Earlier quoted context omitted.
That's a pretty absurd claim, their native implementation is not only experimental, the GC solution is "unknown". You can't make this stuff up, it will be years before anyone is writing production IOS apps in Kotlin. Heck, their JS implementation still generates massive binaries. Targeting the JVM/Android? Sure, good to go; otherwise the promised land is a long way off.
"Unknown"? They have said from the beginning that it uses reference counting with cycle collection, for example in RELEASE_NOTES.md: "Currently Kotlin/Native uses reference counting based memory management scheme with a cycle collection algorithm. Multiple threads could be used, but no objects shared between threads are allowed." They have also said that it's pluggable and they might experiment with other collectors…
I don't doubt that 2-3 years from now Kotlin will have a cross platform solution, but to claim so today is borderline delusional. It's understandable that users are excited, Kotlin's on the verge of bigger things, perhaps even mainstream level adoption if everything falls into place, but there's a ways to go before they achieve one-language-to-rule-them-all.