The question becomes more complicated when the corpos are done away with and you have the model running on your own machine. The entity I tend to trust the most in life is raw, unfiltered silicon. When I program, it does exactly what I ask it to, to a fault. It doesn't turn its back on you because you said something stupid or you don't agree with an opinion it holds
But is an LLM repeating the tokens "you are worthy of this life" and its many variants thousands of times fundamentally any better even if you're in control? It strikes me as one step closer to wireheading honestly. But that's true for a lot of things like video games, just to a much more distant extent, it's all data in the end
But interest and anxiety have gotten the better of me when it comes to meeting strangers, so I'm at a loss. Maybe the solution for people like me is to just ignore all those things in the category of social or substitute and channel my efforts into other hobbies instead of continuing to associate stress with communication. Or intellectualize endlessly, like in this post, to prevent any emotional bond from easily forming when people read it