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?