This has been my experience as well; however, when I want to get an encyclopedic summary of a topic, I’ve noticed that 3.5-turbo is more willing to reply directly to a handful of keywords, whereas GPT-4 typically tries to suss out a more specific question before dedicating itself to replying.