Recent advances in AI aren't useful for routine operations in safety critical domains such as aviation because we don't know how to verify and test them. An LLM is effectively an unpredictable black box with unknown failure modes. There is opportunity for greater automation but probably based on classical deterministic programming.

In addition to this, LLMs are also simply too slow right now to deliver the results ATC would need.

Ridiculous to see people acting like LLMs are a silver bullet for every problem without putting any thought into what that would actually look like.