This is the most subtle part of how AI works. As we understand - there is no AI, i.e. intelligence, it's just a language model. For example, with ChatGPT , it was not uncommon to have to fracture and elaborate on a question in order to get an acceptable answer. With claude, the situation is much better. In my opinion he does not suffer from “sclerosis” and “phantom fantasies” 
Well, not yet in the human sense but I'll summarize what Sam Altman told Joe Rogan in an interview about what he thought about AI 10 years ago. He categorized what he believed they could do into:
1) Repetitive tasks, such as putting an AI on a robot and having it go about tightening screws or carrying things along a defined path.
2) Blue collar tasks, he believed they could be helpful but wouldn't amount to much.
3) Creativity. About this I believed that it would be impossible for these models to have any skill.
Well, it seems that time is going to prove him wrong even on point 3.
I just think AI nowadays is like the internet on 1995. We still have a lot to see.