I haven't messed with any of the other LLMs. So far I've only messed with Copilot, Grok, and ChatGPT.
I have used suno.com to generate some music. That is scary. So fast, lyrics, complete songs, and they are actually good songs and you can intentionally throw weird challenges at it just to see if it can make those songs. That's only going to get better.
I mainly did it to see how it worked. It's crazy. All I used was the free versions. I'm not actually paying for any AI yet. I can say if I decide one is doing a sufficient job to warrant it I might subscribe to one as I do often hit the "you've used too much for awhile, please wait before doing more" type of issues... That's okay it forces me to take a break. It can suck though if I've built up some good momentum.
Self contained LLMs though I wouldn't expect to do as good a job as something that scrapes the web and reads API documents and such as it is helping me code.
If I end up using an LLM it'd likely be to embed it in a game and have it drive NPC dialog or something so it seemed more alive and interactive. That is the likely use case I might consider using it for in the future. At the moment though that is low priority.
I am primarily using it for complex coding. This is especially useful if for example you are taking "Someguy X's Third Person Controller", and you want to integrate it with "Someguy Y's networking protocol for multiplayer". You can say that and it'll go read the API quickly and start telling you how to do it right away. That is literally saving hours and even days of reading the APIs yourself to get just to that point.
RE: Conversations with AI that I could have (perhaps should have) had with people...