Currently the obvious use is to help people express their thoughts in words. It's helped me a lot writing out resumes and cover letters. This can be extended to languages other than our main/first one.

It's also great to narrow down research on a personal scale in areas where if you have no expertise it would be very hard for you to figure out what you are looking for. I've used it to ID plants, insects and diseases successfully. I didn't get a precise result from ChatGPT, but that's not what I asked. I just requested pointers in the right direction. It delivered.

The next obvious implementation is with software interface. I've already used it (unsuccessfully) to work with Unreal Engine and other 3d software. I got half baked results because the models were not trained specifically for the software in question. But if they were, it would be very easy to just ask the software how to do something instead of searching everywhere for potential answers. That doesn't sound too far fetched and I heard it's a feature that will become standard.