which happens to coincide with correctness a lot of the time. When it doesn't you get code that seems good and is therefore very difficult to judge on its merits.

With human written code it's a lot easier to tell if it's good or not.

There are exceptions to this - usually if you have some kind of oracle like that security work that used AddressSanitizer to verify security bugs, or if you're cloning a project you can easily compare the behaviour to the original project. Most of the time you don't have that luxury though.

It's also easy to overwhelm reviewers with far more code than they can possibly review. And it's also the hardest stuff to review where the code at surface level looks totally fine, but takes long hours of actual testing to make sure it works.

Just read the code. There is nothing keeping people from reading the code.

  • I can't find the source code either for the official MeshCore app or for MeshOS -- where can I read the code?

    • It's so weird right? I keep hearing people say it's open source but like... where's the code then? I've tried to find it. I can find stuff for core components, but they lock features behind a delay-wall in the app. If it was open source that stuff would be gone immediately.

      1 reply →

Do folks not write tests and review their own code (AI generated or not)?

Also, citation needed:

> With human written code it's a lot easier to tell if it's good or not