LLMs are great for research, data extraction, language learning, designing a study plan etc. They can answer a complex question quickly, so you don't have to waste time going down a road that leads nowhere. The answer may not be accurate, but the accuracy can often be verified easily.

I find it fascinating that they can also reason as an emergent property of the way they work, even if the reasoning is different from human reasoning; based on pattern matching rather than an internal model of reality. For example they can solve math problems they never encountered in their training.

Most people don't know how to use them. LLMs sound much like a human, but need a different approach to talking to a human.