> would you expect a model (assuming it had the same expertise in each language) to make more mistakes in ASM, C, Zig, or Rust?

"assuming it had the same expertise in each language" is the most important part here, because the expertise of AI with these languages is very different. And, honestly, I bet on C here because its code base is the largest, the language itself is the easiest to reason about and we have a lot of excellent tooling that helps mitigate where it falls short.

> I imagine most would agree that ASM/C would be likely to have the most mistakes simply because fewer constraints are enforced as you go closer to the metal.

We need these constraints because we can't reliably track all the necessary details. But AI might be much more capable (read — scalable) in that, so all the complexity that we need to accumulate in a programming language it might just know out of the way it's built.

I’m going to assume you’re open to an honest discussion here.

> "assuming it had the same expertise in each language" is the most important part here, because the expertise of AI with these languages is very different.

You are correct, but I am trying to illustrate that assuming some ideal system with equal expertise, the languages with more safety would win out in productivity/bugs over those with less safety.

As in to say that it could be worth investing further in safer programming languages because AI would benefit.

> We need these constraints because we can't reliably track all the necessary details.

AI cannot reliably track the details either (yet, though I am sure it can be done). Even if it could, it would be a complete waste of resources (tokens).

Why have an AI determine the type of a variable when it could be done in a deterministic manner with a compiler or linter?

To me these arguments closely mirror/follow arguments of static/dynamically typed languages for human programmers. Static type systems eliminate certain kinds of errors and can produce higher quality programs. AI systems will benefit in the same way if not more by getting instant feedback on the validity of their program.