[zozbot234]

Apparently the write code part works a bit better in languages like Rust and perhaps Swift where the compiler is unforgiving in rejecting outright nonsense and the AI can iterate on any errors it gets. Of course logically flawed code is always possible so this does not replace human review. But code in these languages is also a bit more compact and hopefully easier to understand for a human.

[harambae]

On the flip side, how much training data exists for coding in Swift? Especially when a new version seems to come out every 6 months.