With AI you have to teach one model once. It makes a mistake? Teach it and it will not make it again.
This is a lot cheaper than training every pilot on the planet and its a lot more persistent then training every new pilot
Has this been your experience? Models still invent answers for me. Directives go unrespected, etc.
I expect it to get significantly better in the next few years but I also expect either to see continues improvement or new optimized models for specific tasks.