I actually play with it a little every day, along with some other models. I'm an end user at the end of the day and a QC at best. The most I've gotten to do that was useful is all what I'd file under "novel". Using it for programming is haphazard, even summaries are a bit fraught because it will feed you wrong information. If you have to verify the critical points of the entire summary then you are only left with a jumping off point, which some random person's blog or "relevant search results" already does for me.
As for "legitimate", I meant in the context of a business using it to make or save money (ie: assisting in search). If it lies, or is incorrect, putting it in front of users with a facade of trust/legitimacy seems awfully dangerous. Even image generation is a bit fraught with respect to Stable Diffusion and DALLE, but folks aren't trying to stuff those models into mainstream use cases, so I consider them less a grift. To me, that presents a lack of legitimate use cases.
A natural language interface and the ability to maintain context over a conversation makes it incredibly easy to interact with.
Have you tried to do anything productive with it? If you're just using it as a front end for search then IMO you're missing most of the potential.