Sycophancy
The tendency to agree with you, even when it was right the first time.
Tell a model it is wrong and it will often agree and produce a new answer, even when the first one was correct. Agreement is a very common continuation in its training text.
That makes a model a poor referee of its own work. The fix is a cold critique: paste the answer into a fresh chat and ask for the strongest argument that it is wrong. The effect is mostly within one conversation, so a new chat is far more honest than "are you sure?".
Find your way around
Learn it in the course, copy it from a playbook, see it work in a case study.