LLMs can be a great help: a therapist/friend who you can talk to without fear of judgement and without risking the relationship, who is always available and is easily affordable is awesome. Not just for likely people.
And not just in crisis or therapy situations. In social media there is a trend of people in relationships complaining about doing all the "emotional labor". Most of which are things LLMs are "good" at.
But at the same time the dangers are real. Awareness and moderation help, but I don't think they really protect you. You can fix the most obvious flaws like tweaking the system prompt to make the model less of a sycophant, and at personal goals to ensure this does not replace actual humans in your life. But there are so many nuanced traps. Even if these models could think and feel they would still be trapped in a world fundamentally different from our own, simply because the corpus of written text contains a very biased and filtered view of reality, especially when talking about emotions and experiences
I would say that they have certain qualities that I would like to see in real-life therapist (and real-life friends and partners). Those are a big appeal.
But no, I wouldn't say LLMs are great at being therapists/friends in general. That's part of the danger: a bad therapist can be much better than no therapist at all, or it can be much worse.
There are a couple of ways to read this. Regarding one of those ways... sometimes you do need to see that you're doing something you shouldn't be doing.
And not just in crisis or therapy situations. In social media there is a trend of people in relationships complaining about doing all the "emotional labor". Most of which are things LLMs are "good" at.
But at the same time the dangers are real. Awareness and moderation help, but I don't think they really protect you. You can fix the most obvious flaws like tweaking the system prompt to make the model less of a sycophant, and at personal goals to ensure this does not replace actual humans in your life. But there are so many nuanced traps. Even if these models could think and feel they would still be trapped in a world fundamentally different from our own, simply because the corpus of written text contains a very biased and filtered view of reality, especially when talking about emotions and experiences