I'd only argue that it should be called "emotional support robot" and not "friend"
Assuming people have time to listen, would they be better coders if they explained their problems to human instead? Maybe. But maybe not for them necessarily. E.g. low self-esteem and assuming every criticism is attack on them, human interactions are something expensive to them etc.
It's not a new pattern though. Especially after reading some biographies of famous scientists.
You can't escape that most brains are wired in a way that we are miserable without human connection, but you also can't escape the fact that some people brains are wired differently than others.
Long story short, I don't agree with them but I wouldn't judge them either.
They also run around and are very pleasant to stroke, which is not true of LLMs.
We all know what's going to happen. The content on CIVITAI shows where this will go. Combine it with animation and some personalised responses and many people will find it irresistible.
Horrible it would be if instead one has to work one oneself to become a better human being, a better friend, partner, parent and so on by learning how to be more friendly, outgoing, increasing emotional intelligence etc. All this can be learned, but over weekend (or year).
On the other side, having too much confrontation will destroy your confidence, kill your motivation, blur your plan / vision with uncertainty, etc. It's more likely that those people are facing too much confrontation in their social life that they found AI interaction to be better.
In real life, we need to interact with several people at minimum normally, weekly. Those are having different moral basis and maybe changing daily. It'll be hard to simulate that with AI, that the fact we have the ability to control them means we're in charge of what confrontations are there to stay.
> In real life, we need to interact with several people at minimum normally, weekly.
I think that's one of the problems with social media (aside from AI.) It's too easy to restrict your contacts to only those you agree with.
Their entire business models either heavily incorporate or revolve around exploiting your personal information for their benefit.
But consider iterating such an interaction over the course of, say, 25 years, and comparing the person who was interacting with humans versus the one who interacted with LLMs, and any halfway sensible model of a human will show you what's dangerous about that. Yeah, the former may well have some more bumps and bruises, but on the net they're way ahead. And that's assuming the human who delegated all interaction to LLMs even made it to 25 years.
This argument only holds for LLMs as they stand now; it is not a generalized argument against AI friends. (That would require a lot more work.)
I do think there is higher maximum with other people. But relationships are hard. They take work and there's a decent chance you invest that work in the wrong people.
I can see a life with primarily AI social interaction being an okay life. Which is not the best it can be but also an improvement for some.
If you have the latter option, you also have "leave the bad people behind" as an option because it is made of the things you need in order to "retreat solely into interacting with LLMs" and is in fact simpler.
Cynicism and casting learned helplessness as a virtue are not the solution.
No. Actually it's based on information theory, and probably a better model of what interacting with an LLM would look like a year or five later than the one you are operating on.
Here's a little hint: It has total amnesia. LLMs by their nature scale only so far, and while they may scale larger than ChatGPT, they aren't going to be scaling for an entire lifetime of interaction. (That's going to take another AI technology.)
Ever interacted with someone with advanced dementia but otherwise functioning faculties for any period of time? (I suppose they could well make good therapists too.)
It could be a White Zombie reference: https://www.youtube.com/watch?v=E0E0ynyIUsg