Ask HN: Does ChatGPT 3.5 give you better results if you praise it?

It scores better on tests if you tell it that it's an expert in the field. If you treat it like a dog, it'll act like a dog. It returns the semantic space you point toward. In a way, it's the mirror test for mankind.

It has no understanding of praise or punishment, it is simply a statistical engine that correlates tokens (which are not even words but byte sequences that can be made into words by concatenation).

I made a similar observation. Assuming it was trained on internet content, like fora conversations, it would make sense that a polite conversation is conducive towards usable results.

It's not evolved yet to accept the appreciation to improve results, But It will do only if you say to do so.