How does this surprise anyone?

LLMs are just pattern recognition machines. You give them a sequence of words and they tell you what is the most statistically likely word to follow based solely on probability, no logic or reasoning.