When you say “all popular models” do you mean the commercial models or open source as well? My assumption is that you’d need to run an oss model at the minimum and maybe apply some jailbreaking on top of that.
But wrt how it can impact society, might be worrisome with people emotionally depending on entities they can't have direct real life interaction or experience with. But again that's a separate headache with no real answers as of now. We can only deal with it once it exists.
Emotional dependency on ChatGPT for sfw topics is already here, as you can describe your real life relationships with people and ask it to ask you questions and have it analyze the situation until it says you're right and they're wrong.
Is it materially different if the LLM also says "I put on my robe and my wizard hat..."?
Is it materially different if the LLM also says "I put on my robe and my wizard hat..."?