I think one has to be careful using these models to explore niche topics or complex topics that require humans years of experience just to understand the concepts. They can sound convincing but miss critical details or generate errors that might not be noticed by non-experts. I’ve noticed this in my own work when bouncing ideas off them.