They're still at the compiler stage. LLM features and hardware seem far enough away that it's reasonable to wait to evaluate if that combination is actually practical.