They always sound like an obnoxious American tourist talking through a translator, the chatbot training dataset is the same and foundation models are always built with >50% American English data for some reason.