Comment by macgyverismo

I have come to think of this in the context of 'you can lead a horse to water (but can't make it drink)'. Good therapy is exactly that. There is surprisingly little personal knowledge needed to help someone. It's mostly about steering someone away from negative feedback loops. As much as I believe in human interactions, I actually think that LLMs can be helpful in this regard, provided the LLM provider doesn't prey on us. That last part is why I'm reluctant to advise people to use LLMs for these use cases. I feel that this is not a technical problem, more an alignment issue. Not of AI, but of other people. (in the form of big tech)