I just setup my own instance a few days ago, if it interests you, do give it a shot!
IIRC reading about it, all data for posts is set to be deleted automatically on a schedule. The catch is that schedule is every 6 months, and it is not configurable currently. From what I read, textual posts of lemmy doesn't consume that much, many reported anywhere from 1-10 GB of data over 6 months - ofcourse it all depends on what kinds of communities are subscribed to your instance.
Not sure if you can restrict image sizes or numbers - atleast not through the admin UI, maybe it's possible through config. You can set global rate limit on image uploads though to not go too crazy.
You can set it so only admins can create communities, or admin would have to approve new communities, or free for all.
If you already have a server, try it out. It shouldn't be tricky, particularly if you're familiar with docker.
My primary use of AI is for programming and debugging. It's a great way to get boilerplate code blocks, bootstrap scripts, one-liner shell commands, creating regular expressions etc. More often than not, I've also learned new things because it ends up using something new that I didn't know about, or approaches I didn't know were possible.
I also find it's a good tool to learn about new things or topics. It's very flexible in giving you a high level summary, and then digging deeper into the specifics of something that might interest you. Summarizing articles, and long posts is also helpful.
Of course, it's not always accurate, and it doesn't always work. But for me, it works more often than not and I find that valuable.
Like every technology, it will follow the Gartner Hype Cycle. We are definitely in the times of "everything-AI" or AI for everything - but I'm sure things will calm down and people will find it valuable for a number of specific things.