Most of the people saying the kind of things you say don't know how to properly use an AI code assistant.
That's a very lackluster argument. Even if everything you said was true (and to be clear: it absolutely isn't; LLMs will hallucinate no matter what as it's in their very nature), that's still a flaw in the tool. A good tool must not require convoluted dances to make it work halfway decently.
Sorry but this is demonstrably false. For starters, see the "R's in strawberry" meme.
Ask it to create a program that counts the R's in any text. Now ask it to do it in 20 different languages. It'll get them right immediately.
It's like me asking you who was the proposed King for Finland and calling you a bad programmer because you don't know.
Why are you allowing massive PRs at all, fix your process first.