I had the exact opposite experience with chatGPT (not Bing) this week. I made chatGPT play stockfish a few times. ChatGPT was able to make sensical moves for the first dozen or so turns, but then it would start to go off the rails, like saying that it was moving a pawn to prepare to castle, which it has just done on the last turn. It fianlly started trying to move invisible pieces or make illegal moves. It also played highly defensively, preferring to build a closed fortress in the opening instead of attacking.