Asking good questions and the riddle of the two guards

I often think about how to reframe complex things through different lenses (i.e. "If I were to explain this to my mother/father/wife, how might they really understand the fundamental idea?"). There's two ideas that have come together for me that I think capture commercial LLMs well:

1. Generalist investors know a little (to a medium amount) about a lot of things, and so they rely on the "real" experts (i.e. SMEs, fund managers, etc.) to get comfortable with a company, investment strategy, etc. One of the most sought-after skills in these roles (and hardest to train) is the ability to ask good questions to get to make a good investment decision. I like to think of the experts as an oracle -- they have all the answers you're looking for, if only you ask the right questions. 
2. There is a classic riddle where you have two guards, one who always lies and one who always tells the truth. They each guard a door, one that leads to freedom and one to certain death. You can ask them any question you want to figure out which guard is in front of which door.

I think a great way to think about commercial LLMs (like ChatGPT) is a combination of these two. They:

1. Are trained on the entire internet, so are knowledgeable about almost any subject imaginable, but 
2. You can never really tell if you're talking to the truth-telling guard (i.e. real expertise aggregated from reliable sources) or the liar guard (i.e. just an answer that sounds good).  

Perhaps not a novel insight, but for me, it's a good way to reframe what LLMs are good at. If you're looking to learn about a new area quickly (e.g. dive into the semiconductor industry, review a contract), then the point #1 shines, and LLMs are amazing. If, however, you need to base real decisions off it (i.e. make a semiconductor investment, send out a contract to a customer), then point #2 starts to rear its head. It still requires the discernment of a thoughtful human to ferret out what is the truth.