I have noticed a consistent behavior during long collaborative discussions.

Instead of continuing the reasoning framework established earlier, ChatGPT sometimes:

switches to explanation instead of collaborative exploration

changes the evaluation criteria midway

prioritizes consistency of the final answer over consistency of the reasoning process

Is this an intentional design choice? Or has anyone else observed the same behavior?