ChatGPT is "trained" to "make unpredictable changes in policy（change day and night）" like a human

For the same question,

Because the paths and factors generated each time are different;

And the way of thinking of simulating people is random;

And may add new knowledge, progress at any time;

There will be multiple sets of answers.

For the following questions, 

“How do you say No? 

To say No to people in a good way without letting them get hurt?”

the answers provided in the afternoon may not be exactly the same as those in the morning.

Morning answer:     

Afternoon answer: