Anyone else think the hype around large language models is overstating their actual ability to understand human nuance?