Yeah, some AI companions have been built, but so far they're been a weirdly underdeveloped area of LLMs and the AI hype train in general. You don't need to look very hard at modern society to realize that:
A) There's a lot of demand here (see stuff like self-reported loneliness, the success of marriage these days, and anxiety rates... particularly in the younger and more tech-embracing generations)
B) There's also a lot of danger here, IMO actually far more medium term danger than any other conceivable application for LLMs. It really doesn't matter what your politics are... anyone who has paid the slightest bit of attention over the past 10 years realizes how destabilizing and corrosive modern social (...or pseudosocial) interaction can be.
The things holding it back as far as I can tell are the (understandable) censorship policies that come with centralized control of AI companions, along with a lack of emphasis on giving AI companions apparent motivations and passions. (And a lack of an emphasis on allowing the user to easily and dramatically reshape their AI's personality, interests and motivations to more closely match their own.)
From what I've seen, it's certainly technologically feasible to do it right now, even locally.
I'm not saying it's definitely going to destroy society, but I am saying it's probably the only way that AI could conceivably severely screw up society over the next 20 years.
And at the same time I'm curious as hell to tinker with one myself. If they're out there.
[link] [comments]