An LLM is a model and can be trained for different tasks, so it's not like I can write a one size fits all style section. There's ones for coding, writing, search aggregation and chatbots and probably others. It's a very active scene and anything written tends to go out of date quickly.
It's important to keep in mind that while it overlaps, this isn't quite a "Self-Hosting OT". It's not too hard to set up a personal ChatGPT or the image generating one. That's the easy part. Something like
Serge does the job pretty well. It's more the hardware requirements that can be a hassle here.
The hard part is finding an "ethical" LLM. Which brings is to the discussion of when does an LLM become ethical (as is the main subjest of this OT)? I would argue it would at least have to be open source and trained on open data. Others, in my mind understandably, take a more luddite approach to the topic.
I have written briefly before on topics of preventing and even poisoning your data in regards to data scraping for training LLMs. This would be something I'd also include in a section for AI/LLMs.
That being said I've been watching Home Assistant improve on it's voice assistant to the point it's getting very close to replacing a Google Nest or Alexa. That's something I would write or link to something about. In that case there's a fair tradeoff to be had.
I will do some research into the topic and see if there's anything that I feel makes sense in this thread that would disrupt the likes of ChatGPT etc. I feel it's a relatively advanced topic though where I try to keep things simple for folks.
Oh nice to see you breaking out of a walled garden. If you need something new to check out, Frigate is designed to be integrated with Home Assistant. :p