Not gibberish but definitely I had quite a few back and forth where he would apologize for getting stuff incorrect, try to correct itself and sometimes be wrong again (in that case you can get into a loop where he alternates between two wrong answers). It was not gibberish just code that did not work (either at all or as intended) for various reasons. So I would call it buggy, not gibberish.

Worth noting it was ChatGPT 3.5 so maybe this is less of an issue now.